Since launching the Discovery tier less than a year ago, The I/O Fund’s internal process for identifying new winners has greatly improved. Based on the research we produced from this tier, we added stocks like Bloom Energy, Core Scientific and Oklo to our portfolio, and these new additions became some of our biggest wins in 2025.  

As we continue to build out the Discovery library, we’d like to make it as easy as possible for our readers to follow along. We want to cast a wide net, explore thoroughly, and leave no stone unturned. Therefore, we anticipate our coverage to include dozens of stocks over time. However, we also want to make sure we get down to brass tacks by providing you a clear takeaway by ranking what we’ve dug up every quarter. The ranking will also help clarify which ones we are eyeing an entry for and what setups Knox will cover in his Discovery Tier webinars. This will be called the I/O Fund’s Top 10 New Ideas List.  

The ranking we provide is an estimate, which means all 10 stocks are of interest. Given the nature of momentum stocks, the ranking could shift quickly. Please check back to the Discovery Ranking list provided on the Dashboard for any changes in the interim. 

Access the updated Discovery Ranking list here: Top 10 Watchlist

Themes: 

We have covered the trends below on our Q4 Top 15 AI Stocks Report. Our Discovery list offers names centered in these trends, thus we are repeating some of the information for easy reference, but surfacing several new stocks that stand out this quarter. 

AI Networking: 

Networking is at the heart of the new architecture that Nvidia is shipping now as the increased bandwidth is instrumental in driving higher performance. For example, the NVL72 systems will deliver 4X faster training and 30X faster inference compared to HGX systems. Notably, this is accomplished with many more GPUs from eight to 72 per system. 

To support the new systems, the NVLink domain moves from supporting eight GPUs to 72 GPUs and 36 CPUs with a speed of 1.8 TB/s with 18 NVSwitch ASICs, up from four in the HGX/DGX systems due to increasing the number of GPUs but also due to doubling the per-GPU links. The 5th generation of NVLink supports up to 576 GPUs compared to the fourth generation of up to eight GPUs. 

Scale-up refers to increasing the number of GPUs in an AI system. Proprietary NVLink remains the highest performance option for scale-up interconnects, although PCIe and scale-up Ethernet are also used. For the cabling, copper is used for intra-rack scale-up with up to 5,184 cables per system. Future generations of NVLink are likely to integrate optical I/O so that GPUs can communicate across racks without requiring costly retimers. 

A few parameters around the size of the scale-up opportunity: 

• GB200 NVL72 with 72 GPUs and 36 CPUs has 18 NVSwitch chips and 72 InfiniBand NICs for scale-out networking and 36 Bluefield-3 Ethernet NICs for front-end networking. Compare this to the HGX systems with 8 GPUs and the DGX systems with 8 GPUs and 2 CPUs has 4 NVSwitch chips and 8 InfiniBand NICs. 
• This means the new architecture that Nvidia is shipping now results in 9X more GPUs, 4.5X more NVSwitches, 9X more InfiniBand NICs and 18X more front-end NICs. Each GPU requires its own InfiniBand link for scale-out whereas NVSwitch components grow faster than GPU count as each GPU must talk to every other GPU, therefore, it has more of an exponential growth.  

Although NVLink is proprietary, it acts as a bellwether for the importance of AI networking and lesser-known suppliers. Generally speaking, what we can see from looking more closely at Nvidia’s networking fabric is that networking components are increasing 5X to 9X, and in some cases up to 18X.  

Source: Nvidia Technical Blog, “Nvidia Contributes Nvidia GB200 NVL72 Designs to Open Compute Project”Nvidia Contributes Nvidia GB200 NVL72 Designs to Open Compute Project”

Scale-out racks refer to connecting multiple racks across a cluster. InfiniBand switches and Ethernet switches are used for this purpose. As you can see below, Nvidia offers scalable units called SuperPODs that offer tens of thousands of GPUs. For a very large SuperPOD with 16,834 GPUs and 2048 nodes, there would be hundreds of InfiniBand switches required (or a hyperscaler can also use Ethernet switches) and extensive cabling is also required.  

Source: Nvidia DGX SuperPOD technical blog 

Also consider that as the networking and interconnects market matures, there will be new opportunities to participate, for example, co-package optics are expected to be introduced for the Rubin generation of GPUs in 2026-2027. 

Regardless of the exact networking-to-compute ratio, as we scale up AI systems, the architecture becomes a networking and interconnect problem that must continually be solved for, and we want to be correctly positioned within the networking supply chain. Therefore, AI networking will remain a key focus for the I/O Fund’s Top 10 New Ideas in the coming quarters and years. You can expect our team to deliver additional names in this trend and to increase our allocation as needed.  

AI Energy: 

Data center power demand is expected to grow at an accelerated clip through the end of the decade and beyond, with more powerful GPUs and surging growth in inference two main drivers. 

McKinsey projects data center capacity will rise ~2.5x to 219 GW by 2030, up from 82 GW in 2025, with AI contributing 70% of that demand. This corresponds to total capacity growth of 137 GW over the next five years, with 112 GW coming from AI.

Goldman Sachs estimated global data center power usage at 55 GW in early 2025, far below BCG’s 82 GW figure. However, GS projects power usage to reach 84 GW in 2027 and increase further to 122 GW by 2030, corresponding to total growth of just 67 GW. However, considering 2025 and 2026 capex spend could support >20 GW of new capacity, this forecast may understate the pace of capacity growth.

We covered this trend more closely in a lengthy Pro Tier article entitled: “Why Power is Critical for Data Centers and their Hyperscaler Customers” 

As that analysis pointed out, Nvidia’s Blackwell lineup is bringing a significant increase in power consumption, nearly double the H200’s 70 kW at 120 kW for the GB200 NVL72 and 140 kW for the upcoming GB300 racks.   

Beyond Blackwell, Nvidia’s future design lineup shows continual increases in power consumption. Its Rubin generation is expected to boost thermal design power (TDP) by 50% over Blackwell at up to 180 kW per rack, with the upgraded Vera Rubin then doubling this to 360 kW per rack by 2027.  

In its largest configuration, the Vera Rubin NVL576, dubbed the ‘Kyber’ rack, could draw as much as 600 kW (0.6 MW), or 5x that of the GB200 NVL72 in just a two-year design timeframe. These figures do not include networking, interconnects, cooling and other hardware, which will further boost power draw per rack. 

This rapid increase in power consumption per GPU generation is critical, as existing infrastructure is simply unable to meet these escalating power demands. For example, Applied Digital pointed out that nearly 70% of current data centers “contain racks requiring between four and nine kW of power, and less than two percent of data centers have racks with greater than 50kW.”  

For comparison, Super Micro’s GB200 NVL72 SuperCluster requires 132kW, while the upcoming Kyber rack could more than quadruple that to 600kW. Because Blackwell-based servers are now 15x to 30x the power density, cooling and power delivery strategies have to be redesigned, as liquid cooling now becomes a necessity.   

This sharp rise in power density means current infrastructure may be unable to transition from 4-9kW racks to >130kW racks without incurring significant retrofitting costs, while building new infrastructure bypasses that hurdle and allows for optimization for high-powered racks. 

For example, Vantage’s upcoming 1.4 GW campus in Texas for Oracle is designed for ultra-high density racks up to 250kW, yet this will not be enough power to able to host NVL576 racks in just two to three years’ time. Additionally, a former Microsoft Azure AI executive reportedly said he estimated that the “requirements in terms of power for the data center would probably at least double every three years and maybe exponentially so over a period of time,” further reinforcing this. 

The push to 600kW racks over the next few years means this is not a transient problem, rather it is one that the industry will continue to face, meaning continuous new construction may be needed to handle surging power demand.  

In the latest earnings call with CoreWeave, management agreed with the analysis we have presented to Pro Members, stating: “at the end of the day, right now, it's the powered shells that are the choke point that is causing the struggle to get enough infrastructure online for the demand signals that we are seeing […]” 

Takeaway: When building our portfolio, we have to balance many things when it comes to AI energy stocks. Time to power is paramount as some energy infrastructure is 5-10 years out from commencing operations and generating revenue. Secondly, many energy stocks require significant cash to secure energy sources, get government approval and build the underlying infrastructure, which also includes taking into consideration regional differences, transmission capacity and distances from generation sites to metro areas.

AI Data Layer: 

The initial years of AI development were compute intensive to where training created a compute hierarchy. We still see evidence of this hierarchy as companies with access to GPUs, networking and energy have an advantage, and the barrier to entry is high in both costs and by the limited supply preventing widespread access. 

Nvidia has enjoyed a near-monopoly by being the parallel processing leader decades before AI took over as the primary market that demanded massive compute and data throughput for matrix multiplications and vector math. The lack of supply has afforded the world’s largest companies a head start in training and deploying models while startups and enterprises patiently wait for access. As more frontier LLMs are deployed by R&D labs and Big Tech, the emphasis moving forward will be on inference rather than only on training models.  

The inference market is when enterprises and companies sitting on large private data sets will be able to increase the accuracy of open sourced models and licensed models. There will be an important shift to where companies that can offer domain-specific data in various industries, such as finance, healthcare, manufacturing, will do well by optimizing processes to generate more revenue and achieve better margins. It will start with the Fortune 500, the Global Fortune 2000 and well-funded startups. Meanwhile, investors should also not overlook the fact that R&D labs are growing closer to cracking the consumer market, as well, with apps such as OpenAI’s ChatGPT and Sora, or Perplexity’s search.  

While training is benefiting those who sell the compute or own the infrastructure (and we will continue to own these stocks), there will also be a shift toward companies that manage the data pipes by sitting across the many database and software services that enterprises use. Think of all the ERP systems, CRMs, legacy databases, etc, where private data is stored. There will be an emphasis toward combining the data, keeping it private, yet utilizing it to increase the quality of inference.  

Takeaway: Compute will continue to drive the scale for inference; data will drive the quality of inference. Therefore, a key focus for I/O Fund’s Top 10 New Ideas list over the next few quarters (and years ahead) will be AI data stocks that help private enterprises use their valuable data to feed data-hungry reasoning models. We will also, in tandem with AI networking and AI energy, be looking to build our portfolio with exposure to stocks that will participate in the AI inference market, which spans hardware, software, the data layer, and more.  

The stocks below are new ideas and at time of publishing, the I/O Fund does not own these stocks although they are under strong consideration for the portfolio. To find out the stocks the I/O Fund owns, subscribe to our Pro tier for Research or our flagship tier Advanced with additional research, real-time trade alerts, allocations to stocks and weekly webinars.

Palantir: One of the Strongest Reports from Q3 

Thematic: 9/10 
Fundamentals: 10/10  
Valuation: 1/10 

Brief Overview:  Brief Overview:  

The difference between Palantir and other AI-enabled database competitors is that Palantir is able to answer questions a model cannot answer. Traditional business intelligence companies require a complete data set whereas Palantir is able to tackle situations where there is not a complete data set. You can think of the competitive advantage as being actionable depth, which Palantir has described as “the reasoning that goes into decision-making, not just data.”    

Palantir’s Artificial Intelligence Platform (AIP) integrates generative AI with operational data and workflows, and when combined with Palantir’s other platforms Foundry and Apollo, it provides an AI service mesh that can run hundreds of microservices, scale compute through its Rubix engine and orchestrate updates through Apollo.  Additionally, Palantir’s knowledge graph referred to as Ontology is a distinct advantage. The graph offers better context than a large language model would on its own – or as Palantir states, it’s “the reasoning that goes into decision-making.”   

In Q3, Palantir delivered one of the most outstanding reports across tech, with revenue accelerating nearly 15 points sequentially to almost 63%, with red-hot growth in key metrics and a 50 point acceleration in US Commercial revenue since the start of the year.  

Overall Revenue Growth Overall Revenue Growth 

Palantir reported $1.18 billion in revenue in Q3, up an impressive 18% QoQ and beating estimates by 8.4%, driven by unwavering momentum in US Commercial. Commercial revenue rose 21.5% QoQ and 73% YoY to $548 million, a 26 point acceleration from 47% YoY growth last quarter. Government revenue rose 14.5% QoQ and 55% YoY to $633 million, a six point acceleration from 49% YoY in Q2; Government remained Palantir’s largest segment at ~53.6% of revenue. 

On a YoY basis, overall revenue growth accelerated 14.8 points to 62.8% YoY, the largest sequential acceleration to date and marking to Palantir’s highest growth rate since going public. Over the last nine quarters, topline growth has accelerated ~50 points, from just 12.7% in Q2 2023, a rare feat to accomplish.  

For Q4, Palantir guided for revenue up 60.6% YoY to $1.327 to $1.331 billion, well ahead of estimates for 44.2% growth to $1.19 billion. While this does represent a marginal deceleration at face value, this sequential deceleration is in line with trends from previous quarters.  

For the full year, Palantir raised its revenue outlook to $4.396 to $4.400 billion, pointing to YoY growth of 53.5% at midpoint, a sharp acceleration from 29% growth in 2024. To put in perspective the strength of this acceleration, Palantir had initially guided for just 30.9% growth to $3.76 billion in revenue back in Q4 2024; growth is now more than 22 points faster.  

AI Segment Growth AI Segment Growth 

Palantir’s US Commercial segment is generally seen as the primary vector for its AIP-driven growth, with robust momentum only accelerating further in Q3.   

US Commercial revenue grew 29% QoQ and 121% YoY to $397 million in Q3, accelerating from 93% YoY growth in Q2. Since the start of the year, US Commercial growth has accelerated 50 points, and since the start of 2024, growth has accelerated 81 points.  

For the full year, Palantir significantly boosted its US Commercial growth outlook to >104% YoY, up from 85% previously. This corresponds to revenue of $1.433 billion, up from $1.302 billion previously.  

Key metrics for the segment were very strong: TCV closed (total contract value) surged 342% YoY to a record $1.31 billion, while remaining deal value (RDV) rose 199% YoY and 30% QoQ to $3.63 billion. US Commercial deals closed of >$1 million were up 2X YoY and deals closed of >$5 million were up 5X YoY. 

Additionally, other key metrics outside US Commercial were very strong — net retention rate (NRR) expanded six points sequentially to 134%, and over the past two years, NRR has risen an impressive 27 points, with Palantir noting that AIP is continuing to drive existing expansions and new customer conversions. Total remaining deal value rose 91% YoY and 21% QoQ to $8.6 billion, and Palantir also closed its highest ever TCV quarter at $2.8 billion. 

Earnings Earnings 

Palantir reported $0.18 in GAAP EPS in the quarter, up 200% YoY, while adjusted EPS was $0.21, beating estimates by 25.5% and rising 110% YoY. Palantir did not provide a specific guide for EPS for Q4, though current estimates are pegged at $0.12 in GAAP EPS and $0.22 in adjusted EPS, up 300% YoY and 57% YoY, respectively. 

For FY25, Palantir is expected to earn $0.72 in adjusted EPS, up nearly 76% YoY, before slowing to 39% growth to $1.01 in FY26. 

Margins Margins 

Margins strengthened considerably in the quarter, with adjusted operating margin surpassing 50% with more expansion guided for Q4. Palantir’s Rule of 40 score (revenue growth + adj operating margin) expanded to a wild 114%, up from 94% last quarter and 68% last Q3.  

Gross margin was 82% in Q3, up one point QoQ and two points YoY, while adjusted gross margin was 84%, up two points YoY and QoQ. 

GAAP operating margin was 33%, an impressive 6 point QoQ and 17 point YoY expansion. Adjusted operating margin was 51%, breaking past 50% for the first time and up 5 points QoQ and 13 points YoY. For Q4, Palantir guided for adjusted operating margin to be 52%, showcasing its ability to drive strong margin expansion alongside swift revenue acceleration. Full year adjusted operating margin guidance was raised from 46% to 49%.  

GAAP net margin was 40%, up 7 points QoQ and 20 points YoY. Adjusted net margin was 45%, up 5 points QoQ and 12 points YoY. Palantir is one of the few, if not only, tech companies to have 40% GAAP net margins with revenue growth accelerating to above 60%. 

Cash Cash 

Cash flows were strong, though cash flow margins dipped on a YoY and QoQ basis. Operating cash flow was $507.7 million for a 43% margin, shrinking from a 54% margin in Q2 and 58% in the year ago quarter.  

Adjusted free cash flow was $539.9 million for a 46% margin, down from 57% in Q2 and 60% in the year ago quarter. Palantir raised its adjusted FCF guidance for the year to $1.9 to $2.1 billion, or a 45.5% margin, up from a 42.8% margin previously. 

Cash and equivalents totaled $6.4 billion and debt remained zero.  

Valuation Valuation 

Valuation is the crux for Palantir as the stock trades at 100x forward revenue, nearly triple its five-year average of 36x and in rather uncharted territory for software stocks. On the bottom line, Palantir trades at 256x forward adjusted EPS despite a >40% margin, more than double its average 109x multiple.  

Notable Risks Notable Risks 

The valuation with Palantir is a gamble as the company is attempting to set a new bar for AI software, with >100x forward sales multiples only achieved for short fashion in 2021 for a handful of prior market darlings, whose stocks have yet to return to those prices. This elevated valuation may also present a risk if/when the company reaches peak revenue growth as comps will quickly get tougher.  

Celestica: 800G Switch Demand Accelerating into FY26 with 1.6T Switches Soon Ramping 

Thematic: 10/10
Fundamentals: 7/10
Valuation: 1/10 

Brief Overview: Brief Overview: 

Celestica is an under-the-radar beneficiary of the AI networking trend, capitalizing on strong demand for 800G and 1.6T Ethernet networking switches and leveraging its deep ties to hyperscalers as an original design manufacturer (ODM). 

Celestica guided for one of the most impressive accelerations seen in this last quarter of earnings, underpinned by its 800G switches accelerating next year with 1.6T ramps on deck. For 2026, Celestica expects revenue growth to accelerate around five points to 31% YoY in 2026, whereas consensus had been pegged at just 17% YoY. This strong upside is being driven by networking and custom AI compute platforms with visibility into 2026-2027. 

In terms of AI revenue, Celestica’s Cloud and Connectivity Solutions (CCS) segment is guided to generate $9 billion in revenue in 2025, up ~40% YoY, accounting for nearly 74% of total revenue. CCS, which includes AI networking, server, storage and rack-scale system solutions, is Celestica’s main growth driver and is also expected to grow ~40% annually in 2026 and 2027. 

Celestica is closely linked to Broadcom’s networking platforms as a key vendor, serving major customers such as Google and Meta, with some of its notable product engagements including Google’s TPU server racks, and Meta’s Minerva ASICs servers, Wedge400 switches and also its next-gen Tomahawk5-based 400G AI fabric switch Minipack3. Additionally, management’s commentary suggests that OpenAI could become a key customer as soon as 2027 – read more on this in our Discovery deep dive. Discovery deep dive.  

Revenue: Revenue: 

Celestica reported Q3 revenue of $3.19 billion, up 28% YoY and 10.4% QoQ, coming in more than 6% ahead of estimates for $3.00 billion. Revenue from the CCS segment rose 43% YoY to $2.41 billion, driven by an 82% YoY increase in Communications revenue to $1.94 billion offsetting a (24%) decline in Enterprise to $477 million on an AI program transition with a hyperscaler. Celestica’s other segment, Advanced Technology Solutions (ATS, serving aerospace, defense, industrial and semicap equipment markets) saw revenue decline (4%) YoY to $781 million.  

For Q4, Celestica guided for revenue of $3.325 billion to $3.575 billion, up 35% YoY at midpoint of $3.45 billion, a seven point acceleration. On a QoQ basis, this correlates to 8% growth sequentially.  

As a result, Celestica raised its fiscal 2025 guidance from $11.55 billion to $12.2 billion, a strong 7% raise with just one quarter to go, signaling the strength of demand the company is witnessing in Q3 and Q4. The updated guide points to 26% YoY growth.  

Perhaps more important was Celestica’s initial guidance for fiscal 2026, with the company laying out an initial forecast for $16 billion, for 31% YoY growth, a five point acceleration. This was $1.86 billion ahead of estimates for $14.14 billion, a large 13.2% beat.  

The impressive fiscal 2026 guide and revisions to consensus estimates for fiscal 2027 have taken Celestica’s forward growth CAGR from FY25 to FY27 to a strong 28%, up from 17% prior to the report. 

AI Segment Growth: AI Segment Growth: 

In terms of AI revenue, Celestica’s Cloud and Connectivity Solutions (CCS) segment is forecast to generate $9 billion in revenue in 2025, up ~40% YoY, and accounting for nearly 74% of total revenue. CCS, which includes AI networking, server, storage and rack-scale system solutions, is Celestica’s main growth driver and is expected to grow ~40% annually in 2026 and 2027. 

In Q3, revenue from CCS segment rose 43% YoY to $2.41 billion, driven by an 82% YoY increase in Communications revenue to $1.94 billion offsetting a (24%) decline in Enterprise to $477 million on an AI program transition with a hyperscaler. Communications revenue notably accelerated to 18% QoQ from 15% QoQ in Q2, on strong demand for 800G switch products and solid demand for optical products.   

For Q4, CCS revenue is implied to accelerate nine points to 52% YoY, with Communications growth guided in the high-60s YoY on strong switch demand, and Enterprise guided in the low-20s as the new AI program is set to ramp. Despite the seemingly strong guide in Communications, QoQ growth would be just 1% QoQ, a sharp deceleration from Q3’s 18% QoQ growth. 

For 2026, Celestica guided for approximately 40% YoY growth in CCS to ~$12.6 billion, up from $9 billion guided for 2025, supported by views for accelerating 800G demand, early 1.6T ramps and the ramp of its next-gen AI compute platform to full-volume. Management also hinted that they have visibility and confidence in maintaining at least 40% growth for CCS in 2027 – more on this in our deep dive.  

Earnings Earnings 

Celestica reported GAAP EPS of $2.31 in Q3, beating the $1.38 estimate by 67.6%. Adjusted EPS was $1.58, beating the $1.49 estimate by just 6% and representing growth of 52% YoY.  

For Q4, Celestica guided adjusted EPS to be in the range of $1.65 to $1.81, which, at the $1.73 midpoint, is only marginally ahead of estimates for $1.71. This also corresponds to a slight acceleration to 56% growth. 

For fiscal 2025, Celestica boosted its adjusted EPS outlook by 7.3% to $5.90, from its previous forecast for $5.50 and pointing to 51% YoY growth. For fiscal 2026, Celestica outlined an initial guide for $8.20 in adjusted EPS, up 39% YoY and well ahead of estimates for $7.22.  

Margins Margins 

Margins continued to expand in Q3, with some signs of operating leverage arising from strong Communications growth as operating margin expanded by 4.7 points YoY versus a 2.6 point YoY expansion for gross margin.  

• GAAP gross margin was 13.0% in Q3, up 0.2 points QoQ and 2.6 points YoY. 
• GAAP operating margin of 10.2%, up 0.8 points QoQ and 4.7 points YoY. Adjusted operating margin was 7.6%, up 0.2 points QoQ and 0.8 points YoY. 
• GAAP net margin of 8.4%, up 1.1 points QoQ and 4.8 points YoY. However, adjusted net margin was just 5.7%, up just 0.1 points QoQ and 0.7 points YoY due to a $113 million impact from gains on total return swaps.  

For fiscal 2025, Celestica guided for adjusted operating margin to be 7.4%, and for 2026, only a slight increase to 7.8% despite the 31% growth on the top-line. This suggests that its positioning primarily as an ODM may limit future upside to operating margins.  

Cash Cash 

On the other hand, cash flows are rather thin and fell to the lowest level in a year.  

Operating cash flow was $126.2 million, or a 4% margin, down from 5.3% in Q2 and 4.9% in the year ago quarter. OCF growth was just 2.4% YoY and was also the lowest cash flow since the year ago quarter.  

Adjusted FCF was $89 million for a 2.8% margin in Q3, up 15.6% YoY but also the lowest level since the year ago quarter. Adjusted FCF margin was down from 4.1% in Q2 and 3% in the year ago quarter.  

For fiscal 2025, Celestica raised its adjusted FCF guidance slightly to $425 million, from $400 million prior, for a 3.5% margin, while capex is guided to $200 million, or 1.6% of revenue. Fiscal 2026 adjusted FCF was guided at $500 million, up 18% YoY and for a 3.1% margin, with the margin decline driven by higher capex, guided to rise 50-100% YoY to $300 to $400 million, or 2.2-2.5% of revenue.  

Cash and equivalents totaled $305 million while debt totaled $728 million in term loans. Including an undrawn revolver, total liquidity is approximately $1.1 billion. Celestica’s gross debt to TTM adjusted EBITDA was 0.8x, improving by 0.1 points sequentially and 0.3 points from last year. 

Valuation Valuation 

Celestica is trading close to peak multiples on the top and bottom line. Forward PS is currently at 2.8x, well above the five-year average of 0.75x and 40% above the 2x multiple it commanded at the start of September. Even on the fiscal 2026 guide, shares are at an elevated 2.1x multiple, just below peaks at 2.5x. 

On a forward PE basis, shares are trading at 51x fiscal 2025 adjusted EPS and 41.3x fiscal 2026, well above the five-year average of 15.4x and prior resistance at 25x in late 2024 and early 2025. This is slightly below current peaks around the 60x level from October and November.  

Notable Risks Notable Risks 

Valuation is the primary risk, and while it could be argued that the company is deserving of a material re-rating higher on strong AI-driven growth and a shift to higher-margin, custom rack solutions come 2027, margins remain thin with operating margin only just crossing into double-digit territory. Additionally, its ODM positioning also presents a risk as even a shift to higher complexity, higher value products may be unable to produce continuous margin expansion into the low-teens. 

Celestica’s Communications growth within CCS also presents another key risk for Q4, as the high-60s YoY growth guide would imply QoQ growth of around 1% next quarter. This would mark Communications’ lowest sequential growth in the last two years, and its first time reporting single-digit sequential growth in the last seven quarters, raising a potential red flag considering Communications is primarily driven by networking/800G switches. However, a likely explanation of this could be the strong outperformance in Communications in Q3 – guidance was for low-60s YoY growth, which Celestica beat by ~20 points. As a result, QoQ growth was likely expected to be ~4%, but came in at 18%, possibly representing a much stronger-than-expected ramp of 800G platforms in the quarter.  

Arm: Data Center Royalties Double YoY, Targeting 50% Data Center CPU Share 

Thematic: 9/10
Fundamentals: 9/10 
Valuation: 3/10 

Brief Overview Brief Overview 

AI’s need for high-performance, energy-efficient chips creates a long-term tailwind for Arm, as the company’s heterogenous CPU architectures are seeing rapid adoption in data center applications. This is coming from both next-gen merchant GPU platforms and custom silicon deployments from hyperscalers, with Arm now forecasting its server CPU share to reach 50% in 2025, up from 15% in 2024.  

The company’s license and royalty revenue model had centered around its v9 architecture, as it commanded double the royalty of v8, featuring in “virtually all high-end data center chips” and commanded a majority share in smartphones. For example, Arm’s Neoverse V2 (based on v9) powers Nvidia’s Grace CPU on its Grace Hopper and Grace Blackwell platforms, along with Amazon’s Graviton4 CPUs, Google’s Axion CPUs, and more.  

Arm is now pushing ahead with its Compute Subsystems (CSS) platform to help accelerate time to market for complex chip designs, such as Microsoft’s newest Azure Cobalt 200 CPU rolling out through 2026. CSS notably carries double the royalty rate as v9, which management placed at roughly 10%. Arm also signed three new CSS licenses this quarter, bringing its total CSS licenses up to 19 across 11 companies; five of these companies are already shipping CSS-based chips.  

Revenue Revenue 

Arm’s revenue growth accelerated more than 22 points from 12.1% YoY in fiscal Q1 to 34.5% YoY in fiscal Q2 to $1.13 billion, while QoQ growth rebounded from (15.1%) to 7.8% QoQ. 

Royalty revenue increased 21% to a record $620 million, driven by growth in smartphones, auto and data center, along with increased v9 penetration and the ramp of CSS platforms. However, this was a slight deceleration from 25% growth in the prior quarter. 

Licensing revenue rose 56% YoY to $515 million on normal timing fluctuations, accelerating from (1%) growth in the prior quarter.  

For Q3, Arm guided for revenue of $1.225 billion at midpoint, though this represents a deceleration to 24.6% YoY and 7.9% QoQ growth. 

AI Segment Growth AI Segment Growth 

Arm does not provide specifics into its data center revenue contributions, but noted that data center royalties doubled YoY on continued deployment of Arm-based chips at hyperscalers. Data center Neoverse royalties more than doubled YoY, and Arm expects to reach 50% share in of CPUs deployed by hyperscalers by the end of 2025.  

For another view, Arm’s management explained that it is reasonable to assume cloud and networking would reach 15% to 20% share of royalty revenue for the fiscal year, up from ~10% last year. Assuming Q3 and Q4 see royalty revenue rise ~20% YoY, this could project cloud and networking’s contribution between $394 to $525 million. 

Earnings  Earnings  

Arm delivered strong GAAP EPS growth in Q1 as margins expanded down the line, while adjusted EPS growth was more muted but solid nonetheless.  

GAAP EPS was $0.22 in Q1, up 120% YoY and more than 66% ahead of estimates for $0.13. Adjusted EPS was $0.39, nearly 18% ahead of estimates for $0.33 and representing growth of 30% YoY.  

For Q3, Arm guided for adjusted EPS to be $0.41, +/- $0.04, for YoY growth of just 5%. Q4 is estimated to see growth of just 2.7% YoY to $0.56, before reaccelerating to >29% YoY growth in each quarter of fiscal 2027. 

Margins Margins 

Arm saw strong margin expansion down the line, with operating and net margin expanding at a much larger degree than gross margin in Q2, signaling that adoption of its higher margin v9 and CSS platforms is translating to bottom line strength. 

• GAAP gross margin was 97.4% in Q2, up 1.2 points YoY and 0.2 points QoQ.  
• GAAP operating margin was 14.4%, up 6.8 points YoY and 3.6 points QoQ. Adjusted operating margin was 41.1%, up 2.5 points YoY and 2 points QoQ; for Q3, adjusted operating margin is implied to be ~39.4% at midpoint assuming gross margin is flat QoQ.
• GAAP net margin was 21%, up 8.3 points YoY and 8.7 points QoQ. 

Cash Cash 

Cash flows improved substantially on a YoY basis, and Arm’s balance sheet remains robust and debt-free. 

• Operating cash flow was $567 million for a 50% margin, up from a 0.7% margin in the year ago quarter and a 31.5% margin in the prior quarter. 
• Adjusted free cash flow was $411 million for a 36.2% margin, a significant increase from (7.7%) in the year ago quarter and 14.2% in the prior quarter.  

Cash and equivalents totaled $3.26 billion and debt was zero. 

Valuation Valuation 

Unlike many of the other names on this list, Arm is trading more than 20% below its average multiples on the topline, though the company still commands a premium multiple to many of its semiconductor customers. Arm trades at 24.3x forward PS, below its 31.6x average since IPO and well below its peaks at 50x, though it has traded as low as 16x. 

On the bottom line, Arm trades at 64.5x forward PE, below its average of 79.5x, though shares have traded as low as 40x and as high as 125x.  

Notable Risks Notable Risks 

Despite increasing royalty rates by 2X with each new architecture, from 2.5% under v8 to 5% with v9 and 10% with CSS, Arm’s growth may continue to lag that of peers – the company may need to take the leap into design as IP licensing did not move the needle enough from the mobile era. 

Fabrinet: Key Nvidia Partner with Revenue Accelerating to 30% YoY  

Thematic: 8/10
Fundamentals: 4/10 
Valuation: 4/10 

Brief Overview Brief Overview 

Fabrinet provides advanced optical packaging, high-precision optical and electro-optical manufacturing services to OEMs, with revenue primarily stemming from transceivers, active optical cables, optical subsystems for high-speed networking, and data center interconnect. 

Fabrinet is also a key optical partner for Nvidia, with its contributions said to be for short-reach active optical cables with 800G transceivers for Nvidia’s InfiniBand platforms, and optical engine packaging. Fabrinet was also stated as key partner for Nvidia’s upcoming silicon photonics CPO switch platforms during GTC 2025.  

Nvidia accounted for 27.6% of Fabrinet’s revenue in fiscal 2025, or ~$943.7 million; however, this was down nearly (7%) YoY, potentially due to the Blackwell delays from early 2025. Cisco is Fabrinet’s second largest customer, said to be for optical transceivers, accounting for 18.2% of revenue in fiscal 2025, or $622.3 million, up 61% YoY. 

However, like peers positioned similarly in the contract manufacturing space, margins are quite thin and did not show any expansion in the past quarter, and Fabrinet may be unable to produce substantial margin expansion moving forward.  

Revenue Revenue 

Fabrinet reported record fiscal Q1 revenue of $978.1 million, up 7.5% QoQ and 21.6% YoY. This marked a slight acceleration from 20.7% YoY and 4.3% QoQ growth in FQ4.  

Fabrinet’s Optical Communications (OC) business remained the key driver, accounting for 76.4% of revenue. OC revenue rose 8.4% QoQ and 19.3% YoY to $747 million in Q1, accelerating 15.5% YoY and 4.8% QoQ in Q4. Non-optical Communications revenue, or auto, industrial laser and other end markets, was $231.2 million, up 4.6% QoQ and 30% YoY. 

For fiscal Q2, Fabrinet guided for revenue of $1.05 to $1.10 billion, accelerating more than seven points to 29% YoY at midpoint, while QoQ growth would accelerate 2.4 points to 9.9%. Revenue is expected to accelerate to 30-31% YoY in both Q3 and Q4 to $1.13 billion and $1.19 billion respectively.  

For fiscal 2026, Fabrinet has not provided a set guide, though analysts expect revenue to rise 28% YoY to $4.39 billion, accelerating from 18.6% in fiscal 2025. 

AI Segment Growth AI Segment Growth 

Fabrinet offers a few distinct breakdowns within Optical Communications revenue, highlighting datacom and telecom, and within telecom, data center interconnect (DCI).  

Datacom revenue was $273.1 million in Q1, down (17%) YoY and (1%) QoQ, which was a smaller decline than expected due to a smaller QoQ decline at Fabrinet’s largest datacom customer and larger contributions from other datacom customers.  

DCI revenue was $138.1 million, up 92.3% YoY and 29% QoQ, marking a sharp acceleration from 44.8% YoY and 3.5% QoQ growth in the prior quarter. DCI now accounts for ~14% of total revenue. 

Fabrinet also introduced HPC as a reportable segment under Non-optical Communications in Q1, as the company qualified and started to ramp its first HPC program for AWS, which contributed $15.4 million in revenue. Management believes HPC “will scale considerably over the coming quarters and become a significant driver to our overall growth.” 

Fabrinet did not provide a breakdown on OC revenue by data rate in Q1, though in Q4 (the Aug quarter), revenue from >800G data rates was $312.7 million, up 20.7% YoY and 32.5% QoQ and accounting for 34.4% of total revenue. 

Earnings  Earnings  

Despite a lack of margin expansion, Fabrinet has a rather defensible bottom line and delivered solid EPS growth in Q1. GAAP EPS was $2.66, up 24.9% YoY, while adjusted EPS was a record $2.92, up 22.2% YoY and beating estimates by 3.5%. 

For Q2, Fabrinet guided for GAAP EPS of $2.91 to $3.06, up 25.4% YoY at midpoint, while adjusted EPS was guided to be $3.15 to $3.30, up 23.6% at midpoint. The growth rates here lagging revenue by four to five points suggests that margins may feel a bit of pressure next quarter. 

Adjusted EPS growth is forecast to accelerate into the end of the fiscal year, however, suggesting margins may rebound quickly if there is softness in Q2. Q3 and Q4 are both forecast to see adjusted EPS growth of 36% to 37% YoY, or six to seven points faster than revenue growth. 

For fiscal 2026, Fabrinet is currently expected to generate GAAP EPS of $12.25, up 33.6% YoY, and adjusted EPS of $13.25, up 30.3% YoY. 

Margins Margins 

As noted above, Fabrinet did not deliver much on the margin front in Q1, with gross margin contracting slightly, and operating margin remaining flat YoY but dipping slightly sequentially. Additionally, its contract manufacturing position may not be able to drive meaningful margin upside over the coming quarters. 

• Q1 GAAP gross margin was 11.9%, down 0.4 points YoY and 0.3 points QoQ. Adjusted gross margin was 12.3%, down 0.4 points YoY and 0.2 points QoQ. 
• GAAP operating margin was 9.6%, flat YoY but down 0.2 points QoQ. Adjusted operating margin was 10.6%, down 0.1 points YoY and QoQ, 
• GAAP net margin was 9.8%, up 0.2 points YoY and QoQ. Adjusted net margin was 10.8%, flat YoY but up 0.3 points QoQ. 

Cash Cash 

Fabrinet’s cash flow generation was solid in Q1, with operating cash flow margin expanding and free cash flow rebounding after a soft Q4. Fabrinet also has a healthy balance sheet with nearly $1 billion in cash and no debt. 

• Operating cash flow was $102.6 million in Q1 for a 10.5% margin, up from a 10.3% margin in the year ago quarter and a 6.1% margin in Q4. 
• Free cash flow was $57.3 million for a 5.9% margin, down from a 7.8% margin in the year ago quarter, but up from 0.5% in Q4. The YoY contraction in FCF margin was primarily due to capex, which was $45.3 million, up 123.5% YoY. 

Cash and equivalents totaled $968.8 million and debt was zero. 

Valuation Valuation 

Fabrinet is trading close to peak multiples on the top and bottom line. On a forward PS basis, Fabrinet trades at a 3.8x multiple, below its 4.3x peak from early December and notably well above its five-year average multiple of 2.5x; to note, Fabrinet has spent much of the last two and a half years trading between 2x to 3x, with the re-rating above 3x taking place since October 2025. 

On the bottom line, Fabrinet trades at a 35x forward PE, slightly below its peak of 40x and again well above its five-year average of 23x. Similar to forward PS, Fabrinet has spent the last two and a half years primarily between 18x and 28x forward PE.  

Notable Risks Notable Risks 

The main risks to Fabrinet relate to its positioning as a contract manufacturer for Nvidia, Cisco and other key clients, as the company may be unable to produce substantial margin expansion beyond its current profile, meaning EPS growth will likely be closely correlated with revenue growth rates. Fabrinet is also quite highly exposed to Israel, which accounted for 29% of revenue in fiscal 2025, and continuation of global conflicts could potentially impact growth.  

Fabrinet’s strategic positioning with Nvidia as its key customer may present a risk – while it could see tailwinds from the ramp of the silicon photonics CPO switches in 2026, any delay in Rubin’s ramp could adversely affect shipments and revenue. Analysts have also raised concerns in the past that Fabrinet could face headwinds if hyperscalers ‘unbundle’ from Nvidia’s ecosystem and shift away from Nvidia’s InfiniBand or Ethernet switch options.  

Vertiv: Orders Surge 60% YoY, 20% QoQ in Q3, FY25 Guidance Raised 

Thematic: 8/10
Fundamentals: 8/10
Valuation: 3/10 

Brief Overview Brief Overview 

Vertiv will not win any hypergrowth stock awards, especially as management has previously offered CAGR guidance of 15% to 17% through 2029. Rather, it’s where Vertiv is positioned as an AI infrastructure partner especially as the trend turns toward modular infrastructure that makes this a stock to watch.  Essentially, all roads point toward Vertiv’s power and thermal solutions becoming increasingly important for future generations of rack scale solutions, with the company already preparing 800V DC solutions for Nvidia’s Rubin Ultra platform due in 2027. 

Revenue Revenue 

Vertiv reported revenue up 29% YoY and 1% QoQ to $2.676 billion, well ahead of its original guidance for 23% growth in the third quarter. This was driven by 43% YoY growth in the Americas on accelerated AI demand and 20% growth in APAC.  

For Q4, Vertiv guided for revenue to be $2.81 billion to $2.89 billion, up 6.5% QoQ and 18-22% YoY at the $2.85 billion midpoint. While this was ahead of previous guidance for $2.735 to $2.815 billion, this would still represent a nine point deceleration on the topline at midpoint. Management expects Americas revenue to be up high-30s, APAC up mid-single digits and EMEA down high single digits but up mid-teens QoQ. 

The strong outperformance in Q3 also led to Vertiv hiking its FY25 revenue guidance from $10 billion at midpoint to $10.2 billion at midpoint, pointing to organic growth of 26-28% YoY. Management did not provide any direct insight into FY26, though they did say that based on the “substantial backlog and clear visibility of pipeline, we anticipate continued significant organic sales growth in 2026,” with EMEA potentially reaccelerating in 2H 2026. 

AI Revenue Metrics AI Revenue Metrics 

Vertiv’s backlog rose ~30% YoY and 12% QoQ to $9.5 billion, reaccelerating from 21% YoY growth last quarter. More importantly, the $1 billion sequential increase in backlog was the largest in more than two years. However, one of the stronger metrics was order growth, with Vertiv reporting organic orders up 60% YoY and 20% QoQ in Q3. This drove a ten point rebound in TTM organic order growth to 21% YoY, from 11% in Q2. 

However, starting in Q4, Vertiv will no longer report on quarterly orders and backlog information, and instead will report a new metric “projected full year orders.”  

The following was stated in Q2: “Beginning on our Q4 and full year 2025 earnings call, we will provide projected full year orders rather than quarterly orders and backlog information. We believe this better aligns with how we run our business. We will provide updates on the full year projections quarterly as we progress through the year and as we deem necessary.”  This could create a boost to Vertiv’s stock to remove the lumpiness from quarterly reports and to also be more forward looking in terms of visibility offered to investors.  

Earnings Earnings 

Vertiv reported adjusted EPS up 63% YoY to $1.25 in the quarter, beating the $0.99 estimate by 25%. GAAP EPS of $1.02 beat estimates by 16.7%. For Q4, adjusted EPS was guided to decelerate to 27% growth to $1.26 at midpoint.  

For the full year, Vertiv raised its adjusted EPS forecast to $4.07 to $4.13, up from its prior view for $3.75 to $3.85. At midpoint, this represented a nearly 8% hike, now pointing to 44% YoY growth versus 33% previously.  

Margins Margins 

Vertiv reported expanding margins across the board in Q3, though Q4 is expected to be approximately flat for adjusted operating margin.  

• Gross margin was 37.8%, up 1.3 points YoY and 3.8 points QoQ. 
• GAAP operating margin was 19.3%, up 1.4 points YoY and 2.5 points QoQ. Adjusted operating margin was 22.3%, up 2.2 points YoY and 3.8 points QoQ, driven by tariff mitigation efforts and strong execution addressing operational inefficiencies.  
• Net margin was 14.9%, up 6.4 points YoY and 2.6 points QoQ. 

For Q4, adjusted operating margin was guided to be up 0.9 points YoY and approximately flat QoQ at 22.4%, as “progress addressing operational inefficiencies [is] offset by acceleration in growth investments and negative impact from new tariffs.” This is a rather steep decrease from Q2’s guidance for 23.6%, which would’ve been its best adjusted operating margin print since going public in 2020.  

For FY25, Vertiv slightly raised its adjusted operating margin forecast by 0.2 points at midpoint to 20.2%, representing YoY expansion of 0.8 points. This is strong as it comes in the face of “significant headwinds from tariffs and operational inefficiencies driven by supply chain actions to mitigate tariffs.” Tariff impacts are expected to be materially offset exiting Q1 ’26. 

Cash Cash 

Vertiv reported strong cash flows in Q3, with operating cash flow of $508.7 million, up nearly 36% YoY. OCF margin was 19%, up 1.8 points YoY and 6.8 points QoQ. 

Q3 adjusted free cash flow was $462 million, up 32% YoY. Adjusted FCF margin was 17.3%, up 1.1 points YoY and 6.8 points QoQ. Q4 adjusted FCF was guided to be $496 million for a 17.4% margin, up marginally from Q3. Vertiv boosted its adjusted FCF guidance by $100 million, now forecasting $1.5 billion for the year, up from $1.4 billion previously. This corresponds to a 14.7% margin.  

Accounts receivable dipped (1%) QoQ to $2.81 billion, while inventories rose less than 2% QoQ to $1.43 billion. 

Cash, equivalents and investments totaled $1.94 billion, while debt totaled $2.90 billion. 

Valuation Valuation 

Vertiv is trading around 15% below peak multiples on the top line, and more than 23% below peak on the bottom line. Vertiv’s forward PS is 6.2x, below its recent peak valuation at 7.2x at the end of October and substantially higher than its April low at 2.2x forward PS. 

On the bottom line, Vertiv is trading at 40x forward earnings, slightly below its late October peak at 47x and more than 23% below its late 2024 peaks at 52x. 

Notable Risks Notable Risks 

Vertiv’s extended valuation is a primary risk as the company contends with a sharper deceleration on the top line heading into Q4, as well as a sharp deceleration in EPS growth from 63% in Q3 to 27% in Q4. Margins are also a line item to watch, considering management had guided for a Q4 adjusted operating margin of 23.6% back in Q2 but then subsequently cut that guide to 22.4% in Q3. 

Talen: Q3 Revenue Up 29% QoQ, Operating Margin Strengthens 

Thematic: 9/10 
Fundamentals: 6/10  
Valuation: 2/10 

Brief Overview: Brief Overview: 

Talen is an independent power producer with more than 10GW of generation capacity with 2.2GW of that being nuclear. The company’s assets are primarily located in Pennsylvania, Maryland and now Ohio, yet data center regions and capacity are growing including a long-term power purchase agreement with Amazon to fuel data centers in Pennsylvania. 

Talen is expanding its power production portfolio with recent acquisitions of two combined-cycle gas turbine (CCGT) plants, Freedom Energy Center and Guernsey Power Station, for ~$3.8 billion. The two plants will add 2.8 GW to Talen’s energy assets in the PJM region – both are suitable for hyperscale data center power supply. This comes at a time when data center construction is surging in PJM’s region as its grid faces increasing strain, meaning the plants could be more valuable for meeting near-term hyperscaler power needs. 

Revenue Growth Revenue Growth 

Talen reported revenue of $812 million in Q3, up 28.9% QoQ and 24.9% YoY, with the quarter including the higher 2025/26 capacity pricing of ~$270/MW-day.  Revenue from contracts with customers rose 51.9% YoY to $697 million. 

Looking ahead, revenue is expected to accelerate to 43.2% YoY in Q4 and 170.2% in Q1, though these estimates have been revised lower from 53.4% YoY and 184.5% YoY at the start of November. Talen said that for Q4, “things are a bit better given the market move-up, but we are still projecting to be at the lower end of our guidance range as we previously stated at that September Investor Day.” 

For 2026, Talen said it is “forwards tick up. Gas is up, sparks are expanding and load continues to be strong, all factors that continue to impact commercial positioning on long-term transactions.” However, 2026’s estimated capacity revenue is now $733 million, down $14 million from an estimated $747 million as of Q2.  

AI Revenue AI Revenue 

Talen’s deal with Amazon is contributing minimally so far, with just $0.70 in adjusted FCF per share expected in FY25. By mid-2026, Talen expects to deliver 240MW, and expects adjusted FCF to rise ~121% to $1.55 per share, before rising at a ~27% annually to $2.50 per share by FY28 as capacity scales to 480MW. 

Margins Margins 

Gross margin (operating revenue minus energy expense) improved to 64% in Q3 from 60% in Q2 and 62.3% in the year ago quarter. 

Operating margin was 32.3% in Q3, a strong improvement from 10.5% in Q2 and 24.3% in the year ago quarter. This was likely driven by the $116 million increase in capacity revenue and, to a lesser extent, the $99 million increase in energy revenue, as operating expenses were up only marginally YoY. 

Net margin was 25.5% in Q3, improving from 11.4% in Q2 but down marginally from 25.8% in the year ago quarter. 

Earnings  Earnings  

Talen reported $4.25 in GAAP EPS in the quarter, up 34.5% YoY. Talen did not provide guidance for Q4, yet current estimates point to a (25.9%) QoQ decline to $3.15 in GAAP EPS. For 2025, Talen is expected to report GAAP EPS of $5.13, before rising nearly 290% to $20.00 in 2026. 

Talen also reported adjusted EBITDA of $363 million in Q3, representing a 44.7% margin, and improving substantially from $90 million for a 14.3% margin in Q2. Despite the strong increase, Talen narrowed its adjusted EBITDA guidance for fiscal 2025 to $975 million to $1,000 million, down from $975 million to $1,125 million previously. Management said this was because they are towards the lower end of guidance, due to a lack of price volatility in Q3 and the Susquehanna outage. 

For 2026, Talen reaffirmed its adjusted EBITDA guidance for $1,750 million to $2,050 million, representing 75% to 110% YoY growth.  

Cash and Balance Sheet  Cash and Balance Sheet  

Talen’s cash flow generation improved in Q3, with adjusted FCF margin rising to the high-20% level. Talen also improved its balance sheet, though cash to debt remains upside-down for the moment. 

Operating cash flow was $359 million for a 44.2% margin, a sharp improvement from a (22.9%) margin in Q2 and 14.8% in the year ago quarter. 

Adjusted free cash flow was $223 million for a 27.5% margin, up from (12.4%) in Q2 and 14.9% in the year ago quarter. Talen said Q3’s adjusted FCF also included higher capex related to the extended Susquehanna refueling outage. For fiscal 2025, adjusted FCF is tracking near the middle of Talen’s guided range of $470 million to $490 million (narrowed from $450 million to $540 million). For fiscal 2026, Talen is guiding adjusted free cash flow in the range of $980 million to $1.18 billion, more than doubling YoY. 

Cash and equivalents totaled $497 million, a solid improvement from $135 million in Q2. Long-term debt totaled $2.99 billion as of quarter end, though this does not include recent debt associated with Talen’s acquisition financing. Management also shared that they expect the balance sheet to get stronger over time as the AWS deal ramps, and as more contracts get added, it is expected to “further strengthen the balance sheet and provide for visibility to those cash flows.” 

Management provided an update on total liquidity including revolvers:  

“Our liquidity remains substantial with $1.2 billion of liquidity available for working capital, including approximately $490 million of cash available. Once we close on the Freedom and Guernsey acquisitions, we'll have $200 million more of liquidity as our revolver capacity will increase to $900 million. Excluding the acquisition financing, our leverage ratio is still within our 3.5x net debt to adjusted EBITDA target.” Talen’s current year-end forecast for net leverage ratio is 2.6x, well within its target, though it stands at ~5.7x based on current pro-forma debt (incl acquisition financing). Talen is focusing on debt paydown post-closing to reach its 3.5x target by year-end 2026.  

Valuation Valuation 

Talen’s valuation is trading close to peak levels, though the company has seen substantial multiple expansion over the past two years. Talen is currently valued at 7.2x forward sales, below its peak at 8.5x but well above its 3.4x multiple from March 2025 and its 1.9x valuation from early 2024. 

On the bottom line, Talen trades at 74x forward EPS for FY25, more than double its 34x average, though the strong earnings growth in FY26 to the $20 range brings this down to 19x for next year, well below 2024’s peaks of 38x and in line with its five-year average. 

Notable Risks Notable Risks 

Though Talen looks to prioritize rapid debt deleveraging following the closing of the Freedom and Guernsey acquisitions, debt to cash remains stretched very thin. The valuation does appear stretched when looking at FY25 multiples, though Talen is expected to grow into its multiples on the bottom line quickly next year. 

SanDisk: Shares Up 559% In 2025 On NAND Flash, Enterprise SSD Tailwinds 

Thematic: 9/10
Fundamentals: 6/10 
Valuation: 3/10 

Brief Overview Brief Overview 

NAND flash-based data center (enterprise) solid state drives (SSDs) are an often overlooked but equally important memory component when it comes to AI training and inference. This is because data center SSDs offer higher read-write speeds critical for accessing and transferring data rapidly, along with higher performance and energy efficiency, making them vital for larger-scale AI training and inference workloads.   

NVMe (Non-Volatile Memory Express) is a protocol designed specifically for NAND-flash based SSDs that optimizes performance by reducing latency and increasing data transfer speeds by utilizing the PCIe bus. This helps provide the high throughput and fast data transfer speeds necessary for AI workloads – NVMe SSDs can increase performance by more than 2X versus SATA SSDs.  

Now operating independently after being spun out of Western Digital in February 2025, SanDisk is eyeing growth in the enterprise SSD market with its ‘Stargate’ NVMe SSD products, based on its BiCS8 3D NAND architecture jointly-developed with Kioxia, offering industry-leading capacity, energy efficiency and performance. SanDisk’s Stargate line debuted this year with 64TB and 128TB capacity, but will scale to 512TB by 2027, suitable for managing massive AI datasets and workloads. Data center remains a smaller portion of SanDisk’s revenue, with client (PC/smartphones) and consumer products (SD cards/USB) remaining core to its business.  

Revenue Revenue 

SanDisk reported a strong sequential revenue acceleration in its fiscal Q1, driven by NAND demand outpacing supply and increasing demand in its data center, edge and consumer end markets. Q1 revenue increased 22.6% YoY and 21.4% QoQ to $2.31 billion, accelerating from 8% YoY and 12.2% QoQ growth in fiscal Q4. Higher-than-expected bit growth drove the outperformance in the quarter relative to guidance of $2.1-2.2 billion.  

SanDisk’s Edge segment was the primary growth driver in Q1 with revenue up 30% YoY and 26% QoQ to $1.39 billion, driven by increasing NAND content in PCs and smartphones and a positive PC refresh cycle. Consumer revenue rose 27% YoY and 11% QoQ to $652 million, while data ce3nter revenue was down (10%) YoY but up 26% QoQ to $269 million. 

Q2 revenue was guided to be $2.55 to $2.65 billion, up 38.6% YoY and 12.6% QoQ at midpoint. Revenue growth is expected to accelerate further to 53% YoY in fiscal Q3 and then decelerate slightly to 49% in Q4.  

For fiscal 2026, SanDisk is currently expected to generate revenue of $10.53 billion, up 43.2% YoY. Management expects demand to outpace supply through 2026, creating stronger tailwinds for pricing and growth through the year. 

AI Segment Growth AI Segment Growth 

SanDisk’s data center revenue, as mentioned above, declined (10%) YoY but rose 26% QoQ to $269 million, driven by increasing demand for its ‘Stargate’ enterprise SSD product line. However, revenue contribution remains small, at less than 12% of revenue. 

SanDisk did not provide a numerical guide for Q2 for data center, but management noted that they are expecting sequential growth in the segment throughout fiscal 2026, with two hyperscaler qualifications underway and qualifications with an additional hyperscaler and major storage OEM planned for calendar 2026.  

Management also increased their forecast for data center exabyte growth, explaining that last quarter, exabyte growth expectations were in the mid-20% range, but now are in the mid-40% range. As a result, data center Is expected to be the largest market in NAND on an exabyte basis in 2026, surpassing mobile. 

Earnings  Earnings  

SanDisk stands out for its strong expected earnings growth through fiscal 2026 and fiscal 2027, with adjusted EPS expected to reach almost $21 by then, nearly 7X higher than the $2.99 it earned in fiscal 2025. 

Q1 GAAP EPS was $0.75, a strong improvement from a ($0.16) loss in Q4, though this was down (49%) YoY from $1.46 in the year ago quarter as margins remained lower YoY. Adjusted EPS was $1.22, up 321% QoQ but down (33%) YoY. 

For Q2, SanDisk guided for adjusted EPS of $3.00 to $3.40, up more than 162% QoQ. Adjusted EPS is expected to further increase to $3.67 in fiscal Q3 and $4.67 in fiscal Q4.  

For fiscal 2026, SanDisk is expected to generate $13.02 in adjusted EPS, up almost 336% YoY, while GAAP EPS is projected to be $11.49, up from ($11.32) in FY25 due to the spin off. Fiscal 2027 is expected to see earnings power surpass $20, with GAAP EPS estimated to be up nearly 76% to $20.20 and adjusted EPS up nearly 59% to $20.68. 

Margins Margins 

Margins are lower YoY compared to pre-spinoff margins, but Q1 saw strong sequential margin expansion that is expected to accelerate in Q2.  

• Q1 GAAP gross margin was 29.8%, down 8.8 points YoY but up 3.6 points QoQ. Adjusted gross margin was 29.9%, down 9 points YoY but up 3.5 points QoQ. 
• GAAP operating margin was 8.3%, down 8.3 points YoY but up 5.6 points QoQ. Adjusted operating margin was 10.6%, down 8.2 points YoY but up 5.3 points QoQ. 
• GAAP net margin was 4.9%, down 6.3 points YoY but up 2.7 points QoQ, and adjusted net margin was 7.8%. 

For Q2, SanDisk guided adjusted gross margin to be 41-43%, or up just over 12 points QoQ at midpoint, while adjusted operating margin is implied to be 24.2% at the midpoint of opex guidance, or up 13.6 points QoQ.  

Cash Cash 

SanDisk noted that in Q1 it reached a net cash position, six months ahead of schedule, though debt is still almost equivalent to its cash on hand. Cash flows were quite strong, and adjusted FCF margin showed strong expansion. 

• Operating cash flow was $488 million in Q1 for a 21.1% margin, up from a (7%) margin in the year ago quarter and a 4.9% margin in Q4. 
• Adjusted free cash flow was $438 million in Q1 for a 19% margin, up from a (10.5%) margin in the year ago quarter and 2.6% in Q4. 

Cash and equivalents totaled $1.44 billion while debt totaled $1.35 billion. 

Valuation Valuation 

SanDisk’s valuation is somewhat hard to pin down given the company’s limited history on the public markets after its February spinoff, and its rapid 362% ascent since the end of August. 

SanDisk trades at 3.3x forward PS, having peaked at 4x in November and having traded as low as 0.6x in the summer, prior to its sharp rally. For comparison, this is a lower multiple than its former parent Western Digital at 5.1x forward PS, though the two are focused on different memory market segments with WDC primarily in hard disk drives.  

For forward PE, SanDisk currently trades at an 18.4x multiple, slightly above its 15.8x average from the second half of fiscal 2025 prior to its fiscal year readjustment in June. Since then, shares have traded as high as 21.7x and as low as 3x due to the sharp earnings increase expected in fiscal 2026.   

Notable Risks Notable Risks 

The NAND flash market has historically been quite volatile, and is shifting from significant oversupply in 2023 to expectations for substantial supply shortages through 2026. However, if NAND capacity begins to come online quickly through next year, or if demand for PCs and smartphones falters due to rising memory prices,  the NAND cycle could reverse and lead to pricing pressures cutting into revenue growth and margins. 

Teradyne: Quiet Beneficiary of Growing AI Compute, Memory Demand 

Thematic: 8/10
Fundamentals: 6/10 
Valuation: 1/10 

Brief Overview Brief Overview 

Surging demand for AI compute and memory chips is most obvious in the reports of leading chipmakers such as Nvidia and Micron, yet there are numerous behind-the-scenes beneficiaries of this powerful trend, such as Teradyne. Teradyne primarily provides automated test equipment (ATE) for the semiconductor industry, spanning high-performance processors and networking devices, as well as DRAM/HBM and SSD manufacturing.  

For example, Teradyne’s UltraFLEXplus test system was architected specifically for high-performance AI processors and networking devices, enabling high-efficiency volume production and reducing time to market by up to 20%. Its Magnum 7H is a multi-generational HBM test platform, serving HBM3e and HBM4 needs with upgradability to service HBM4e and HBM5 when these products arrive.  

Teradyne sees strong tailwinds from AI compute and memory through the end of the year, while its robotics revenue remains challenged. Management explained in Q3 that its view for 2H 2025 compute revenue is >50% higher than its expectations just three months prior, while memory test sales more than doubled QoQ in the quarter.  

Over the longer term, the increasing complexity of chips, shift to chiplet or multi-chip modules, and increasing die sizes all increase test intensity. For example, the cost of scrapping racks escalates from the NVL72 to the upcoming NVL144 and NVL576 platforms due to the increase in complexity and size, creating long-term tailwinds for Teradyne’s high-performance SoC and memory test products.  

Revenue Revenue 

Teradyne reported revenue up 4% YoY in Q3 to $769.2 million, accelerating from an (11%) decline in Q2. Sequential growth was quite strong at 18% QoQ, though this was against a slightly soft comp of (5%) from Q2.  

The primary driver of growth was Semiconductor Test Equipment, which saw revenue rise 7% YoY and 23% QoQ to $606 million, or nearly 79% of total revenue. This accelerated from a (12%) YoY decline in Q2. 

Product Test Equipment revenue was $88 million, up 10% YoY and 4% QoQ, while Robotics revenue was $75 million, down (15%) YoY and flat QoQ. 

For Q4, Teradyne guided for revenue to be $920 million to $1 billion, a strong acceleration to 28% YoY and 25% QoQ growth at midpoint. Analysts currently estimate revenue growth to continue accelerating to 41% YoY by fiscal Q2 (June 2026).  

AI Segment Growth AI Segment Growth 

Teradyne’s strongest AI-driven growth came from Memory Test Equipment (MTE), which rose more than 110% QoQ to $128 million, or ~16.6% of total revenue. However, this was down (15%) YoY. MTE revenue was driven primarily by DRAM (LPDDR and HBM performance test demand), which accounted for roughly 75% of revenue, with the remaining 25% coming from flash, primarily SSDs. As mentioned above, Teradyne’s Magnum 7H product for HBM3e, HBM4/4e and HBM5, began volume shipments in Q3.  

System-on-Chip (SOC) revenue saw growth driven by AI compute and networking products, with revenue up 11% QoQ and 12% YoY to $440 million, or more than 57% of total revenue.  

Management did not break out Q4’s guidance for these two segments, noting that AI-driven test demand remains robust across compute, memory and networking. Analysts from UBS placed the QoQ growth in Semi Test Equipment at ~$200 million as a whole, with management explaining that the growth will likely be 2/3 compute and networking-driven and 1/3 memory-driven, primarily by HBM. 

Earnings  Earnings  

Teradyne reported strong sequential EPS growth in Q3 that will continue in Q4, though EPS was lower YoY as margins were a couple of points lower compared to last year.  

In Q3, Teradyne reported GAAP EPS up 53% QoQ but down (16%) YoY to $0.75. Adjusted EPS was $0.85, up 49% QoQ but down (6%) YoY, and beating estimates by 7.5%. 

For Q4, Teradyne guided for GAAP EPS to be $1.12 to $1.39, pointing to QoQ growth of 67% and YoY growth of 39% at midpoint. Adjusted EPS was guided to be $1.20 to $1.46, up more than 56% QoQ and 40% YoY at midpoint. Analysts expected adjusted EPS growth to accelerate sharply over the next two quarters, with current consensus estimates calling for growth of 61% in fiscal Q1 and 112% YoY in fiscal Q2. 

Adjusted EPS for fiscal 2025 is projected to be $3.52 for 9.3% YoY growth, before accelerating to nearly 46% growth to $5.13 in fiscal 2026. 

Margins Margins 

While margins were down YoY, Teradyne reported strong sequential growth and improving operating leverage versus Q2, with the expansion in operating margin at 4X the rate of gross margin. Operating margins are guided to continue expanding at a similar rate in Q4.  

• GAAP gross margin was down 0.8 points YoY but up 1.2 points QoQ to 58.4%. Adjusted gross margin was down 1.2 points YoY but up 1.2 points QoQ to 58.5% 
• GAAP operating margin was down 1.7 points YoY but up 5 points QoQ to 18.9%. Adjusted operating margin was down 2 points YoY but up 5.3 points QoQ to 20.4%.  
• GAAP net margin was down 4.2 points YoY but up 3.5 points QoQ to 15.5%. Adjusted net margin was down 2.3 points YoY but up 3.6 points QoQ to 17.7%.  

For Q4, Teradyne guided for adjusted gross margin to be 57-58%, down 1 point at midpoint on onetime supply costs to meet accelerated demand. Adjusted operating margin was guided to be 24-27%, up 3.6 to 6.6 points at midpoint on a much lower opex run rate at 31-33% of revenue, versus 38.1% in Q3. 

Cash Cash 

Cash flow margins contracted sharply Q3, though this was primarily driven by a large QoQ increase in accounts receivable, providing an extra layer of confidence in the upcoming revenue acceleration in the next couple of quarters.  

• Operating cash flow was $49.1 million for a 6.4% margin, though OCF margin had been >22% for the past five quarters. The sharp contraction was primarily due to a $161 million sequential increase in accounts receivable.  
• Free cash flow was $2.4 million for a 0.3% margin, down from 20.2% in the prior quarter due to the jump in AR.  

Cash and equivalents were $297.7 million in Q3, down from $367.9 million in Q2, while Teradyne took on new debt of $200 million in the quarter (with this being its only debt). 

Valuation Valuation 

Teradyne is trading at peak multiples, with shares currently at 9.9x forward PS, well above its five-year average of 6.6x and just shy of its 10.4x peak. 

On a forward PE basis, Teradyne trades at 55x forward earnings, more than 50% above its five-year average of 35.7x and just off its peak of 58x. Even looking ahead to fiscal 2026, with the ~36 point acceleration for adjusted EPS, Teradyne is trading at a 38x multiple, still above its average. 

Notable Risks Notable Risks 

It is still early in this AI-driven inflection in growth for Teradyne, and management has pointed out that SoC test growth in Q3 and Q4 should not be extrapolated into the early part of next year, as timing for Q1 and Q2 remains uncertain and growth may be lumpy. The thin cash flows are not necessarily a red flag yet, but if margins remain depressed, this could be more of a risk to watch moving forward. 

Dell: FY25 AI Server Revenue Raised to $25 Billion, Up 150% YoY 

Thematic: 8/10
Fundamentals: 6/10 
Valuation: 7/10 

Brief Overview Brief Overview 

Dell’s story is primarily centered around its AI server growth opportunities, benefiting from the ramp of Nvidia’s Blackwell and Blackwell Ultra GPUs, though the company does have other outlets into storage, networking, and commercials PCs and workstations. Dell is working closely with Nvidia to help accelerate enterprise AI adoption via the Dell AI Factory, and notably was the first to deliver Nvidia’s GB300 NVL72 racks to CoreWeave back in July. 

Riding Nvidia’s coattails this past quarter, Dell reported strong AI server metrics in Q3, booking record orders and reaching a new record for backlog, while guiding for record shipments next quarter. Dell also raised its AI server revenue forecast by another $5 billion, now seeing $25 billion this year, up 150% YoY and up $10 billion from its initial $15 billion forecast at the start of the year. Nvidia’s strong visibility into Blackwell and Rubin sales through 2026 also hints that this AI server momentum may persist through next year.  

However, growth in Dell’s consumer and commercial PC business remains low, and the company must battle rather thin AI server margins in a competitive market. Dell did note that AI server margins improved sequentially and helped drive some margin expansion, but the company must prove that this dynamic can be maintained. 

Revenue Revenue 

Dell reported a solid Q3 with revenue of $27.0 billion, though this was around 1.1% shy of consensus estimates for $27.3 billion. Revenue was up 10.7% YoY but down (9.7%) QoQ, as Q2 was positively impacted by a sharp ~356% QoQ increase in AI server shipments to $8.2 billion due to timing of fulfillment.  

Dell’s Infrastructure Solutions Group (ISG) revenue increased by double-digits YoY for a sixth consecutive quarter, up 24% YoY in Q3 to $14.1 billion. This slowed from 44% growth in Q2 to $16.8 billion, though again this was impacted by the strong QoQ growth on shipment timing. Within ISG, Server and Networking revenue was up 37% YoY to a record $10.1 billion, while Storage revenue was down (1%) YoY to $4 billion. 

Dell’s Client Solutions Group (CSG) revenue increased 3% YoY to $12.5 billion, a slight acceleration from 1% YoY growth in Q2, as Dell noted robust international demand and continued growth in Commercial demand. Commercial revenue was up 5% YoY to $10.6 billion, acceleration from 2% YoY in Q2, while Consumer revenue was down (7%) YoY to $1.9 billion.  

For Q4, Dell guided for $31-32 billion in revenue, marking a sharp acceleration to 32% YoY at midpoint and 16.7% QoQ, with nearly 30% of this revenue coming from AI servers. Full-year fiscal 2026 revenue was guided to be $111.2-112.2 billion, up 17% YoY at midpoint. This was a solid $4.7 billion raise from prior guidance for 12% growth to $107 billion at midpoint.  

AI Segment Growth AI Segment Growth 

Dell’s AI server momentum was robust in Q3, with the company reporting record orders and backlog, with a strong QoQ increase in revenue guided for Q4. Dell also raised its full-year AI server revenue forecast to $25 billion, up ~150% YoY.  

AI server revenue was $5.6 billion in Q3, up 93% YoY but down nearly (32%) QoQ as Dell fulfilled some larger orders in Q2. For Q4, Dell guided for $9.4 billion in AI server revenue, a fresh record, and representing YoY growth of 348% and QoQ growth of 68%. 

AI server orders reached a record $12.3 billion, up ~120% QoQ and also surpassing Q1’s $12.1 billion. Fiscal 2026 YTD orders reached $30 billion, more than 3X higher than the $9.4 billion recorded in the same period last year. 

AI server ending backlog rose ~57% QoQ to a record $18.4 billion with a significant shift to the GB300 in the quarter. Dell added that its five-quarter pipeline remains multiples of its backlog and grew sequentially across neoclouds, sovereigns and enterprises.  

Earnings  Earnings  

Dell reported a strong GAAP EPS beat in Q3, with earnings of $2.28, up 39% YoY and 15.7% ahead of estimates for $1.97. Adjusted EPS also beat by 4.5%, rising 17% YoY to $2.59, driven by improved profitability in AI servers and storage.  

For Q4, Dell guided for GAAP EPS growth to accelerate to 42% YoY to $3.05 at midpoint, while adjusted EPS was guided to accelerate to 31% YoY to $3.50 at midpoint. This accelerate is poised to continue into Q1, with adjusted EPS forecast to rise 52.4% YoY to $2.36. 

For fiscal 2026, Dell guided GAAP EPS to be $8.38 at midpoint, up 31% YoY, a $0.40 raise from its Q2 guide of $7.98 for 25% growth. Adjusted EPS was guided to be $9.92 for 22% YoY growth, also a $0.37 raise from its prior guide for $9.55.  

Margins Margins 

Margins were mixed for Dell in Q3, with gross margin rebounding QoQ but remaining down YoY, and operating margin expanding on both a YoY and QoQ basis.  

• Q3 GAAP gross margin was 20.7%, down 1.3 points YoY but up 1.4 points QoQ. Adjusted gross margin was 21.1%, down 1.4 points YoY but up 2.4 points QoQ. 
• Q3 GAAP operating margin was 7.8%, up 0.7 points YoY and 1.8 points QoQ. Adjusted operating margin was 9.3%, up 0.1 points YoY and 1.6 points QoQ. 
• Q3 GAAP net margin was 5.7%, up 0.9 points YoY and 1.8 points QoQ. Adjusted net margin was 6.5%, flat YoY and up 1.2 points QoQ. 

By segment:  

• ISG adjusted operating margin was 12.4%, down 0.9 points YoY but up 3.6 points QoQ. This was driven by storage and sequential improvement AI server profitability, with management saying AI server margins were mid-single-digit, up from an implied 2.5% last quarter.  
• CSG operating margin was 6.0%, down 0.2 points YoY and 0.4 points QoQ.  

Cash Cash 

Operating and free cash flow both declined YoY and QoQ, though adjusted FCF more than doubled YoY due to financing receivables.  

• Operating cash flow was $1.17 billion, down (25%) YoY. OCF margin was 4.3%, down from 6.4% in the year ago quarter and 8.5% in Q2.  
• Free cash flow was $506 million, down (45%) YoY. FCF margin was 1.9%, down from 3.8% in the year ago quarter and 6.3% in the prior quarter. Adjusted FCF, however, was up 133% YoY to $1.67 billion for a 6.2% margin, up from 2.9% in the year ago quarter. 

Cash and equivalents totaled $9.57 billion, while debt totaled $31.24 billion, with $7.39 billion being current.  

Valuation Valuation 

Dell trades slightly above its average forward PS multiple at 0.76x currently versus its five-year average of 0.69x, though the company has rarely traded above 1x over the past two years. 

On the bottom line, Dell trades at 12.9x forward PE, also slightly above its five-year average of 12x, but well below its October high at 16.6x and previous resistance around the 18x level.  

Notable Risks Notable Risks 

The main risk to watch for Dell is AI server margins, and whether or not the company can maintain mid-single-digit margins moving through 2026, considering the competitiveness of the industry between Super Micro and Taiwanese ODMs such as Foxconn and Quanta. 

Additionally, the sharp surge in memory prices is expected to cause substantial PC price hikes to preserve margins – Dell is said to be raising commercial PC prices by as much as 30% as a result. These price hikes could weigh on demand and cause PC growth to slow next year, with IDC forecasting the industry to decline (5%) in a conservative scenario and as much as (9%).  

Micron: HBM, LP Server DRAM Driving Strong Growth, FQ2 to Accelerate to 37% QoQ 

Thematic: 9/10
Fundamentals: 10/10 
Valuation: 6/10 

Brief Overview: Brief Overview: 

Micron is a primary beneficiary of rapidly increasing dollar content of high-bandwidth memory (HBM) chips with each new generation of GPUs. AI training and inference rely heavily on HBM for the massive memory bandwidth that complex models require. AI servers also use more DRAM and NAND than a traditional server. These are reasons that Micron’s cyclical fundamentals could become more secular as the AI economy is built out.  

In fiscal 2025, Micron’s HBM, high-capacity dual in-line memory modules (DIMMs) and low-power (LP) server DRAM revenue reached $10 billion, up more than fivefold from the prior year, while HBM alone reached $2 billion in revenue in Q4.  

Management expects robust AI server demand, the shift to HBM4 and tight DRAM supply to drive substantial records for revenue, gross margin, EPS and free cash flow for fiscal 2026 with business strengthening throughout the year. Micron also expects these tight market conditions to persist beyond 2026. 

What Micron will need to answer is if the cyclical nature of the memory market will smooth out as the dollar content of memory is rapidly increasing. Data center is already proving to be a strong driver of growth for Micron, accounting for 56% of sales in FY25, up from 35% in FY24. On a dollar basis, data center revenue surged 137% YoY to $20.75 billion for the year.  

Overall Revenue Growth Overall Revenue Growth 

Micron reported record Q1 revenue of $13.64 billion, beating estimates by 5.9% and accelerating 10.7 points to 56.7% YoY growth. Sequentially, growth was 20.6% QoQ, just one point slower than Q4’s 21.7% QoQ growth. DRAM products (within that HBM and LPDDR5X) were the primary driver of Q1’s results, with revenue up 69% YoY and 20% QoQ to $10.8 billion, or 79% of revenue. 

Micron also delivered one of the largest beats in AI semiconductors outside of Nvidia’s May 2023 report with its Q2 guidance. Management forecast revenue of $18.7 billion, +/- $0.4 billion for next quarter, more than 31% above estimates for $14.23 billion. This corresponds to a very sharp 75.5 point acceleration to 132.2% YoY growth, while QoQ growth would accelerate to 37.1% at the midpoint of the guide.  

Micron has not provided a full-year guide for revenue, but current consensus estimates call for 98% growth to $73.98 billion in revenue. 

AI Revenue Growth AI Revenue Growth 

In Q1, Micron's data center revenue rose 55% YoY to $7.66 billion (56% of company revenue), with growth primarily driven by DRAM (HBM/LPDDR5X) products and aided by data center SSDs and NAND components.  

Micron’s Cloud Memory Business Unit (CMBU), which consists of its HBM, high-capacity dual in-line memory modules (DIMMs), and low-power server DRAM solutions, saw Q1 revenue of $5.28 billion, up 99.5% YoY. 

Micron’s other data center unit, its Core Data Center Business Unit (CDBU), consists primarily of data center SSDs and NAND components. This unit saw Q1 revenue grow just 4% YoY to $2.38 billion.  

Earnings Earnings 

Micron is expected to see robust earnings growth this fiscal year as margins are rapidly expanding on surging prices. In Q1, Micron reported GAAP EPS of $4.60, up 175% YoY; this also is a sharp uptick from $2.83 in Q4.  

For Q2, Micron guided for GAAP EPS to be $8.19, +/- $0.20, nearly 74% ahead of estimates for $4.71 and corresponding to YoY growth of almost 481%, a 306 point acceleration. GAAP EPS growth is expected to remain >250% for both Q3 and Q4 to $9.37 and $10.04. 

For the full year, Micron is expected to deliver GAAP EPS of $31.17, more than quadrupling from $7.59 in fiscal 2025. Earnings estimates also moved more than 60% higher following Q1’s report and Q2’s blowout guide, moving from $19.42 to the now $31.17 estimate.  

Margins Margins 

Micron’s margin turnaround story has been impressive, with gross margin up nearly 57 points over the last two years and operating margin up 69 points. Micron also guided for substantial records for gross and operating margin in Q2, on the backs of strong pricing. 

GAAP gross margin in Q1 was 56%, up 17.6 points YoY, aided by the strong growth in CMBU which carried a 66% gross margin in the quarter. For Q2, GAAP gross margin was guided to be 67% at midpoint, an 11 point sequential expansion and up 31.2 points YoY.  

GAAP operating margin was 45%, up 12.7 points QoQ and 20 points YoY, again aided by CMBU which carried a 55% margin in the quarter. For Q2, Micron implied operating margin to be 58.7%, up 12.7 points QoQ and 36.7 points YoY, signaling strong tailwinds from surging DRAM prices. 

GAAP net margin was 38.4% in Q1, up 10.1 points QoQ and nearly 17 points YoY.  

Cash Cash 

Operating cash flow was $8.41 billion in Q1, up more than 159% YoY and nearly 47% QoQ. OCF margin was 61.7%, up 10 points QoQ and up 24.4 points YoY.  

Adjusted free cash flow was $3.91 billion in Q1, up sharply from $803 million in Q4 and $112 million in the year ago quarter. Adjusted FCF margin was 28.6%, up from 7.1% in the prior quarter and 1.3% in the year ago quarter. 

Micron reported total cash and equivalents of $12.0 billion and total debt of $11.76 billion. 

Valuation Valuation 

Despite its recent rally, Micron trades at somewhat reasonable multiples, well below its peak from 2024. On the top line, Micron trades at 4.4x forward revenue, 22% above its average 3.6x multiple and far below its peak of 6.8x from mid 2024. 

On the bottom line, Micron trades at 9x forward earnings, though its 40.9x average is skewed higher by 2024’s >100x multiples when margins were razor-thin. Since the start of 2025, Micron has traded as high as 16x forward earnings, and as low as 3x on its fiscal-year reset in September. 

Notable Risks Notable Risks 

Micron’s growth to this point and beyond has been centered around HBM and data center DRAM products, both on the top and bottom lines. Rising HBM demand in 2026 as next-gen GPU systems ramp and content growth in LP server DRAM are strong tailwinds for growth, yet sharply rising DRAM prices from tight supply could cut into demand for consumer electronics products. This is Micron’s second largest segment and growth driver (Mobile and Client), with nearly $4.3 billion in revenue and a 47% operating margin in Q1, and any demand softness from price hikes could be felt more acutely in 2026.

Conclusion: 

We are thrilled about our new tier Discovery as we’ve seen immediate results in 2025 after delivering a separate tier for new ideas. We quickly spotted the limitations around running an active portfolio that does not dedicate a separate effort to new idea generation as the market moves fast with new winners emerging every year. It would be easy to miss up-and-coming momentum stocks without this tier – especially for enthusiastic AI investors, such as ourselves.  

We believe pointing out names that are not often spoken of (yet) while also reiterating those that are well-known yet remain strong quarter-after-quarter allows us to go beyond the I/O Fund portfolio to offer maximum value from our research efforts. 

As we look at Q4 and beyond, we believe this quarterly analysis combined with an actively managed Top 10 list will become a strong offering in 2026. Our cumulative record proves we are one of the strongest teams in the world on AI stocks and 2025 was no exception. Moving forward, our goal is to use our proven methodology to deliver additional value add as we participate heavily in the once-in-a-lifetime trend of AI.  

In the coming weeks, we expect things to shift rapidly as new information is published daily during earnings season. Please reference our Top 10 Watchlist spreadsheet and incoming analysis as critical tools for staying on top of the Must Know stocks in the space. We will also cover these stocks in a Discovery webinar hosted by Knox Ridley early next month. 

Stay tuned for frequent updates!

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock and crypto entries and exits. Beth Kindig offers weekly deep dives including lesser-known cryptocurrencies and AI stocks, plus the team offers trade alerts. The I/O Fund team is one of the only audited portfolios available to individual investors. If you’d like to subscribe to the Advanced Market Signals plan, email us at premium@io-fund.compremium@io-fund.com.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• The I/O Fund’s Top 10 New Ideas List for Q4 2025
• Celestica Eyes FY26 Acceleration on Strong Networking Switch Demand
• Nebius: Financing its Data Center Ambitions Will be Challenging
• Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth

The difference between Palantir and other AI-enabled database competitors is that Palantir is able to go further in answering questions that a model would struggle to answer. Traditional business intelligence companies require a complete data set whereas Palantir is able to tackle situations where there is not a complete data set. You can think of the competitive advantage as being actionable depth, which Palantir has described as “the reasoning that goes into decision-making, not just data.”    

Palantir’s Artificial Intelligence Platform (AIP) integrates generative AI with operational data and workflows, and when combined with Palantir’s other platforms Foundry and Apollo, it provides an AI service mesh that can run hundreds of microservices, scale compute through its Rubix engine and orchestrate updates through Apollo.   

Additionally, Palantir’s knowledge graph referred to as Ontology is a distinct advantage. The graph offers better context than a large language model would on its own – or as Palantir states, it’s “the reasoning that goes into decision-making.” Palantir made some key upgrades to AIP with the introduction of AI forward-deployed engineers (FDEs) and AI Hivemind, and brought Ontology to the edge, allowing it to be deployed on mobile devices. 

Fueled once again by AIP, Palantir delivered one of the best reports across tech in the third quarter, with revenue accelerating nearly 15 points sequentially to almost 63%, with strong growth in key metrics and a 50 point acceleration in US Commercial revenue since the start of the year. We break this down and more below. 

New Product Upgrades – AIP FDEs, Hivemind and Edge Ontology 

Palantir made a handful of upgrades to its AIP and Ontology in Q3, unveiling AIP forward-deployed engineers (FDEs) in beta, AI Hivemind, and Edge Ontology, all aimed at accelerating AI deployment with its customers.  

AI FDEs build upon Palantir’s take on software engineers, the forward deployed engineer, which sits at the intersection of software, sales, and platform engineering, embedding within customer teams to closely develop and tailor AI software and solutions directly to their needs. Palantir now brings this to AIP, with the AI FDEs being its AIP-native deployment AI agent that “understands how to connect to data sources, how to integrate and transform data, how to create ontologies and functions and build applications.” AI FDEs function in conversational commands, allowing customers to easily turn requests into autonomously-executed Foundry operations. 

Palantir says the AI FDEs are increasing productivity for customers, noting that at one customer, two of its human FDEs utilized AI FDEs to migrate a legacy data warehouse in five days, a task which Palantir says would normally have taken up to two years.  

AI Hivemind is a new tool within AIP that Palantir says will orchestrate a “swarm of dynamically generated agents to tackle hard problem solving, idea generation refinement and executable proposal generation that is integrated with Ontology and therefore aware of the context of your enterprise.” Palantir says the tool was developed to help government clients solve extremely complex problems with classified data (such as generating intricate mission plans), but it has already been tested by commercial clients to help “identify bottlenecks to their supply chain, proactively developing possible solutions and then leveraging AI FDE to code that up into an actual solution.” 

Edge Ontology is Palantir’s new, lightweight Ontology that allows it to run on mobile devices, letting customers build mobile apps or embedded software for hardware such as robots or drones. Edge Ontology is also fully integrated with AIP. While Palantir was very thin on details, it’s likely tied to its existing partnership with Qualcomm, where the two brought Ontology to edge devices powered by Qualcomm’s Dragonwing processors. This partnership focused on the industrial, auto and manufacturing sectors, and remote and offline environments. 

Together, the new product upgrades reflect Palantir’s dedication to continuously improve its platform and help customers consistently solve problems daily. Financially, the three new upgrades can help accelerate deployments and adoption of AIP, help secure more (and potentially larger) contracts, and tap into new markets such as IoT at the edge. 

Financials 

Revenue Accelerates Nearly 15 Points to 63% YoY, Up 18% QoQ 

Palantir reported $1.18 billion in revenue in Q3, up an impressive 18% QoQ and beating estimates by 8.4%, driven by unwavering momentum in US Commercial. On a YoY basis, revenue growth accelerated 14.8 points to 62.8% YoY, the largest sequential acceleration to date and marking Palantir’s highest growth rate since going public. Over the last nine quarters, topline growth has accelerated ~50 points, from just 12.7% in Q2 2023, a rare feat to accomplish. 

For Q4, Palantir guided for revenue up 60.6% YoY to $1.327 to $1.331 billion, well ahead of estimates for 44.2% growth to $1.19 billion. While this does represent a marginal deceleration at face value, this sequential deceleration is in line with trends from previous quarters.  

For the full year, Palantir raised its revenue outlook to $4.396 to $4.400 billion, pointing to YoY growth of 53.5% at midpoint, a sharp acceleration from 29% growth in 2024. To put in perspective the strength of this acceleration, Palantir had initially guided for just 30.9% growth to $3.76 billion in revenue back in Q4 2024; growth is now more than 22 points faster.  

Impressive 28 Point Acceleration in US Commercial to 121% YoY 

Palantir’s US Commercial segment is generally seen as the primary vector for its AIP-driven growth, with robust momentum only accelerating further in Q3.   

US Commercial revenue grew 29% QoQ and 121% YoY to $397 million in Q3, accelerating from 93% YoY growth in Q2. Since the start of the year, US Commercial growth has accelerated 50 points, and since the start of 2024, growth has accelerated 81 points.  

For the full year, Palantir significantly boosted its US Commercial growth outlook to >104% YoY, up from 85% previously. This corresponds to revenue of $1.433 billion, up from $1.302 billion previously.  

Key metrics for the segment were very strong: TCV closed (total contract value) surged 342% YoY to a record $1.31 billion, and TTM TCV was $3.8 billion, up 217% YoY. Remaining deal value (RDV) rose 199% YoY and 30% QoQ to $3.63 billion. US Commercial deals closed of >$1 million were up 2X YoY and deals closed of >$5 million were up 5X YoY. 

Key Segments – Government and Commercial Revenue 

Q3 also marked the first time Commercial growth outpaced Government segment growth since Q2 2024, fueled by the robust momentum and sharp acceleration in US Commercial as discussed above.   

Commercial revenue rose 21.5% QoQ and 73% YoY to $548 million, a 26 point acceleration from 47% YoY growth last quarter. International Commercial revenue was ~$151.7 million in Q3, up ~10% YoY and 5% QoQ. 

Government revenue rose 14.5% QoQ and 55% YoY to $633 million, a six point acceleration from 49% YoY in Q2. Government remained Palantir’s largest segment at ~53.6% of revenue. US government revenue was up 52% YoY and 14% QoQ to $485.9 million, while International Government revenue was ~$146.8 million, up nearly 66% YoY and 16% QoQ.

Other Key Metrics – NRR Expands, Strong Customer Growth 

Palantir had a handful of other strong key metrics that support strong revenue growth continuing despite the sharp acceleration the company delivered in Q3.   

Palantir’s net retention rate (NRR) expanded six points sequentially to 134%, and over the past two years, NRR has risen an impressive 27 points, with Palantir noting that AIP is continuing to drive existing expansions and new customer conversions. Palantir also continued to emphasize that NRR does not include revenue from new customers acquired over the last twelve months, and accelerating momentum in quarterly deals closed supports more upside to NRR in 2026. 

Moving to deals closed, Palantir closed 201 deals of >$1 million in Q3, up 30% QoQ; this was a sharp acceleration from 13% QoQ growth in Q2. Palantir also signed more deals in all of its cohorts (>$1M, >$5M and >$10M) in Q3 than it had in Q2.  

To put deal growth in the context of NRR, Palantir has signed 629 >$1M deals over the last twelve months, up more than 61% from 390 in the same period last year – with none of these new deals appearing yet in NRR. 

However, there were a few blemishes within key metrics – billings growth decelerated from 53.5% YoY in Q2 to 48.8% YoY in Q3 to $1.23 billion, with QoQ growth also decelerating from 21.8% QoQ to 11.2% QoQ. RPO growth decelerated from 77% YoY and 27% QoQ in Q2 to 66% YoY and 8% QoQ in Q3 at $2.60 billion. 

Margins – Q3’s Rule of 40 of 114%, from 94% in Q2 

Margins strengthened considerably in the quarter, with adjusted operating margin surpassing 50% with more expansion guided for Q4. Palantir’s Rule of 40 score (revenue growth + adj operating margin) expanded to a wild 114%, up from 94% last quarter and 68% last Q3.  

Gross margin was 82% in Q3, up one point QoQ and two points YoY, while adjusted gross margin was 84%, up two points YoY and QoQ. 

GAAP operating margin was 33%, an impressive 6 point QoQ and 17 point YoY expansion. Adjusted operating margin was 51%, breaking past 50% for the first time and up 5 points QoQ and 13 points YoY. For Q4, Palantir guided for adjusted operating margin to be 52%, showcasing its ability to drive strong margin expansion alongside swift revenue acceleration. Full year adjusted operating margin guidance was raised from 46% to 49%.  

GAAP net margin was 40%, up 7 points QoQ and 20 points YoY. Adjusted net margin was 45%, up 5 points QoQ and 12 points YoY. Palantir is one of the few, if not only, tech companies to have 40% GAAP net margins with revenue growth accelerating above 60%. 

Earnings 

Palantir reported $0.18 in GAAP EPS in the quarter, up 200% YoY, while adjusted EPS was $0.21, beating estimates by 25.5% and rising 110% YoY. Palantir did not provide a specific guide for EPS for Q4, though current estimates are pegged at $0.12 in GAAP EPS and $0.22 in adjusted EPS, up 300% YoY and 57% YoY, respectively. 

For FY25, Palantir is expected to earn $0.72 in adjusted EPS, up nearly 76% YoY, before slowing to 39% growth to $1.01 in FY26. 

Cash 

Cash flows were strong, though cash flow margins dipped on a YoY and QoQ basis. Operating cash flow was $507.7 million for a 43% margin, shrinking from a 54% margin in Q2 and 58% in the year ago quarter.  

Adjusted free cash flow was $539.9 million for a 46% margin, down from 57% in Q2 and 60% in the year ago quarter. Palantir raised its adjusted FCF guidance for the year to $1.9 to $2.1 billion, or a 45.5% margin, up from a 42.8% margin previously. 

Cash and equivalents totaled $6.4 billion and debt remained zero.  

Valuation 

Valuation is the crux for Palantir as the stock trades at 100x forward revenue, nearly triple its five-year average of 36x and in rather uncharted territory for software stocks. On the bottom line, Palantir trades at 256x forward adjusted EPS despite a >40% margin, more than double its average 109x multiple.  

The valuation with Palantir is a gamble as the company is attempting to set a new bar for AI software, with >100x forward sales multiples only achieved for short fashion in 2021 for a handful of prior market darlings, whose stocks have yet to return to those prices. This elevated valuation may also present a risk if/when the company reaches peak revenue growth as comps will quickly get tougher.  

Conclusion 

Palantir delivered one of the strongest earnings reports in all of tech in Q3, with revenue growth of nearly 63% and a strong 28 point acceleration in its AI-driven US Commercial segment. Since the start of 2025, revenue growth has accelerated more than 23 points, with US Commercial revenue accelerating 50 points. Trends in other key metrics such as NRR and quarterly deals closed remain robust, although billings and RPO growth decelerated in the quarter. However, what Palantir has to contend with is an extended valuation, in uncharted territory even by its own measures, and where the market will ultimately price its shares.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund do not own shares in PLTR at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

• Positions Report – October 2025
• Cloudflare: Revenue Accelerates to >30%, Key Metrics Strengthen
• The AI Memory Boom Has Arrived
• Meta: Growth is Quietly Benefitting from AI, Though Margin Risks Weigh Heavy

As we have discussed for our Discovery and Pro members, AI networking is one of the strongest trends for this year and next, driven by scale-up and scale-out networking to support larger GPU racks and accelerating GPU cluster sizes. Celestica is an under-the-radar beneficiary of this trend, capitalizing on strong demand for 800G and 1.6T Ethernet networking switches and leveraging its deep ties to hyperscalers.  

Celestica guided for one of the most impressive accelerations seen in this last quarter of earnings, underpinned by its 800G switches accelerating next year with 1.6T on deck. For 2026, Celestica expects revenue growth to accelerate around five points to 31% YoY in 2026, whereas consensus had been pegged at just 17% YoY. This strong upside is being driven by networking and custom AI compute platforms with visibility into 2026-2027. 

In terms of AI revenue, Celestica’s Cloud and Connectivity Solutions (CCS) segment is guided to generate $9 billion in revenue in 2025, up ~40% YoY, accounting for nearly 74% of total revenue. CCS, which includes AI networking, server, storage and rack-scale system solutions, is Celestica’s main growth driver, expected to grow ~40% annually in 2026 and 2027. 

Celestica is closely linked to Broadcom’s networking platforms as a key vendor, serving major customers such as Google and Meta, with some of its notable product engagements including Google’s TPU server racks, and Meta’s Minerva ASICs servers, Wedge400 switches and also its next-gen Tomahawk5-based 400G AI fabric switch Minipack3. Additionally, management’s commentary suggests that OpenAI could become a key customer as soon as 2027. 

Below, we cover Celestica’s strategic positioning in the high-bandwidth Ethernet market, its engagements with hyperscalers and upcoming platform ramps, its updated growth outlook for FY26 and beyond, and more.  

Celestica’s Strategic Positioning in Custom Networking Switches 

Celestica is strategically positioned in the AI supply chain as it provides hyperscalers with highly customized data center networking switches, servers and storage platforms, alongside custom rack-scale integration services.  

Celestica is also closely aligned with Broadcom, as a preferred provider offering customized high-performance Ethernet switches based on its Tomahawk platform and integrated XPU-based racks and systems. CEO Rob Mionis explained that when Broadcom launches new silicon, such as its newest Tomahawk6, “they’ll work with us to develop products, and those products end up in the major hyperscalers.”  

Growth opportunities are primarily centered around its high-bandwidth Ethernet switch portfolio focused on back-end networking, with the company being the leading supplier with 41% share of the >200G switch market through Q2, and with 55% share of the custom switch market (up from 40% in 2024).  The back-end networking positioning is important for Celestica as it means the company is exposed to the faster-growing segment of Ethernet switching – the back-end TAM is forecast to grow at a 56% CAGR through 2029 on scale-out, and potentially soon, scale-up demand, whereas front-end (user-facing) is forecast to grow at a 20% CAGR.  

Per management, the back-end also sees a much faster refresh rate of every 18-24 months versus >5 years for front-end deployments, and it adopts the newest and fastest bandwidths (800G and soon 1.6T) due to the greater performance and reliability requirements of GPU-to-GPU and rack-to-rack communications.  

Celestica’s primary products include scalable top-of-rack switches and high-bandwidth Ethernet switches (>400G). Its 100G and 400G switches are optimized for data center leaf-and-spine deployments.  

For 800G switches, Celestica’s DS4100/DS4101 are based on Broadcom’s Tomahawk4 portfolio and the DS5000 is based on the Tomahawk5, targeting high bandwidth data center leaf-spine, and top-of-rack applications. For additional clarity on 800G switch dynamics, management explained that they have seen “tremendous growth in 800G this year to the point where we'll end 2025 with roughly a 50% split between 800G and 400G in terms of the products that we're delivering. As we look into 2026, we're seeing the 800G demand, in particular, accelerating.” 

For 1.6T, Celestica will offer the DS6000 and DS6001, based on Broadcom’s upcoming Tomahawk6 offering 102.4Tbps bandwidth, with availability later in 2026. The DS6000 comes in an air-cooled version with linear pluggable optics (LPO) to improve power efficiency, while the DS6001 features hybrid-cooling and the first to integrate direct-to-chip liquid cooling.  Both 1.6T switches are optimized for AI back-end networking (scale-out and scale-up), as well as large-scale AI fabrics for AI training and inference for frontier model sizes. Management expects the 1.6T upgrade cycle to emerge in late 2026 but primarily land in 2027, with one customer giving visibility to a back-half 2026 ramp and multiple other ramps occurring through 2027. 

Celestica is also already making early investments for 400G SerDes to support 3.2T switch platforms, though it does not expect 3.2T mass production to arrive until 2028. Management is also preparing for co-packaged optics (CPO) and other interconnect types such as co-packaged copper (CPC), and while it sees some potential CPO shipments with 1.6T, it does not expect CPO to emerge in full-force until the 3.2T cycle.  

Although these switch products can be highly customized, they support open-source networking stacks such as SONiC, offering hyperscalers flexibility in deployments, facilitating integration into existing software and hardware ecosystems, and letting customers avoid vendor lock-in. This is furthered with Celestica’s extensive Circular Services offering, spanning hardware lifecycle management, remanufacturing, and refurbishment to extend hardware lifetimes and reduce TCO.  

ODM Pivot Driving Hyperscaler Growth, with Google and Meta Key Customers (and Soon Likely OpenAI) 

This positioning in custom Ethernet switches also pushes Celestica towards more of an ODM (original design manufacturer) model from a traditional EMS contract manufacturer, as it engages more deeply across the design and engineering phase, tailoring products exactly to its hyperscaler customers’ needs. Some of its notable confirmed/implied hyperscaler product engagements include Google’s TPU server racks, and Meta’s Minerva ASICs servers, Wedge400 switches and also its next-gen Tomahawk5-based 400G AI fabric switch Minipack3. 

Here is an example of what Celestica’s involvement would look like, such as on Meta’s Wedge400, its top-of-rack networking switch based on Broadcom’s Tomahawk3 from 2021:  

• Celestica works with Meta on system requirements and finalizes system level architecture.   
• Celestica is fully responsible for hardware design and complete prototyping, along with functional and reliability tests and diagnostic software development. 
• After this, Meta reviews Celestica’s engineering design and test reports and moves to production. 

Management’s discussion on its first-of-kind rack-scale liquid cooled 1.6T networking win with a hyperscaler (likely to be Meta for its upcoming Santa Barbara racks) also shed light on why Celestica continues to win these engagements: 

“The customer required an accelerated road map to allow the solution to be early to market, leveraging Broadcom's Tomahawk 6 SC silicon, making speed to market a key consideration. In addition, the customer required multi-node manufacturing capabilities in Asia and the U.S. to support the delivery of the program. As with many of our key engagements, managing complexity was a defining factor. 

Celestica was awarded the program earlier this year based on a strong working relationship with the customer and their confidence in our industry-leading design engineering. They also valued our advanced manufacturing capabilities, specifically our ability to operationalize highly complex production lines for liquid cooled racks at scale and to do this faster and more seamlessly than other potential partners. 

After receiving initial Tomahawk 6 samples earlier this year, we quickly stood up an operational prototype for the 1.6T switch and believe we were the first team anywhere to have done so. The program is scheduled to begin mass production next year.” 

This ODM pivot and ability to co-design and manufacture highly customized, next-gen networking switch and rack solutions at speed is quite visible in Celestica’s hyperscaler growth trajectory, as hyperscalers are expected to account for ~$6.93 billion of revenue in 2025, up from $2.19 billion in 2022, an incredible ~47% CAGR. However, the ODM positioning also presents a risk as even a shift to higher complexity, higher value products may be unable to produce continuous margin expansion into the low-teens. 

Also aiding this hyperscaler growth is a high level of stickiness and deep customer engagement across data rate upgrades. Management explained that they have been able to consistently upgrade all customer engagements each cycle:  

“When you look at where we carved out this industry-leading position in networking, it started in 400-gig, and we were able to translate all of those engagements into 800-gig. And those engagements have been expanding incrementally to new opportunities, and we fully plan to translate all of our 800-gig engagements into 1.6T as well, and we're on track to do that.” Management quietly dropped that they currently have ten programs underway with 1.6T. 

Digital Native Customer Win and 2027 Ramp (Hint: It’s Likely OpenAI) 

Back in January, Celestica announced a “a significant new HPS win with a leading digital native company, who is a pioneer in the commercialization of AI applications,” collaborating with this company to “deliver a full rack, which is an optimized AI system solution built around the customer's custom ASIC accelerator.” The rack-scale solution also will include Celestica’s 1.6T switches and rack-level cooling and connectivity. At the time, Celestica said that “production for this program is expected to begin ramping in the latter part of 2026.” 

In Q3, Celestica provided an update, saying that the “design work for this program is well underway, and we expect to receive initial XPU deliveries in the second half of 2026 to support early test deployments with full-scale production expected to commence in 2027.” Management did add that 2026’s $16 billion revenue outlook does not include any contribution from this customer, but if the custom silicon “is available sooner for mass production, then we may be able to produce sooner.” 

Broadcom’s discussion around its 10GW commitment from OpenAI likely confirms that OpenAI is Celestica’s new digital native customer, with Broadcom saying that 2026 contribution from OpenAI is expected to be minimal with the 10GW deployments concentrated in 2027 through 2029.   

This would potentially be a landmark deal for Celestica in the long-run, as it is an entirely new customer with the potential to add several billion in annual revenue; management said in the original announcement that “demand from this customer at scale could achieve a level similar to those of our largest hyperscaler customers today.” For context, this would compare to Celestica’s largest hyperscaler contributing 28% of revenue in 2024, or ~$2.7 billion, implying OpenAI’s revenue could match that once it ramps.

Financials 

Celestica’s financials are somewhat mixed – on one hand, the company expects strong switch demand to drive its Cloud and Connectivity Solutions (CCS) segment revenue up ~40% YoY in both 2026 and 2027, yet gross and operating margins are quite low compared to its key supplier Broadcom.   

CCS Revenue to Grow ~40% Annually Through 2027 

Celestica was also one of the few companies to really provide a solid long-term growth outlook in its AI-related segment this past quarter, with management confident in maintaining ~40% annual growth in CCS through 2027. While this is already reflected by consensus estimates (meaning Celestica will need substantial upside as growth expectations are already being baked in to shares), there are five main factors that could push growth above and beyond this guide – more on this below. 

For 2026, Celestica guided for approximately 40% YoY growth in CCS to ~$12.6 billion, up from $9 billion guided for 2025, supported by views for accelerating 800G demand, early 1.6T ramps and the ramp of its next-gen AI compute platform to full-volume.  

Driven by this growth in CCS, Celestica guided for initial revenue of $16 billion in 2026 during Q3’s report, nearly 18% ahead of the consensus estimate for $13.6 billion. This would correspond to ~31.1% YoY growth, a five point acceleration from FY25.Supporting this, management says they “currently have about 12 to 15 months of real solid forecast inputs and demand inputs from our customers,” and in many cases, visibility extends beyond that and longer into 2027. For example, some customers have a “certain amount of ASICs, for example, that they may have committed to, and it gives us some assurance as to the longevity and the size of the overall program.”  

Also backing up the guidance is capacity, with Celestica explaining in Q2 that it can support “$3 billion to $4 billion of additional revenue” across its footprint in Thailand, Malaysia, Mexico and the US; Celestica is also aiming to expand production of 800G switches and add capacity for thousands of advanced AI racks annually by 2027. 

For 2027, management explained that it is around 12 months too early to provide concrete numbers, but “right now, we think at least 40% [CCS growth] into 2027 is what we have visibility to,” at least 40% [CCS growth] into 2027 is what we have visibility to,” with opportunities to potentially accelerate that growth. Underscoring this strong outlook includes solid visibility into significant new program ramps starting in 2027 (multiple 1.6T ramps with hyperscalers), scale-up engagements translating to production and revenue, a next-gen custom ASIC platform, and mass-production of the rack-scale custom AI system with the new digital-native customer (likely OpenAI).  

Thus, assuming ~40% growth in CCS and comments for high-single digit (6-8%) YoY growth in its Advanced Technology Solutions segment (ATS/focused on aerospace, industrial and semicap equipment), a reasonable initial estimate for Celestica’s 2027 revenue would be ~$21.3 billion. This would mark a slight two point acceleration to 33% YoY. 

Five Factors Supporting Growth Accelerating Beyond 31-33% 

There are five main factors that support Celestica’s revenue growth being materially faster than the initial 31% and 33% implied growth for 2026 and 2027, as each of these five factors all exhibit growth rates in excess of Celestica’s guidance. 

While there is no guarantee that Celestica can match or exceed some of the growth rates in the opportunities below, these five factors provide ample evidence of market conditions that can support higher growth.  

1) Ethernet Switch Demand Growing for Back-End Networking 

As we had discussed in our Top 10 New Ideas report, networking is at the heart of the new architecture that Nvidia is shipping now as the increased bandwidth is instrumental in driving higher performance. The majority of growth is expected to be driven by back-end networking — scale-up and scale-out networks.  

Scale-out is where the near-term growth opportunities for Ethernet switches lie, despite Ethernet adoption and revenue share historically lagging InfiniBand by a wide margin. High-bandwidth Ethernet switches are seeing strong demand in recent quarters as hyperscalers pivot away from Nvidia’s lock-in ecosystem of GPU + InfiniBand. Arista has said that momentum for Ethernet “has really shifted in the last year” while Nvidia touted that its new Spectrum-X Ethernet is annualizing at $10 billion in revenue. This is also validated by some of the largest operational GPU clusters of today, such as xAI’s Colossus, utilizing Ethernet for the back-end fabric (Nvidia’s Spectrum-X).   

These proof points support bullish growth forecasts for the Ethernet switching market over the next few years. Through 2025 to 2029, the high-bandwidth Ethernet switch TAM is projected to rise at a 30% CAGR, driven by >800G rates rising at a 54% CAGR. In dollar terms, the market is expanding from ~$18 billion to nearly $50 billion over the period. 

It is still quite early for the scale-up opportunity, as Broadcom and others just introduced the ESUN consortium (Ethernet for Scale-Up Networking) a few months ago. Scale-up is inherently linked to Broadcom’s 102.4T Tomahawk6 platform, which, as we had discussed in our recent newsletter, Broadcom Stock: The Silent Winner in the AI Monetization Supercycle, paves the way for >100K to 1 million accelerator clusters by allowing larger leaf-spine fabrics to be constructed, while drawing less power and keeping latency low.  

Broadcom’s management points toward the flattening of the AI cluster as an important catalyst for this product, stating: “[…] Tomahawk 6 enables clusters of more than 100,000 AI accelerators to be deployed in just two tiers instead of three … this flattening of the AI cluster is huge because it enables much better performance in training next-generation frontier models through a lower latency, higher bandwidth and lower power.”  

Broadcom already sees multiple >100K accelerator deployments using Tomahawk 6 for both scale-out and scale-up interconnect, with bookings at record rates. As such, Celestica sees scale-up as an “emerging multibillion-dollar new market opportunity,” having already secured some program wins for its first scale-up solutions leveraging Tomahawk6.  

Main takeaway: A majority of Celestica’s switch deployments already go to back-end networking, with >800G growth expected to rise at a 54% CAGR through 2030, 14 points faster than CCS revenue. Main takeaway: A majority of Celestica’s switch deployments already go to back-end networking, with >800G growth expected to rise at a 54% CAGR through 2030, 14 points faster than CCS revenue.  

2) Broadcom’s $73 Billion Backlog and AI Revenue CAGR 

Broadcom provided an update on its backlog in early December with its fiscal Q4 results, saying that its total AI semiconductor backlog was >$73 billion, with AI switch backlog exceeding $10 billion. Tomahawk6 was booking at record rates, with management later clarifying that TH6 is one of the “fastest-growing products in terms of deployment that we've ever seen of any switch products.”  

Broadcom expects the $73 billion backlog to be delivered over the next six quarters, and this is also expected to be a baseline, with CEO Hock Tan explaining that “we fully expect more bookings to come in over that period of time.”  

Broadcom also provided a strong AI revenue guide for FQ1 of $8.2 billion, implying a 100% YoY and 26% QoQ growth, primarily driven by custom AI accelerators and Ethernet switches. For 2026 and 2027 AI revenue, BofA analysts are already laying the tracks for $50 billion and $100 billion, implying growth at a 122% CAGR if this pans out. This is notably 2X faster than Broadcom’s previously-laid-out 60% serviceable addressable market CAGR of 60%.  

Analysts from RBC also believe that Broadcom’s AI semiconductor revenue acceleration “bodes well” for Celestica’s Q4 as the numbers were “’directionally positive’ for Celestica’s near-term business momentum” in CCS. 

Main takeaway: Celestica’s close ties to Broadcom suggest that its accelerating XPU and networking driven momentum and backlog growth could drive a stronger acceleration for CLS.Main takeaway: Celestica’s close ties to Broadcom suggest that its accelerating XPU and networking driven momentum and backlog growth could drive a stronger acceleration for CLS. 

3) ASICs CAGR Forecasts  

The ASICs market is forecast to see substantial growth over the next few years, as a handful of major hyperscalers pursue ASICs-based AI platform roadmaps. Marvell has projected the ASICs market to rise at a 47% CAGR from 2023 through 2028, rising from $6 billion to $40.8 billion, or nearly 7X growth in five years.  

On the other hand, 650 Group has a slightly more bullish forecast, projecting the ASICs TAM to rise at a 54% CAGR from $18 billion in 2025 to $104 billion in 2029, or a roughly 6X increase over the next four years. Broadcom’s rising backlog and increasingly large ASICs orders — $10 billion and $11 billion with Anthropic and the 10GW commitment from OpenAI – support strong growth as Google and Meta continue to build expand their ASICs platforms.  

Main takeaway: Rising demand for ASICs over the next few years can drive stronger growth for Celestica as its solutions are almost exclusively focused on ASICs platforms. Main takeaway: Rising demand for ASICs over the next few years can drive stronger growth for Celestica as its solutions are almost exclusively focused on ASICs platforms.  

4) Google’s TPU Shipments Accelerating 

While the market continues to debate the TPU vs GPU story, there are some reports from analysts that see TPU volumes accelerating over the next few years. For example, Morgan Stanley projects Google’s TPU shipments to be 1.75 million in 2025, with initial contribution of ~0.5 million from TPU v7 Ironwood.  

For 2026, the firm projected Ironwood shipments to rise 5X to 2.5 million, driving total TPU shipments up ~83% YoY to 3.2 million. For 2027, Morgan Stanley boosted its shipment forecast by ~67%, from 3 million to 5 million, driven mostly by TPU v8 and future generations, while its 2028 forecast was boosted 120% from ~3.2 million to 7 million. This would imply 2027 and 2028 growth of ~56% and ~40%, up from (6%) and 19% previously.  

There are some unsubstantiated claims that Celestica could generate $500 million in revenue per 1 million TPUs shipped, and while this is unverified, if Celestica can capture that amount ($500) per chip, the TPU linked opportunity could be increasingly large over the next few years.  

Main takeaway: Celestica could see solid revenue tailwinds linked to Google’s TPUs if shipments accelerate per some analyst estimates, assuming it remains engaged on the platform.Main takeaway: Celestica could see solid revenue tailwinds linked to Google’s TPUs if shipments accelerate per some analyst estimates, assuming it remains engaged on the platform. 

5) Meta’s Capex 

Meta’s capex strategy is to “aggressively front-load building capacity” to prepare for the most optimistic cases on when AI superintelligence will arrive, with the company outlining substantial capex growth in 2026. Meta aims to meet these capacity needs by “both building our own infrastructure and contracting with third party cloud providers.” 

As a result, Meta expects capex dollar growth to be “notably larger in 2026 than 2025,” implying 2026 capex of at least $103 billion, as current guidance for 2025 at $70-72 billion implies a minimum of ~$32 billion YoY growth. However, considering management’s comments for notably larger dollar growth, there is potential for capex to come in at or above $110 billion, up ~55% YoY, above current estimates for $107.9 billion.  

Main takeaway: Meta’s capex is expected to grow a minimum of >45% in 2026 as the company spends heavily on AI data center infrastructure, potentially driving faster growth for Celestica as it has worked closely with Meta on prior products (and potentially the upcoming next-gen AI rack ramp in Q4). Main takeaway: Meta’s capex is expected to grow a minimum of >45% in 2026 as the company spends heavily on AI data center infrastructure, potentially driving faster growth for Celestica as it has worked closely with Meta on prior products (and potentially the upcoming next-gen AI rack ramp in Q4).  

Key Risk – Communications Growth Decelerating Sharply QoQ 

Celestica’s CCS segment is broken down further into two other subsegments – Communications (networking) and Enterprise (servers and storage). This breakdown highlights a key risk moving to Q4, as guidance implies Communications revenue will decelerate sharply QoQ, an odd print considering the strong ramp in switching products. 

In Q3, revenue from CCS segment rose 43% YoY to $2.41 billion, driven by an 82% YoY increase in Communications revenue to $1.94 billion offsetting a (24%) decline in Enterprise to $477 million on an AI program transition with a hyperscaler.   

For Q4, CCS revenue is implied to accelerate nine points to 52% YoY, with Communications growth guided in the high-60s YoY and Enterprise guided in the low-20s as the new AI program is set to ramp. Despite the seemingly strong guide in Communications, QoQ growth would be just 1% QoQ, a sharp deceleration from Q3’s 18% QoQ growth. 

This would mark Communications’ lowest sequential growth in the last two years, and its first time reporting single-digit sequential growth in the last seven quarters, raising a potential red flag considering Communications is primarily driven by networking/800G switches.  

A likely explanation of this could be the strong outperformance in Communications in Q3 – guidance was for low-60s YoY growth, which Celestica beat by ~20 points. As a result, QoQ growth was likely expected to be ~4%, but came in at 18%, possibly representing a much stronger-than-expected ramp of 800G platforms in the quarter.  

Margins 

Margins continued to expand in Q3, with some signs of operating leverage arising from strong Communications growth as operating margin expanded by 4.7 points YoY versus a 2.6 point YoY expansion for gross margin.  

• GAAP gross margin was 13.0% in Q3, up 0.2 points QoQ and 2.6 points YoY. 
• GAAP operating margin of 10.2%, up 0.8 points QoQ and 4.7 points YoY. Adjusted operating margin was 7.6%, up 0.2 points QoQ and 0.8 points YoY.  
• GAAP net margin of 8.4%, up 1.1 points QoQ and 4.8 points YoY. However, adjusted net margin was just 5.7%, up just 0.1 points QoQ and 0.7 points YoY due to a $113 million impact from gains on total return swaps.  

For a segment breakdown: 

• CCS adjusted gross margin was 12.1% in Q3, up from 11.9% a year ago. CCS adjusted operating margin was 8.3%, up from 7.6% a year ago. Positioning as an ODM may not find much more margin upside even as higher-margin products such as advanced AI rack systems ramp.  
• ATS adjusted gross margin was 10.6%, up from 8.4% a year ago. ATS adjusted operating margin was 5.5%, up from 4.9% a year ago. 

For fiscal 2025, Celestica guided for adjusted operating margin to be 7.4%, and for 2026, only a slight 0.4 point increase to 7.8% despite the 31% growth on the top-line. This suggests that its positioning primarily as an ODM may limit future upside to operating margins even as hyperscaler-linked revenue and higher-margin, higher-complexity designs increase in its mix:  

“So we continue to see the benefits of both operating leverage as well as positive mix in our numbers, on track for about 7.4% operating margin at the company level for 2025, and we're guiding that, that can expand now going into 2026. We do continue to believe that there's opportunities for even more margin expansion. But again, I'm not giving formal numbers for '27 at this point. 

When you look at our ATS business, the business has done very well on doing some selective pruning in order to really focus on the highest value engagement. And so we're really happy with the margin expansion that we've seen in ATS already. And we think that there's opportunities to continue to expand and get it above 6%, hopefully in the near to medium term. 

On the CCS side, which is operating in the low 8s right now, what's working to our favor is the fact that we will continue to be seeing growth in networking, which are primarily our HPS products. And our HPS products are accretive to the company and accretive to CCS. And so as we see growth in that area, we will continue to see some margin upside. 

That being said, we do continue to evaluate how we can support our customers on multiple areas such as doing complex rack integration work. And so sometimes that will be margin dilutive.” 

Earnings 

Celestica reported GAAP EPS of $2.31 in Q3, beating the $1.38 estimate by 67.6%. Adjusted EPS was $1.58, beating the $1.49 estimate by just 6% and representing growth of 52% YoY.  

For Q4, Celestica guided adjusted EPS to be in the range of $1.65 to $1.81, which, at the $1.73 midpoint, is only marginally ahead of estimates for $1.71. This also corresponds to a slight acceleration to 55.9% growth. Looking ahead to Q1 and Q2, estimates point to 52.3% growth and 41.5% growth, decelerating in both quarters, likely driven by margin expansion slowing. 

For fiscal 2025, Celestica boosted its adjusted EPS outlook by 7.3% to $5.90, from its previous forecast for $5.50 and pointing to 51% YoY growth. For fiscal 2026, Celestica outlined an initial guide for $8.20 in adjusted EPS, up 39% YoY and well ahead of estimates for $7.22.  

Cash 

On the other hand, cash flows are rather thin and fell to the lowest level in a year.  

Operating cash flow was $126.2 million, or a 4% margin, down from 5.3% in Q2 and 4.9% in the year ago quarter. OCF growth was just 2.4% YoY and was also the lowest cash flow since the year ago quarter.  

Adjusted FCF was $89 million for a 2.8% margin in Q3, up 15.6% YoY but also the lowest level since the year ago quarter. Adjusted FCF margin was down from 4.1% in Q2 and 3% in the year ago quarter.  

For fiscal 2025, Celestica raised its adjusted FCF guidance slightly to $425 million, from $400 million prior, for a 3.5% margin, while capex is guided to $200 million, or 1.6% of revenue. Fiscal 2026 adjusted FCF was guided at $500 million, up 18% YoY and for a 3.1% margin, with the margin decline driven by higher capex, guided to rise 50-100% YoY to $300 to $400 million, or 2.2-2.5% of revenue.  

Cash and equivalents totaled $305 million while debt totaled $728 million in term loans. Including an undrawn revolver, total liquidity is approximately $1.1 billion. Celestica’s gross debt to TTM adjusted EBITDA was 0.8x, improving by 0.1 points sequentially and 0.3 points from last year. 

Inventories were $2.05 billion, up nearly 7% QoQ, while accounts receivable totaled $2.44 billion, also up nearly 7% QoQ. 

Valuation 

Celestica is trading just off peak multiples on the top and bottom line following this recent pullback after Broadcom’s earnings. Celestica’s forward PS is currently at 2.5x, well above the five-year average of 0.75x and 25% above the 2x multiple it commanded at the start of September. Even on the fiscal 2026 guide, shares are at a 1.9x multiple. 

On a forward PE basis, shares are trading at 45.7x fiscal 2025 adjusted EPS and 32.3x fiscal 2026, well above its five-year average forward PE of 15.4x and prior resistance at 25x in late 2024 and early 2025. The company has been seeing a re-rating higher as it captures AI-growth tailwinds, but any hint of softness in growth could easily see Celestica re-rated lower given growth through 2027 is visible and may already be priced in.  

Conclusion 

Celestica is benefiting from strong market demand for 800G switches with its Tomahawk6-based 1.6T switches on deck for availability later in 2026. The company guided for an impressive five-point acceleration in 2026 during Q3’s report, outlining more than 31% growth to $16 billion in revenue. 2027 was implied to accelerate slightly to around 33% YoY to surpass $21 billion in revenue, again on strong switch demand, the ramp of 1.6T programs, and a new custom rack-scale solution with a digital native customer entering the picture, likely OpenAI. 

However, Celestica’s valuation remains quite stretched, with the company sitting well above its five-year average multiples on the top and bottom line as shares are being re-rated higher for its visible topline acceleration and strong 40% growth momentum in CCS – this extended valuation will need to be watched considering growth expectations could be getting priced in already given the high level of visibility into 2027.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• The I/O Fund’s Top 10 New Ideas List for Q4 2025
• Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth
• Nebius: Financing its Data Center Ambitions Will be Challenging
• Cloudflare: Revenue Accelerates to >30%, Key Metrics Strengthen

Memory is typically a cyclical industry that is lower margin and lumpy, yet it is seeing a newfound resurgence from AI that is strong enough to transform commoditized hardware into a secular trend as the AI economy is built out. AI servers use more DRAM and NAND than traditional servers, relying heavily on high-bandwidth memory (HBM) for training and inference.   

We first touched upon the rising importance of the memory market in AI GPUs in the summer of 2023 within our AMD and Lam Research analyses, and provided a closer look in November 2023 in the analysis, 2024 Trend: Memory and PC Rebound. We also dove further into HBM’s growth opportunities in December 2023 with our Micron deep dive, Micron: AI Offers a Multifaceted Secular Growth Tailwind. 

As is stands, the AI-driven demand for memory (especially HBM and high-performance DRAM) is still in the early stages of a multiyear growth cycle. The company’s CEO and Chairman, Sanjay Mehrotra, also mentioned in the September earnings call, “Memory is very much at the heart of this AI revolution. This means a tremendous opportunity for memory and certainly a tremendous opportunity for HBM."  

The HBM market is projected to reach $35 billion this year, doubling YoY, with Micron’s September results confirming that the market was well on track to be over $30 billion as of Q3. Looking ahead, the shift to HBM4 with Nvidia’s Rubin architecture and AMD’s MI400 series will represent another important growth lever come 2026 as HBM content per GPU and per rack surges, paving the way for HBM to potentially triple again by as early as 2028. 

Not only is HBM a focal point due to its rising importance and thus increasing content per GPU, but other memory products are quickly coming to the forefront, notably low-power DDR5 memory (LPDDR5X) and data center solid state drives (SSDs). 

Below, we look at the memory products front and center of this AI-driven cycle, structural drivers behind rising AI memory demand such as AI inference, supply and inventory constraints driving prices rapidly higher, long-term growth outlooks and more. 

Overview: DRAM and NAND’s Role in AI 

Demand for high-capacity memory is driven by generative AI and LLMs, which both require significant amounts of computing power and substantial amounts of DRAM to meet elevated performance requirements. Within DRAM, demand is focused more specifically around high-bandwidth memory (HBM) and double-data rate 5 DRAM (DDR5) for its increasing content in AI accelerators — SK Hynix’s head of DRAM marketing Park Myung-soo has explained in the past that “an AI server requires 500-gigabyte (GB) or larger high bandwidth memory (HBM) chips and at least 2-terabyte (TB) DDR5 chips.” Now, we’re seeing nearly that amount of HBM being put on a single GPU rather than an 8-GPU system.  

High bandwidth memory (HBM) offers higher bandwidth, capacity, performance, and lower power by vertically stacking up to twelve DRAM memory chips to shorten how far data has to travel, while also allowing for smaller form factors. Stacked memory chips are connected through something called “through silicon vias” or TSVs.  

HBM is now mission-critical, especially for inference, as increasing bandwidth and capacity per HBM generation paves the way for significant leaps in throughput with each GPU generation. Currently, the leading AI accelerators from Nvidia and AMD utilize HBM3e, an enhanced HBM3, while the next-gen Rubin and MI400 architectures are set to bring HBM4 mainstream next year – more on this below. 

DDR5 DRAM, or double data rate 5, is aimed to double bandwidth and data transfer speeds at a lower latency and power consumption than its predecessor, DDR4. DDR5 memory chips can be mounted on circuit boards to create memory modules, for use in servers or PCs. DDR5’s increased bandwidth allows for faster processing for memory-intensive applications, such as generative AI and training LLMs. Memory giant SK Hynix saw high-capacity DDR5 (>128GB) revenue more than double QoQ in Q3. 

Demand for a low-power DDR5 variant, LPDDR5X, is rising sharply due to its role in Nvidia’s Grace and Vera CPUs, as LPDDR5x is delivering up to 5X better throughput, a more than 35% increase in memory bandwidth with up to 77% lower power consumption. This combination can improve system power efficiency (performance per watt) by up to 10%, per Micron.  

Emerging with Nvidia’s GB300 racks (and soon Rubin racks) are SOCAMM modules, which combine LPDDR5X with a Compression Attached Memory Module (CAMM). SOCAMM modules can deliver up to 2.5X higher bandwidth with lower power consumption and a smaller footprint versus traditional RDIMMs (registered dual in-line memory modules).  

Not to be forgotten is memory’s second half, NAND, as it also plays a vital yet less visible role in AI, as its high-capacity, reliable storage is increasingly important in meeting growing inference demands. NAND’s importance is primarily concentrated to NAND-flash based data center solid-state drives (SSD). 

SSDs can boast superior performance in speed, latency, energy efficiency and reliability compared to hard disk drives (HDD). The high-speed read capabilities of SSDs help process vast datasets in training large LLMs and multi-modal models, as well as store model checkpoints. For inference, SSDs are typically used to store ‘hot’ data, or data needing to be accessed frequently, making them crucial for inference workloads.  

HBM3e, HBM4 and the Need for Increased Memory Bandwidth 

HBM3e is the primary version of HBM shipping currently, supporting both Nvidia’s H200 and Blackwell architectures, as well as AMD’s Instinct MI350 series and Google’s TPU v7 Ironwood. HBM4, the next generation, is expected to support Nvidia’s upcoming Rubin platform later in 2026, along with AMD’s Instinct MI400 series.  

The reason that AI accelerators are quickly upgrading to the next generation of HBM is because HBM capacity and bandwidth are consistently increasing, which, when combined with increasing capacity per chip, translates directly to massive leaps in throughput, or tokens processed per second. This means that newer chip generations, such as the shift from Nvidia’s Hopper generation to Blackwell, are exponentially more performant on LLM inference workloads.  

For example, HBM2e, used on Nvidia’s H100, delivered a modest 3.6 Gb/s data rate (speed of data transfer), leading to 461 GB/s of bandwidth per HBM cube. With HBM3, data rates improved to 6.4 Gb/s and stack heights moved from 12 to 16, thus boosting bandwidth by more than 75% to 819 GB/s.  

Source: Rambus 

With HBM3e, data rates increased substantially to 9.6 Gb/s, boosting bandwidth to 1.23TB/s, or nearly 3X that of HBM2e. Translating this to Nvidia’s H200 meant that it could deliver 1.4X to 2X faster LLM inference versus the H100 as bandwidth per chip rose from 3TB/s to 4.8TB/s and HBM capacity rose 1.76X from 80GB to 141GB.  

Source: Nvidia 

With HBM4, the main upgrade is a doubling of interface bits from 1,024 to 2,048, or the number of data bits that can be transferred simultaneously between the memory chip and the GPU. This means that even at the JEDEC standard of 8 Gb/s data rate, a modest decline from HBM3e, bandwidth per HBM4 stack rises to more than 2 TB/s, a 2.5X boost from HBM3 and a ~66% increase from HBM3e. However, Micron claims that its HBM4 boosts data rate to 11 Gb/s, delivering 40% higher bandwidth at 2.8 TB/s per stack, along with 20% better power efficiency and 60% better performance versus HBM3e.  

While these increases may not seem significant when looking simply at upgrades per HBM generation, looking at the exponential increases in bandwidth and inference performance per GPU generation shows a better picture. 

Nvidia’s 8-GPU HGX H100 system delivered a mere 24 TB/s of aggregate memory bandwidth, yet the HGX B200 system boosted that ~2.6X to 62 TB/s with the shift to HBM3e and more HBM3e content (more on this below). 

The scale-up architecture of Nvidia’s GB200 and GB300 NVL72 brought a massive boost to aggregate memory, with both rack-scale solutions offering 576 TB/s, or 24X more than the HGX H100 systems. Nvidia says the GB200 can offer throughput of up to 116 tokens/s on GPT-MoE-1.8T model, a 30X improvement on real-time LLM inference versus the HGX H100, with performance gains also aided by improvements in NVLink and CX8 network interface cards. 

Source: Nvidia 

Nvidia’s upcoming Vera Rubin architecture will boost aggregate memory bandwidth by as much as 8X from here over the next two years. 

Nvidia’s upcoming Vera Rubin NVL144 is expected to take aggregate memory bandwidth to 1.4 PB/s, and 1.7 PB/s with the CPX platform. This is the equivalent of 1,400 TB/s to 1,700 TB/s, or a ~2.4X to ~3X increase versus the GB200/GB300 racks. Jensen Huang claims that the NVL144 system bandwidth is “the entire data usage of the Internet in one second.” 

With the NVL576, aggregate memory bandwidth will continue to surge, with the rack boasting 4.6 PB/s, or 4,600 TB/s of bandwidth. This is another roughly 3X boost to the NVL144, and compared to the GB200/GB300, a massive 8X increase in just two years.  

HBM’s Longer-Term Tailwind: Capacity per Chip Surging 

HBM capacity per chip continues to rise with each new generation of GPU, and this is a primary contributing factor behind the surging aggregate memory bandwidth discussed above, paving the way for accelerated throughput gains and inference performance.  For example, we’ve seen a ~3.5x increase in HBM content in short fashion on Nvidia’s GPUs within about three years’ time frame:  

• The H100 featured 80GB of HBM2e content per chip. This chip began shipping in Q4 2022 and ramped in early 2023.  
• The H200 featured 141GB of HBM3e content per chip, 1.76x higher than its predecessor.  
• The B200 features 180GB of HBM3e content, more than double the H100 and a 28% increase versus the H200. In an 8-GPU server configuration, the B200 boasted 1.44TB of HBM content.   
• The B300 boasts 288GB of HBM3e content, a 60% increase versus the B200 and over 3.5x more than the H100. In an 8-server configuration, the B300 has 2.3TB of HBM content. This chip is beginning to ship now in Q3-Q4 2025.  
• The upcoming Rubin chip will remain at 288GB, but transition to HBM4 for more bandwidth. 

Putting in context Nvidia’s rack-scale solutions, the GB200 and GB300 NVL72, shows just how rapidly HBM content is increasing. The GB200 supports up to 13.4TB of HBM content, while the GB300 supports up to 21.7TB of HBM, nearly 34X higher than the 640GB of HBM content in the 8-GPU DGX H100 servers.  

AMD is also showing surging memory requirements, to the tune of 3.5X across two main generations:   

• The Instinct MI250 featured 128GB of HBM2e memory.  
• The MI350X featured 288GB of HBM3e memory, a 125% increase versus the MI250 and on par with Nvidia’s Blackwell Ultra.   
• The MI400 series is expected to feature 432GB of HBM4 memory, a 50% increase versus the MI350X and the Blackwell Ultra. In the Helios rack configuration slated for 2026, the MI400 will boast 31.1TB of HBM content, 1.5x more than the GB300 NVL72.   

Packing more HBM per chip is also not exclusive to GPUs, with Google’s TPUs notably seeing a 6X jump in HBM capacity over one generation (one year) and a 12X increase in two generations: 

• Google’s TPU v5e, released to general availability in 2023, featured 16GB of HBM capacity. 
• TPU v6e (Trillium), released in 2024, doubled HBM capacity per chip to 32GB. 
• TPU v7 (Ironwood), released this year, boosted HBM capacity by 6X over Trillium to 192GB per chip, or 12X growth from v5e. 

This surge is expected to continue through 2027 as HBM4 and then HBM4e come online – it has been estimated that in a 20-high configuration could pack 80GB of memory per HBM chip, up from 36GB per 12-high HBM3e cube today. Assuming similar usage of eight cubes, this could take memory per GPU up from 288GB in Nvidia’s B300 to 640GB in future chips.   

HBM Market Doubling in 2025, Expected to Triple Again by as Early as 2028 

The HBM market is forecast to double this year to approximately $35 billion, up from less than $18 billion in 2024, with growth driven by increasing content per GPU such as that with Blackwell and Blackwell Ultra as well as from capacity constraints. As of Q3, HBM is now likely above a $30 billion annualized run rate, supported by comments from Micron last quarter that its HBM revenue grew to almost $2 billion; with HBM share expected to nearly match its DRAM share (of 22% in Q2 to 25.7% in calendar Q3), this would imply the HBM market is likely in the mid-$8 billion range, or around $32-33 billion annualized.  HBM’s share of DRAM revenue is also rising sharply, expected to rise ten points this year, from 18% in 2024 to 28% in 2025, with more growth ahead through 2030.  

Through 2026 and 2027, the outlook for HBM remains fairly positive, with SK , SK Hynix, Samsung and Micron already selling out of HBM3e and HBM4 capacity through the end of 2026. This underscores the robust demand environment stemming from AI accelerators, with Micron seeing HBM bit shipments outpacing DRAM bit growth, but also may limit revenue upside as prices have been contracted over the next four quarters.  

On pricing, HBM4 is expected to carry a significant premium to HBM3e, currently used for Nvidia’s Grace Blackwell chips. Analysts from UBS had estimated that HBM4’s price premium could be as much as 30%, though reports of Samsung’s discussions over HBM4 supply with Nvidia dwarfed that – Samsung was said to be targeting price parity with SK Hynix on HBM4 around $500, up ~50% from the mid-$300s for HBM3e. These price increases will support strong growth as HBM4 volumes ramp. 

Looking forward, industry analysts project the HBM market to reach $98 billion to $100 billion by 2030, representing a 31.5% CAGR from 2024’s $18 billion, outpacing DRAM’s growth by 3X, which is expected to rise at an 11.7% CAGR to $194 billion. As a result, HBM’s share of DRAM revenue is expected to surpass 50%.  

However, in its Q1 report, Micron said it now expects the HBM TAM to reach $100 billion as early as 2028, two years sooner than its prior forecast. This would represent a ~42% CAGR from $35 billion, or more than 10 points faster than the base case forecasts.  

HBM’s Challenges: No Tail-End to Shipments, Supplier Shifts  

The reason that HBM can be such a challenging market is two-fold – supplier qualifications can (and do) change rather quickly between generations for SK Hynix, Samsung and Micron, and the winner is oftentimes determined by time-to-market, or whichever company can hit mass production first.  

Micron executives explained that HBM is unlike standard memory products, and that they do not expect a long tail in these products, meaning that once the next generation comes online and ramps (HBM3 to HBM3e, HBM3e to HBM4, etc), demand for the old generation dissipates quickly. This in turn means that whichever suppliers can either qualify first and reach mass production first have an advantage when it comes to revenue and even margins. For example, SK Hynix was the main supplier of HBM for Nvidia’s H100, yet Micron was the main supplier for the H200.  

For HBM4, SK Hynix said in early September that it had finished development of HBM4 and was ready for mass production, while Micron also announced that month that it had begun shipping HBM4 samples to customers. On the other hand, Samsung just finished development of HBM4 in early December and began shipping samples to Nvidia. However, SK Hynix is reportedly delaying the start of mass production from Q2 2026 to Q3 2026 to better align with Rubin’s ramp.  

Rising Demand for LPDDR5X and the DDR5 Profitability Dilemma 

Demand for  LPDDR5X is rising sharply due to its role in Nvidia’s Grace and Vera CPUs, as LPDDR5x is delivering up to 5X better throughput, a more than 35% increase in memory bandwidth with up to 77% lower power consumption versus typical DDR5.  

Low power is critical with Nvidia’s GB racks as well as Rubin, as power consumption has been surging per rack, and already pushing the upper boundaries of what current data center infrastructure can handle. Current builds, such as Vantage’s upcoming 1.4 GW campus in Texas for Oracle, are only designed for ultra-high density racks up to 250kW, meaning these new facilities could quickly be phased out and require new infrastructure to accommodate increasingly power hungry racks. 

Nvidia’s GB300 and Rubin platforms will use a purpose-built SOCAMM module optimized for AI servers, which combines LPDDR5X with a Compression Attached Memory Module (CAMM), aimed at maximizing performance and reducing power consumption. This is currently provided by Micron, which reported 50% QoQ in LPDDR revenue to a new record last quarter.  

When comparing to smartphones, the usual destination for LPDDR memory, the content demands for AI servers are profoundly large:  

“Indeed, each Grace CPU in today's platform is equipped with 480 GB of LPDDR5X memory (a premium smartphone uses 16 GB of LPDDR5X), but this is going to at least double with Vera CPUs, possibly straining LPDDR5X supply.” 

So not only do you have Nvidia’s Blackwell and Blackwell Ultra lines ramping, with those consuming 30X LPDDR5X memory as a typical smartphone, but that gap is poised to widen tremendously later next year as Rubin ramps, with the Vera CPU expected to contain 1.5TB of LPDDR5X, more than 3X the Grace CPU and as much as almost 94 smartphones.  

Keep in mind that the 480GB for the Grace CPU and the 1.5TB for Vera CPU are per chip, per chip, meaning that the GB200 NVL72 rack featuring 36 CPUs will consume 17.28TB per rack. per rack. For the Rubin NVL144, with the same 36 CPU count, LPDDR5X content would surge to 54TB per rack, and in the NVL576, with 144 CPUs, content quadruples to 218TB per rack218TB per rack. That is the equivalent of 13,625 premium smartphones.  

Nvidia’s demand needs are expected to place substantial upward pressure on prices, as Counterpoint Research believes it now is an LPDDR “customer on the scale of a major smartphone maker — a seismic shift for the supply chain which can’t easily absorb this scale of demand.”  

Global DRAM Market Surges 31% QoQ in Q3, Q4 Pricing to Remain Strong 

You would be hard pressed to find another segment of the AI data center industry posting growth to this degree on a sequential basis. Data from TrendForce estimates that the global DRAM market recorded growth of 30.9% QoQ in calendar Q3 to $41.4 billion. In dollar terms, this represented QoQ growth of ~$9.7 billion, or nearly as large of a QoQ jump as Nvidia reported in its most recent quarter.  This growth was driven by “significant increases in conventional DRAM contract prices, higher bit shipments, and growing HBM volumes.”  

For a supplier breakdown, SK Hynix’s revenue grew 12.4% QoQ to $13.75 billion, fueled by seasonal price increases and significant bit shipment growth. Samsung also reported similar significant growth in bit shipments, with revenue up 30.4% QoQ to $13.50 billion. Micron followed with a substantial 53.2% QoQ increase to $10.65 billion, per TrendForce (note that this is for calendar Q3 which does not align with Micron’s fiscal year calendar).  

As of November, TrendForce estimates that DRAM contract prices will accelerate into Q4, predicting conventional DRAM contract prices will surge by another 45% to 50% QoQ, while total contract prices (which includes HBM) will increase by 50% to 55% QoQ – this is a substantial uplift from projections for 18-23% QoQ growth in Q4 at the end of October.  

Contributing to strong pricing is DDR5 DRAM, where prices rapidly skyrocketed – from late September to early November, prices have as much as quadrupled, with impacts felt most on consumer products. Samsung also reportedly just boosted DDR5 prices by 100%, citing no stock left.  

However, revenue growth in Q4 will likely be lower than pricing as bit shipments are projected to decline sequentially due to rapid inventory depletion. DRAM supplier inventory levels are projected to range between two to four weeks, a major crunch from 5.5 weeks on average last quarter and more than 15 weeks at the start of the year.  

Turning to 2025 as a whole, HBM is expected to be a primary growth contributor for the DRAM industry. Current projections have DRAM revenue rising ~35% YoY, or $32 billion, to $127 billion in 2025, meaning HBM is contributing more than half of that dollar growth at ~$17 billion. This is also marking a rapid recovery from 2023’s trough of $52 billion, with the $127 billion projection representing two-year growth of 148%.  

For 2026, it is these tailwinds above, along with tight supply, that can continue to drive strong growth in the DRAM market moving through the year, especially as HBM4 begins to ramp initially with Nvidia’s Rubin platform along with AMD’s MI400 platform.  

Inference is Creating a Secular Tailwind for Data Center NVMe SSDs  

Data center solid state drives (SSDs), such as those based on NAND flash memory, are an often overlooked but equally important memory component when it comes to AI training and inference. This is because data center SSDs offer higher read-write speeds critical for accessing and transferring data rapidly, along with higher performance and energy efficiency, making them vital for larger-scale AI training and inference workloads.  

NVMe (Non-Volatile Memory Express) is a protocol designed specifically for NAND-flash based SSDs that optimizes performance by reducing latency and increasing data transfer speeds by utilizing the PCIe bus. This helps provide the high throughput and fast data transfer speeds necessary for AI workloads – NVMe SSDs can increase performance by more than 2X versus SATA SSDs. 

There are five main types of NAND flash used in SSDs, delineated by the number of bits of data that can be stored per cell: 

• SLC (single-level cell): Stores one bit of data per cell, meaning data can be retrieved faster. SLC offers the best performance and highest endurance, though it is typically the most expensive.  
• MLC (multi-level cell): Stores two bits of data per cell, allowing for a higher data density or higher capacity, though this comes at the expense of performance and endurance. MLC is typically found in consumer NAND products. 
• TLC (triple-level cell): Stores three bits per cell, increasing density and capacity and reducing cost, but also increasing chance for error. 
• QLC (quad-level cell): Stores four bits per cell, providing significant storage capacity (4X that of SLC) and lower costs, making QLC suitable for large-capacity solid-state drives. Meta has made the case for QLC SSDs in data center applications due to the higher density and improved power efficiency versus TLCs at a price that allows for significant scaling, though it says price is not yet competitive enough for broader deployments. 
• PLC (penta-level cell): The next evolution of NAND flash that stores five bits per cell, aiming for significant high density storage but also facing high error rates.    

AI training and inference are two main long-term drivers for SSD demand, which is projected to rise ~6X from 2024 to 2030, from 181 exabytes (EB, or equal to 181,000,000 TB) to 1,078 EB, under McKinsey’s base case scenario. Training demand projected to rise at a 62% CAGR to from 7 EB in 2024 to 127 EB by 2030. On the flipside, demand from AI inference is expected to grow at a 105% CAGR from 6 EB to 447 EB by 2030, giving inference a 41% share of demand versus less than 12% for training. 

This is not only driven by development of more LLMs, but also the increasing size and complexity of frontier models, where training data sets and context windows for inference are getting increasingly large.  

For example, EpochAI estimates that training data set sizes are rising 3.7X per year on average, or nearly doubling every six months. There are some models that are scaling much quicker. For example, Meta’s Llama2-70B from 2023 was trained on 2 trillion tokens, while Llama3-70B, from 2024, was trained on 15 trillion tokens, a 7.5X increase.  Multi-modal models, those integrating audio, video, image or more, are also likely to require significantly more SSD storage, with McKinsey estimating in the hundreds of TBs depending on the mix of data needing to be stored.  

Source: EpochAI 

Looking at tensor parallelism from a memory perspective also shows why the ability to distribute workloads across tens to thousands of GPUs is such a game-changer for AI training and inference. After accounting for memory required to store model parameters and for the activation buffer, a single AMD MI300X GPU can handle a max request of ~6,500 tokens on Llama-70B, per TensorWave. However, when you distribute model parameters across 8 GPUs along with the same buffer, that 8-GPU server could now handle a max request of 523,000 tokens, an ~80X increase, with gains that only compound as server size and memory increase.  

The increasing size and complexity of models also ties directly to a major pain point when it comes to inference: “As models grow in complexity and require longer contexts, their memory footprint expands beyond what a single GPU can handle. This results in inefficiencies where GPUs are memory-starved, causing significant bottlenecks in AI token generation.” 

To put this GPU memory bottleneck in real-life application for inference, AI inference acceleration startup WEKA states that when it tested Llama-3.1 70B with no optimizations, a 100K token prompt took 24 seconds to pre-fill into the model before any output could be generated, but “extending GPU memory to ultra-fast storage [NVMe SSDs] can dramatically improve token processing efficiency.” When configuring an Nvidia DGX H100 server with an 8-node exabyte-scale NVMe SSD pod, WEKA says its “tests demonstrated a staggering 41x reduction in prefill time on LLaMA3.1-70B, dropping from 23.97 seconds to just 0.58 seconds,” significantly improving model efficiency with zero optimizations – simply from adding SSDs to extend GPU memory. This pre-fill time would feel near instantaneous for an end user versus a 24 second delay.  

Other long-term growth vectors include increased adoption of retrieval augmented generation (RAG), which “assembles companies’ own data into vectorized databases, which models then refer to, improving the accuracy and specificity of outputs.” RAG would then require two forms of storage – active storage of useful data, and vector database storage to organize that active data to be accessible by LLMs.  

Source: McKinsey 

A more rapid uptake of RAG or faster multi-modal model adoption could push data center SSD demand up to a 42% CAGR through 2030, reaching 1,490 EB or ~8X 2024’s demand, while slower uptake could see demand rise at a 25% CAGR to 702 EB, or ~4X 2024’s demand.  

Data Center SSD Revenue Up 28% QoQ in Q3 

Similar to DRAM, data center SSD shipments and prices were strong in Q3, driven by hyperscaler demand for AI infrastructure and general-purpose servers. Revenue from the top five companies – Samsung, SK, Micron, Kioxia and SanDisk – rose ~28% QoQ to a new record at $6.54 billion, per TrendForce. Notably, this was broad-based strength, with growth at the five firms all ranging between 26-30% QoQ.   

For Q4, there are a few dynamics in play that are likely to keep prices and thus revenue growth strong. For example, supplier inventories are expected to have fallen sharply, from 10-15 weeks in early Q3 to just 7-10 weeks at the start of Q4, which was said to be ‘below healthy levels’, with enterprise SSD supply growth substantially lagging demand. SanDisk says that its storage-focused SSD is “growing in demand with 2 hyperscaler qualifications underway and a third hyperscaler along with a major storage OEM planned for calendar year '26.” 

In November, TLC and QLC SSDs reportedly experienced strong price increases, with 1 TB TLC SSDs seeing sharp increases and the “most significant shortage due to persistent enterprise SSD demand.” 512 GB TLCs were estimated to see the most significant price hikes at ~65% MoM, while the QLC supply chain tightened and forced prices higher.  

Additionally, TrendForce points out that these inventory and demand dynamics mean “supply shortages in 2026 are becoming increasingly apparent,” providing an additional lever for SSD prices to rise through next year and support more revenue growth as long as inventories and bit shipments do not hinder that.  

Can the Memory Boom Last Through 2028? 

There have been rising discussions regarding the strength of this current memory boom, and whether it can stretch through 2027 or even 2028, as reports from Korea and analysts from Morgan Stanley now estimate. The industry currently has the necessary ingredients for a sustained upcycle: strong demand, supply shortages combined with lean inventories, and strong pricing trends. A multi-year supercycle would likely require persistent supply shortages driving strong pricing power, stemming from elevated demand. such as strong HBM and LPDDR5X content growth with next-gen GPU racks, and strong inference-led tailwinds for SSD growth.  

There are signs emerging that support such a view. Micron said in November that it is seeing “much more supply-demand tightness than we expected” in September and expects this tightness “to continue beyond 2026.” However, perhaps the most important comment from Micron came from Q1’s call this past week, with management saying that “in the medium term, we are only able to meet about 50% to 2/3 of our demand from several key customers.” SK Hynix also believes it will be “difficult to resolve the supply shortage by the first half of 2027.” More specifically on NAND, SanDisk says that demand “continued to outpace our supply, a dynamic we expect to persist through the end of calendar year '26 and beyond.”  

Samsung and SK Hynix have both been rather straightforward about wanting to avoid oversupply, as this could cut the current cycle short and eat into profitability quickly. Samsung executives have said that they will “minimize the risk of oversupply through a capital expenditure strategy that balances customer demand and pricing," and instead of rapidly expanding production, they will focus on profitability.” SK Hynix is on a similar page, though reports have suggested it could boost 1c DRAM production by ~8X by 2026, from 20K units per month to 160K, in order to meet rising SOCAMM and GDDR7 demand.  

The profitability point ties into capacity allocations and exhibits why supply remains tight. For example, HBM3e and DDR5 share production capacity, and through the first part of 2025, HBM3e “commanded a price premium more than four times that of DDR5.” However, with the recent surge in DDR5, profitability is now on track to surpass HBM3e by Q1, meaning suppliers may shift HBM3e capacity to DDR5 to boost profits. Samsung is already planning this shift from HBM3e to DDR5, with the expectation that it will shift ~80K wafers per month, while Micron is shelving its consumer DRAM and SSD unit, Crucial, to focus on HBM, DDR5 and enterprise SSDs.  

Competitive risks aside, evidence of the size of this boom will be visible within revenue growth trajectories and margins. But perhaps the most important question for this cycle is, can the combination of tight supply, low inventories, sharply rising prices and strong (and rising) demand drive margins and earnings power to surpass 2018 levels in a sustainable way? 

Currently, Micron’s revenue estimates and revisions give two primary takeaways into the duration and size of the cycle – analysts are more bullish about the boom lasting into 2028, though they are essentially completely divided on the overall strength of it, with revisions showing a massive range between low and high end forecasts. 

Above shows revenue estimates heading into Micron’s fiscal Q1 report on December 17, with fiscal 2027 and fiscal 2028 both seeing estimates revised 41-43% higher since July. FY27 estimates had moved from $48 billion to $68 billion, while FY28 moved from $51 billion to $72 billion. A majority of the upward revisions have come since September, aligning with surging DRAM prices.  

However, Micron gave a blowout Q2 guide, forecasting revenue of $18.7 billion at midpoint, more than 31% above consensus for $14.23 billion and representing growth of 37.1% QoQ and 132.2% YoY. This has pushed estimates even higher – FY26 and FY27 already see revisions ~$16 billion higher to $74.1 billion (+98% YoY), and $84.3 billion (+15%), while FY28 rose $11 billion and points to flat growth.

Analysts remain essentially completed divided on the potential strength of the cycle, with the gap between the low and high end of revenue estimates doubling from fiscal 2026 to fiscal 2027 and 2028.  

For example, estimates for fiscal 2026 range from $53 billion on the low end to $82 billion on the high end, or a $29 billion range. For fiscal 2027, the low end falls to $46 billion, potentially on pricing peaking much sooner than expected, while the high end rises to $106 billion, a $59 billion range. Fiscal 2028 also sees a $61 billion range between the high and low end of $53 billion to $114.5 billion. 

Source: Seeking Alpha 

More impressively, Micron is showing that gross and operating margins have already surpassed the 2018 peaks, and commentary for expansion through the year suggests some potential upside to already strong earnings estimates.   

For example, Micron’s TTM gross and operating margins, below, have rapidly recovered from 2023’s trough and already pushed past prior cyclical peaks (2010/2015), at 45.6% and 33% respectively as of fiscal Q1 (ending Aug). On a quarterly view, Micron’s FQ1 margins were 56% and 45% respectively, up 11.3 and 12.7 points QoQ. For comparison, SK Hynix reported operating margin at nearly 47% in Q3, up more than five points QoQ. 

Compare this to the 2018 cycle, where DRAM prices tripled over the course of six to eight quarters. Micron’s gross and operating margins peaked at 60% and 50% respectively, and Q2 was guided to far surpass that at 67% and 58.7%, respectively. Again for comparison, SK Hynix’s operating margin peaked at 57% on a quarterly basis in Q3 2018, more than ten points higher than current margins.  

Potential earnings power is where this boom gets interesting, especially for Micron, given the wide range for revenue estimates and the rapid ascent in margins to above >65%/>55%. Considering Micron’s management explained that they “would expect gross margins to expand beyond fiscal Q2” though at a more gradual pace than the last few quarters, it is reasonable to assume upside towards 70-72% and potentially 61-63% on operating margin, assuming similar fall-through. Supporting this would be evidence of strong AI-driven product demand in HBM and LPDDR5X (non-existent factors in the prior 2018 cycle) and strong DRAM pricing. 

Assuming Micron ramps into this margin profile of ~70%/61% by year-end and maintains that through fiscal 2027 (Aug ’27) on tight supply dynamics and demand growth, rough back-of napkin math would place FY27 GAAP EPS at $40.35, or ~7.6% above consensus for $37.50 (although it should be noted that this was $20.77 prior to earnings, or an ~80.5% raise now). 

If FY27 revenue moves to the upper end of the estimated range, or ~$105.8 billion, driven by factors such as strong HBM demand from next-gen platforms and strong LPDDR5X content growth, earnings power could be even stronger. Assuming the same peak margin profile of 70/61/54%, FY27 GAAP EPS could reach $50, or ~33% higher than consensus, though such a scenario could be challenging to execute. 

However, it would be remiss to cover memory without discussing the cyclicality of the industry and risks to the ‘supercycle’ narrative. Some of the main factors that could end this cycle include potential oversupply from capacity additions, or price reverting lower after its current L-shaped trajectory.  

While 16Gb DDR4 and DDR5 prices have seen an “unprecedented spot price rally” to record levels, time and time again DRAM prices have always reverted lower, although cycle timing can differ. The swiftness of the current price rally has already outpaced 2018’s rise, though the duration of the price rally has not nearly been long enough to see when or where it could peak. The next-gen DDR6 is not expected to reach the mass market until 2027, suggesting there is ample runway for DDR5 pricing to remain strong through 2026.   

A more hidden risk to the thesis emerges from consumer electronics. Although Micron has exited its consumer memory business and the focus for the trio of Micron, SK Hynix and Samsung remains squarely on AI, consumer electronics (smartphones and PCs) are still strong drivers of DRAM and NAND demand. For example, some analysts have placed consumer electronics at ~37% of DRAM and ~56% of NAND demand.   

The surging DRAM prices are placing upwards pressure on bill-of-materials content for PCs and smartphones, with Lenovo, Dell, HP, Asus and others already hiking PC prices as a result, estimated at around 15-20%. These dynamics could lead to inventory buildup in consumer electronics markets, or potentially some degree of demand erosion, both major headwinds for pricing strength moving through 2026 and 2027. 

While inventory rebuild and oversupply have previously ended past booms, manufacturers are aiming to preserve strong profitability and avoid flooding the market to keep this cycle intact. However, there can be no assurance that these fears will remain on the back burner come 2027. 

Conclusion 

While this may be a lot to unpack, the primary takeaway here is that the memory market is seeing strong, structural tailwinds from rising HBM and LPDDR5X content in GPUs and SSD use in AI applications. Some of the primary companies located at the heart of this trend include Micron, SK Hynix and Samsung as the primary HBM manufacturers; for enterprise SSDs, the market leaders include Micron, Samsung, SK, Kioxia and SanDisk.  

To help narrow down on this trend, we plan to dive deeper into one of a leading Memory stock to our Discovery tier members the first week of January. 

Subscribe to Discovery and get the Top 10 Emerging Tech Watchlist delivered monthly. Our incoming Top 10 list will be published January 2nd with many new names including a lesser-known memory stock. Current Pro and Advanced Members: To subscribe to Discovery with 30% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY30

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• The I/O Fund’s Top 15 AI Stocks for Q4 2025
• Meta: Growth is Quietly Benefitting from AI, Though Margin Risks Weigh Heavy
• Broadcom FQ4 Earnings: $73B AI Backlog with Visibility; $162B Consolidated Backlog
• Coherent: Indium Phosphide Capacity to Double, Data Center to Reaccelerate to 10% QoQ

Below are our current positions and corresponding theses. In most cases, we have written about the stock many times. What is listed below are the most pertinent analysis for becoming acquainted with the stocks we currently hold. If you want to read more, please use our search bar by also ticking Pro in the filters and search by stock name to pull up more archived articles.

This list will be updated and refreshed when positions are added or removed. Please check back often for updates!

Audited Returns

• 2025 Full Year Audited Returns
• 2024 Full Year Audited Returns 
• 2023 Full Year Audited Returns
• 2022 Full Year Audited Returns
• 2021 Full Year Audited Returns
• 1-Year and YTD Audited Returns for 2021
• 2020 Audited Returns, LTBH Update and Site Update
• The Harsh Truth: Retail Investors Take the Brunt of Market Losses 
• The Importance of Verified Returns and Risk Management for Retail Investors

Quarterly Webinars and Analysis

• The I/O Fund’s Top 15 Stocks for Q2 2026
• The I/O Fund’s Top 15 Stocks for Q1 2026
• The I/O Fund’s Top 15 AI Stocks for Q4 2025
• The I/O Fund’s Top 15 Stocks for Q3 2025
• Q2 2025 Quarterly Kickoff Webinar
• Q1 2025 Webinar with Beth Kindig
• Q4 2024 Earnings Kickoff Webinar Replay
• Q3 2024 Earnings Kickoff Webinar Replay
• Q2 2024 Earnings Kickoff Webinar Replay
• Q1 Earnings Kickoff Webinar
• 2023 Year in Review: I/O Fund Webinar
• Q4 Earnings Kickoff Webinar Replay

Nvidia

• Nvidia Q4: Stellar Report; Stock Remains Range Bound
• Nvidia Fiscal Q1: Perfect Quarter, Imperfect Catalysts

Astera Labs

• Astera Labs: Important QoQ Acceleration, Product Road Map is Loaded
• Astera Labs Q3 Earnings: Blowout Report Meets UALink Uncertainty

Alphabet 

• Alphabet Q4: Cloud Sees 14 Point Acceleration to 48% Growth, FY26 Capex to Nearly Double
• Google’s Q1: TPUs Go Merchant and Cloud Accelerates to 63%
• Alphabet Q4: Cloud Sees 14 Point Acceleration to 48% Growth, FY26 Capex to Nearly Double

Applied Optoelectronics

• Applied Optoelectronics Q1: Management Guides to 141% YoY Growth; Execution Comes Next

Arm

• Arm FQ4: AGI CPU Demand Hits $2B, Revenue Outlook Stays at $1B

AMD

• AMD Q1: Doubled CPU TAM, Helios Incoming for Q4

Bloom Energy

• Bloom Q4: $20B Backlog, Guides for 58% Revenue Growth
• Bloom Energy Q1: Beat/Raise and Customer List is Growing
• Bloom Energy Q3: Doubling Capacity in FY2026 for “4X 2025 Revenue”

Broadcom

• Broadcom Fiscal Q1: $100 Billion+ in AI Chip Revenue in 2027
• Broadcom Offers Strong AI Growth at Scale; Yet Enters Circular Investing

Coherent

• Coherent FQ3: InP Capacity Doubling to Drive CY26 Inflection
• Coherent: Indium Phosphide Capacity to Double, Data Center to Reaccelerate to 10% QoQ
• Coherent Fiscal Q2: Strong Visibility for Back-Half of 2026 and Beyond

GE Vernova

• GE Vernova Q1 Earnings: Backlog and Pricing Point Higher
• GE Vernova Q4 Results: AI Demand Fuels Record Backlog and Strong Visibility
• GE Vernova: All Roads Point to the Nat Gas Behemoth

Lumentum

• Lumentum FQ3: Firing on All Cylinders Despite Stiff Supply Constraints Across EMLs, Pump Lasers
• Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth

Micron

• Micron Fiscal Q2: Record-Breaking Fundamentals
• Micron Stock Up 120% YTD: What the HBM Memory Leader Plans for 2026

Palantir

• Palantir Q1: Strong Headline Numbers; TCV to be Watched
• Palantir Q4: Highest Growth As Public Company; US Commercial To Accelerate

SanDisk

• SanDisk Fiscal Q3: Data Center Inflects 233% QoQ while New Business Models (NBMs) Weigh on the Stock
• SanDisk Q2: Blowout On All Metrics

Last updated on 06/18/2026Last updated on 06/18/2026

The trend toward neoclouds is a high risk/high reward opportunity for investors. Nebius is similar to CoreWeave, dubbing itself as AI native cloud infrastructure, which means the infrastructure was created specifically for AI workloads with architectures built on bare metal servers instead of hypervisor layers, for example, creating an important differentiation from the Big 3 which we’ve previously covered here and here. 

Nebius offers an Nvidia-optimized cloud platform for teams that need to adjust compute resources, want high-performance storage and an easy-to-use AI environment yet do not want to manage the infrastructure or AI operations. 

Nebius shares surged in September following the up to $19.4 billion deal with Microsoft. Recently, the company expanded its hyperscaler customer list recently with a $3 billion deal with Meta expected to ramp over the next three months. Backed by these two deals ramping through 2026, Nebius now projects reaching $7 billion to $9 billion in annualized run rate revenue (ARR) by the end of next year, up more than 13X from its current ARR of $551 million from Q3.  

However, Nebius is a high-risk stock given its success depends on how much capital the company can raise, and the current financials do not support an easy path to ramping capacity and reaching these targets. Capex needs have already moved much higher, and meeting management’s more aggressive capacity targets for the end of 2026 will require significant capital to deploy the necessary GPUs – there is a chance for capex needs to remain at 7-8X of revenue in 2026, making financing a major challenge. 

Regardless of how Nebius executes relative to CoreWeave, it remains an AI bubble stock as the company has to hope the stock price goes up to raise cash, which will dilute shareholders (or raise debt). It’s a vicious cycle as during any months/quarters that AI stocks are soft, Nebius stock will carry outsized execution risk. 

For a closer look at Nebius’ deal with Microsoft, its vertical integration and custom servers, read more from our Advanced analysis, Nebius: Mega Microsoft Deal, 5x Growth in Power Fueling AI Cloud Hypergrowth, But High Risk Remains.Nebius: Mega Microsoft Deal, 5x Growth in Power Fueling AI Cloud Hypergrowth, But High Risk Remains. 

Revenue Growth Decelerates in Q3, but Expected to Reaccelerate 

Revenue decelerated to 355% YoY and 39% QoQ in Q3 to $146.1 million in Q3, down from 605% YoY and 90% QoQ in Q2. Core AI Infrastructure was the primary driver as revenue grew 400% YoY and 40% QoQ to ~$131.5 million, or ~90% of total revenue in Q3. 

On that note, Nebius tightened its 2025 revenue outlook, from $450-630 million down to $500-550 million, citing timing of capacity as the primary reason. This is also below current estimates for 372% growth to $555 million. 

On the other hand, revenue growth is expected to reaccelerate to 549% to $246.1 million in Q4 and further to 749% to $469 million by Q1, with more capacity coming online next quarter to support the Microsoft deal ramping through 2026 and its new $3 billion, five-year deal with Meta, with capacity rolling out over the next three months.  

For fiscal 2026, revenue is expected to reaccelerate to 521% to $3.45 billion, with estimates having doubled since Nebius signed the Microsoft deal in September. Nebius said this quarter that it plans to provide full-year guidance for 2026 in the upcoming quarter.  

However, reaching these targets will likely require strong/perfect execution as the company must build and deliver substantial capacity for both the Microsoft and Meta deal, as well as additional capacity to meet external demand. Doing so will likely require capex at >7X of revenue next year, an incredibly challenging position to be in as peers are spending far less yet still struggling to find funding. 

Ambitious $7-9B ARR Target by End of 2026, up >7X YoY 

Nebius provided an ambitious new medium-term annualized run rate revenue (ARR) target in Q3, forecasting reaching $7 billion to $9 billion in ARR by the end of 2026. For comparison, this would represent >7X YoY growth from 2025’s target of $1 billion at midpoint if it materializes.  

Nebius’ two hyperscaler deals with Microsoft and Meta would account for more than half of this at ~$4.1 billion in ARR at full scale combined (approx. $3.5B for Microsoft and $0.6B for Meta). However, this would require Nebius to generate ~$3.9 billion in ARR at midpoint from other customers and other capacity, a challenging task given how costly its capacity expansion plans will be.  

Additionally, the QoQ decline in incremental ARR, from $181 million in Q2 to ~$121 million in Q3, highlights a major challenge – this growth boils down to timing, and when Nebius can bring capacity online. This could be power bottlenecks, GPU supply bottlenecks, delays in building out physical infrastructure, funding bottlenecks, etc. 

Thus, if the company cannot raise enough capital to afford its ambitious plan of reaching 2.5GW of contracted power and 0.8-1.0GW of connected power by the end of next year, these revenue targets may be unattainable.  

Capex Forecast Raised by 2.5X to $5 Billion, 9X of Revenue 

Capex requirements will be the number one focal point for Nebius considering it does not have the same depth of cash as hyperscalers, yet is competing with them on securing power, GPUs, and AI workloads.  

On this note, Nebius is projected to spend ~9X its estimated 2025 revenue on capex after boosting its 2025 capex forecast by 2.5X to $5 billion, versus its prior view for $2 billion. This capex-revenue ratio is far above what peers such as CoreWeave and even Oracle are spending, at 2.5X and 0.75X. It’s also important to note that the latter two are struggling to in the financing department – CoreWeave is having to take on debt at >9% interest rates, while Oracle’s 520% debt-to-equity ratio has pushed its credit default swaps up to the highest level since 2009. 

Financing this capex will be the primary challenge, and it is critical considering the bulk of this capex will go to revenue-generating GPUs. To put this in perspective for 2025 and 2026, here’s what CEO Arkady Volozh explained about capex and Nebius’ connected power targets: 

And if we look at it from the CapEx point of view, roughly speaking, it breaks into 3 spending blocks. So first stage, securing land and power. It's pretty cheap. It's around — again, it depends on the scale, but it's around 1% of total CapEx for securing those blocks and electricity. The second stage, building the data centers, building connected power is something around, I don't know, 18%, 20%. And the remaining 80%, the main part is for deploying the actual GPUs. This is the main part of CapEx. So if we want to build as much as our capital will allow us, what should we do?1% of total CapEx for securing those blocks and electricity. The second stage, building the data centers, building connected power is something around, I don't know, 18%, 20%. And the remaining 80%, the main part is for deploying the actual GPUs. This is the main part of CapEx. So if we want to build as much as our capital will allow us, what should we do? 

First, we should secure as much capacity as we can because the cost [is] immaterial at this scale. Second, we should build as much as our capital allows us. And third, we will fill GPUs in line with contracted or clearly visible demand. We will need this massive 80% spend [that] will come only when we see real demand. That's why we say that in 2026, we will be securing 2.5 gigawatts total contracted capacity. And we are planning to physically build 800 to 1 gigawatt of connected data centers. This will be done by the end of next year.” 

For the updated 2025 capex guide, this would assume ~$50 million towards land, and ~$1 billion towards the physical data center shell and necessary equipment to connect power, and the remaining $4 billion for GPUs. This aligns with management’s forecast to have 220MW of connected power (not yet active) and 100MW of active power by year-end. This would be around ~$10 million per MW of active power based on the comments above, slightly below averages around $12 to 14 million.  

Looking ahead to 2026, Nebius is planning to have 2.5GW of contracted power and 0.8-1.0GW of connected power (up 2.5X from a prior view for 1GW contracted). Per management, this includes scaling its existing data centers in the UK, US, and Israel, new data centers under development in the US and Europe, and several large sites with up to hundreds of MW under review, with the chance that some go online by the end of 2026. 

Building out this capacity pipeline to reach the connected power targets with powered shells would likely require approximately $3 to $4 billion in capex, while GPUs would likely require $20 billion to $24 billion, potentially higher, depending on mix and how much of that connected power Nebius aims to have active. This is anchored by GPU costs for next-gen hardware now running at $25 billion to $30 billion per GW.  

Financing Potential >4X Capex Growth, Still 7-8X of Revenue 

There is a likelihood that Nebius’ capex needs for 2026 rise at a multiple of >4X YoY to meet these aggressive capacity expansion targets. This would also be ~7-8X current estimated revenue of $3.45 billion, meaning the company will be unable to break free of this extremely elevated capex cycle next year.  

Compare this to Nebius’ balance sheet, which currently show $4.79 billion in cash on hand and $4.1 billion in debt following the company’s recent combined debt and equity offering raising $4.2 billion. This is not likely not even 20% of what the company could need to reach its capacity targets, and thus its revenue targets.  

CFO Dado Alonso covered the financing aspect, though it is unlikely that these financing avenues could raise the necessary amount needed for these capacity targets without significantly stressing the company’s balance sheet:  

“In order to support our aggressive growth plans in 2026 and to maintain this pace of growth in 2027, we will be utilizing at least 3 sources: corporate debt, asset-backed financing and equity. We are in the process of raising asset-backed debt, which we'll be able to secure with attractive terms supported by creditworthiness of our largest customers. Tomorrow, November 12, we will be putting in place an at-the-market equity program for up to 25 million Class A shares and plan to file a prospectus supplement. We will evaluate the program regularly based on our capital needs. The program enables us to access equity funding on an efficient ongoing basis. However, we will remain dilution sensitive as we prepare to finance future growth opportunities.” 

Alonso’s comments suggest that the current cash on the balance sheet likely will go towards other data center opportunities outside of its hyperscaler deals, with the asset-backed debt more likely to fund the Microsoft and/or Meta buildouts in similar fashion to some Bitcoin miners raising substantial cash via asset-secured debt.  

The at-the-market program could provide around $2.25 to $2.5 billion in capital around current share prices, if exercised in full between $90 to $100/share, while leading to approximately 10% dilution. Considering management’s goal of remaining dilution sensitive, the ATM program may be utilized at higher share prices to raise more capital, or as a second source behind debt. Again, raising only a few billion via 10% dilution to shareholders is still far from enough from meeting estimated capex needs next year.  

While Nebius remains in a better position than CoreWeave at the moment in terms of debt-to-equity, at 0.94x versus 4.85x, there is a very high likelihood that this ratio will move rapidly in CoreWeave’s direction through 2026 given the capital intensity of building this capacity at an accelerated pace. There is also the potentiality for Nebius to be unable to raise $20B+ in capex given that it is approximately equivalent to the company’s current valuation and still 7-8X of revenue. 

Why Nebius Must Spend Aggressively – Capacity is Sold Out 

The reason that this capex growth is necessary, at least from management’s point of view, is because demand continues to far outstrip the capacity that Nebius can offer. Put another way – it is a spend or get left behind market.  

Management emphasized numerous times in Q3’s call that capacity is the main bottleneck to revenue growth, and their current main focus is adding capacity to remove this bottleneck. CEO Arkady Volozh explained that demand was very strong in Q3 with Nebius selling out of all available capacity, and each time capacity was brought online, it was sold. Nebius is currently “selling the remnants of Q4, but [also] now preselling new capacity being delivered in future quarters” in 2026, helping lock in future revenue growth. This would also include the company’s trance of GB300s coming online in Finland in Q4.  

One of the more important comments this quarter related to the demand pipeline. Management stated that pipeline generation, or customers wanting to buy capacity, expanded 70% QoQ to $4 billion in Q3, yet they “were only able to convert a portion of that given to the constraints of our capacity.” For context, core AI infrastructure revenue was approximately $131 million in Q3, so the pipeline would be ~30X its current run rate. It makes sense why Nebius is aiming to aggressively grow capacity given it witnessed well over $1.5 billion in sequential growth in the pipeline, as meeting a larger portion of this pipeline via more capacity would quickly translate to revenue growth.   

To that point, management said that lead times from power connection and start of GPU deployments to revenue generation range “anywhere from 6 to 12 weeks” or potentially faster in existing facilities, again underscoring why they are willing to pursue this rapid capacity growth through 2026 as they can quickly shift to revenue generation from power connection. 

However, this has highlighted a major downside to a capacity-constrained model. Nebius stated that it is “learning to say no to customers as we routinely sell out and have to actually let them down lightly and try to convince them to purchase in the future,” but there is an equal chance that these customers will simply go to CoreWeave or a hyperscaler who has capacity, get locked in to that ecosystem and not return to Nebius.  

It also sheds light on the puts and takes of Nebius’ deployment strategy at this small scale. By prioritizing hyperscaler deals with Microsoft and now Meta (though this was ultimately constrained by capacity), Nebius is locking in strong future revenue streams over the next few years, but turning away these smaller startup customers by also locking up a larger portion of its near term capacity to the hyperscalers.   

Building Core AI Cloud Business  

As we discussed in our prior analysis on Nebius for Advanced members, Nebius: Mega Microsoft Deal, 5x Growth in Power Fueling AI Cloud Hypergrowth, But High Risk Remains, the company believes its vertical integration and proprietary cloud serve as a key advantage and the ‘real future opportunity’: 

“In the longer term, Nebius believes that its vertical integration with a full stack of AI services will help broaden its customer base, increase platform stickiness and capture higher margin revenue and services. Nebius offers a proprietary cloud platform with managed MLops services, low downtime and high cost efficiency, combined with its inferencing platform AI Studio. With AI Studio, Nebius says it can offer up to 3x token savings with low latency, and up to 4.5x faster time to token versus other competitors in Europe.”offer up to 3x token savings with low latency, and up to 4.5x faster time to token versus other competitors in Europe.” 

As such, the company is prioritizing building out its core AI cloud platform and continuously adding new features, with the predominant goal currently being geared towards enterprise adoption. Two main features from Q3 include Nebius’ new enterprise-ready cloud platform, dubbed Aether, as well as Nebius Token Factory.  

With Aether, Nebius brings enterprise-grade security and compliance features along with a comprehensive observability suite, developer tools and more, in an effort to make its platform more attractive and accessible for large enterprise customers. Nebius also is focusing on improving reliability of its network with active health checks to reduce maintenance tasks, and boosting performance and storage speeds.  

Token Factory builds on Nebius’ AI Studio and embeds enterprise-grade security in a production-scale inference platform, letting customers run open-source AI models from OpenAI, Alibaba, Meta, DeepSeek and others with 99.9% uptime, per the company. Management says Token Factory will help customers “transform open source models into optimized production-ready systems with guaranteed performance and transparent cost per token” and the “best total cost of ownership.” While it is too early to see how Token Factory contributes to monetization, Nebius’ dedication to improve its platform and expand its suite of offerings can serve as a strong differentiation and potentially aid in customer acquisition over rival platforms. 

Financials 

Revenue to Reaccelerate 

Following the deceleration in Q3, Nebius tightened its 2025 revenue outlook, from $450-630 million down to $500-550 million, citing timing of capacity as the primary reason. This is also below current estimates for 372% growth to $555 million. 

For fiscal 2026, however, revenue is expected to reaccelerate to 521% to $3.45 billion, with estimates having doubled since Nebius signed the Microsoft deal in September. Nebius said this quarter that it plans to provide full-year guidance for 2026 in the upcoming quarter. 

Margins Improving, but Widely Negative 

Gross margin was 70.6%, down slightly from 71.3% in the prior quarter and up from 69.2% in the year ago quarter. 

Operating margin began to show signs of improvement, coming in at (89.1%), compared to (105.8%) in the prior quarter and (251.1%) in the prior year. At this rate, Nebius could break even in five quarters assuming it can maintain such improvements consistently, though this may be challenging as capacity ramps up rapidly. 

Net margin was (81.9%), not comparable to the prior quarter’s 556% on Toloka’s deconsolidation but up from (293.5%) in the year ago quarter. Adjusted net margin was (68.7%), improving from (123.7%) in the year ago quarter. However, it is important to note that net losses has widened (from ($39.7 million) to ($100.4 million) for adjusted net loss) though margins are showing improvement from the rapid revenue ramp. 

Adjusted EBITDA improved to just ($5.2 million) or a (4%) margin, up from a (20%) margin last quarter. Nebius noted that its Core AI Infrastructure business continued to generate positive adjusted EBITDA at a nearly 19% margin in Q3, with the metric weighed down by Avride and TripleTen investments.  

Nebius lags CoreWeave by a significant degree for adjusted EBITDA, with CoreWeave posting a 61% adjusted EBITDA margin in Q3, down 4 points YoY. This suggests that there is room for substantial expansion over the upcoming quarters as the business scales to a much larger size. 

Earnings Remain Far From Profitability 

Nebius reported a 23% beat on adjusted EPS in Q3, though the company remains far from profitability and is not expected to reach profitability for quite some time.  

Q3 adjusted EPS was ($0.40), beating estimates for ($0.52) but widening slightly from ($0.38) in the prior quarter. Looking ahead to Q4, adjusted EPS is expected to be ($0.58), before widening to ($0.74) in Q1. 

For fiscal 2025, adjusted EPS is projected to be ($1.73) before widening to ($2.29) in 2026, likely driven by increasing expenditures to quickly ramp capacity. 

Cash Needs Increasing 

The challenge for Nebius is very similar to that of CoreWeave, with the neocloud spending significantly on GPUs and raising substantial debt to fund said spending. As a reminder, Nebius spent nearly $1 billion in capex in Q3, up from $510 million in Q2 and representing more than 6X its revenue. Capex is on pace to be >9X of revenue this year and potentially remain at 7-8X of revenue next year. 

Operating cash flow was ($80.6 million) for a (55.2%) margin, while free cash flow was ($1.04 billion) for a (709.1%) margin. This widened from free cash flow of ($678.3 million) in the prior quarter as capex surged more than 87% QoQ to $955.5 million. 

Cash and equivalents totaled $4.8 billion, while debt was $4.1 billion. As mentioned previously, surging capex this year and the potential for tens of billions next year means Nebius will likely turn to debt markets for significant funding. Debt to equity sat at 0.94X in Q3, though this is likely to worsen significantly moving through 2026 based on estimated capex needs north of $20 billion.  

Valuation 

Nebius trades at 37x forward PS ratio, slightly above its average of 32x, though data is limited considering its recent launch on the public markets post-Yandex breakup. Shares have traded as low as 8.6x forward PS and as high as 61.5x. 

Rapid revenue growth in 2026 is expected to bring forward PS down to 5.9x next year, though this remains a premium to CoreWeave at 3.0x next year’s revenue estimate of $12.07 billion. 

Conclusion 

Nebius is on a trajectory of high growth-high debt for the foreseeable future, with the company spending nearly $1 billion this quarter on capex alone, in preparation for capacity ramps for Meta and Microsoft occurring in quick succession.  

On one hand, revenue growth is expected to accelerate sharply to 750% over the next two quarters and maintain a hypergrowth profile with >380% growth for the next five quarters, while YoY growth is expected to accelerate 148 points to 521% in 2026. However, capex guidance for 2025 was raised to ~9X of revenue, and aggressive capacity expansion targets for 2026 mean capex will likely remain 7-8X of revenue, far above peers and making finding financing a significant challenge. 

Ultimately, even if high-beta stocks catch a bid, there are far cleaner and less capital-intensive ways to gain exposure to the AI buildout. Until Nebius can demonstrate that its growth is not being funded by an expanding cash shortfall, we see limited risk-reward and will remain on the sidelines.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• The I/O Fund’s Top 10 New Ideas List for Q4 2025
• Lumentum: EMLs Driving Results, CW Lasers Ramping with Q2 Guided for 22% QoQ Growth
• Cloudflare: Revenue Accelerates to >30%, Key Metrics Strengthen
• Talen Energy: Acquisitions May Unlock Data Center Deals, Amazon Ramping Soon

While much attention is given to Nvidia and the AI semiconductor ecosystem for visible AI-driven hypergrowth trajectories, there is another quiet AI beneficiary emerging in Meta, with advertising revenue accelerating to the mid-20% range and YoY dollar growth surpassing $10 billion in Q3. Revenue forecasts continue to strengthen over the next few years with revisions of up to $30 billion, or up 10-14% since July, underscoring greater confidence in Meta’s ability to leverage AI to improve monetization.   

More impressively, Meta’s AI ads automation platform has quietly reached a $60 billion run rate in Q3 in three and a half years from its launch. This is 3X Broadcom’s AI revenue and also 3X of OpenAI’s projected ARR of $20 billion for 2025, emphasizing how large of a platform Advantage+ is. Compared to Microsoft, AI contributed 16 points of growth in the April quarter, implying a run rate of around $16 billion assuming growth at a similar degree as the January quarter’s 175% YoY to $13 billion.  

However, the thorn in Meta’s side stems from the compute and capex side, as the company is aggressively building data center capacity to prepare itself for the most optimistic scenarios of reaching superintelligence. Not only is capex guided to surge to over $100 billion in 2026, potentially creating another cash flow crunch reminiscent of 2022’s metaverse-linked spending spree, but expense growth is also expected to outpace revenue growth by a wide degree and weigh heavily on operating margin.  

Advantage+ Reaches $60 Billion ARR 

Meta’s flagship AI ads automation platform Advantage+, powered by Andromeda, is also key to the company’s strong ads performance, as Meta has been straightforward about the end-to-end platform driving strong return on ad spend for advertisers. Advantage+ automates campaign targeting, budget allocation, and creative generation, providing advertisers with an easy-to-use tool that integrates generative AI directly into Meta’s ad ecosystem.  

Meta revealed earlier this year that “for every dollar spent on its AI-enabled Advantage+ products, advertisers generate on average $4.52 in revenue for their businesses,” or an increase of ~22% versus typical campaigns, highlighting how ad performance improves while using the platform. 

In Q3, Meta emphasized that “Advantage+ continues to drive performance gains, [and] advertisers who run lead campaigns using Advantage+ are seeing a 14% lower cost per lead on average than those who are not.” This compares to a 10% lower cost per lead as of April, showing that the platform continues to drive results while lowering costs per lead.  

More importantly, Meta disclosed that Advantage+ has surpassed a $60 billion annual run rate, just three and a half years after its launch. CFO Susan Li sees room to continue growing this run rate for Advantage+ and expanding adoption of the platform by focusing on driving continued performance improvements.  

First Monetization Lever: Improving Ad Performance 

Meta outlined two monetization levers in Q3 – improving ad performance, and an ability to continue delivering engaging content to users. 

This first monetization lever stems from improving ad performance for its advertisers, mostly driven by the company’s three foundational models as well as its end-to-end ads automation platform Advantage+. Meta opts for tracking conversions to gauge ad performance, despite it being a complex metric to track considering advertisers can optimize for different types of conversions. CFO Susan Li stated that value-weighted conversion rates showed “very strong” YoY growth in Q3, outpacing impressions.  

Meta’s three foundational models all serve a different function, with the same end goal of improving ad quality and conversions to drive higher ROI for advertisers: 

• GEM (Generative Ads Recommendation Model) is described by Meta as the ‘super brain’ that can rapidly process, catalog and analyze trillions of data points, to then recognize subtle patterns in user activity to provide the most relevant ads at the right time. Meta says GEM was rolled out more broadly earlier this year after initial testing on Reels saw GEM boost conversions by up to ~5%. GEM delivered a 5% increase in conversions on Instagram and a 3% increase on Facebook in Q2, and in Q3 Meta “doubled the performance benefit we get from adding a given amount of data and compute” to continue scaling training capacity at an attractive ROI.
• Lattice is described as a ‘giant library’ that generalizes learnings across different campaign objectives (clicks, views, etc), surfaces (Reels, Story, Feed, etc) and subjects, in order to predict an ad’s performance. Lattice increases ad efficiency as it runs fewer models, while the knowledge-sharing effect increases ad quality and conversions – Meta said earlier this year that Lattice has increased ad quality by 12% and conversions by 6%. In Q3, Meta rolled out Lattice to app ads, driving a ~3% gain in conversions on that objective.
• Andromeda is described as a ‘personal concierge’, or Meta’s vast ML ad recommendation and prediction system that, at its core, aims to predict exactly which ads a user will find the most interesting. For Andromeda, Meta says, “Imagine having a personal concierge who knows your tastes so well that they don’t just understand that you covet shoes, but that you like to wear red flip flops at the beach.” Meta said that in Q3, it significantly improved Andromeda’s performance by combining retrieval and early-stage ranking models, driving a 14% increase in ad quality on Facebook. Andromeda is also the core engine powering Advantage+ automation tools. 

Moving to 2026, Meta discussed that it is “working on combining these 3 major AI systems into a single unified AI system that will effectively run our family of apps and business using increasing intelligence to improve the trillions of recommendations that it will make for people every day.”  

A single model that combines the strengths of GEM, Andromeda and Lattice could theoretically understand user preferences and activity at a much deeper level, improve ad ranking quality, relevance and conversions across its family of apps, and save on inference. For example, Meta does not use GEM for inference as its size makes it too cost-prohibitive, rather it transfers knowledge to smaller run-time models; a single model incorporating GEM’s knowledge could potentially run inference in a more cost-effective manner.  

Second Monetization Lever: Increasing User Engagement 

On the second of increasing engagement, Meta is executing quite well, with improvements in its recommendation models helping drive time spent on its apps higher. More time spent then allows ad impressions to grow without substantially increasing ad load, underpinning this reacceleration in impressions growth seen in Q3 and more growth moving forward.  

Management pointed out that “overall time spent on Facebook and Instagram grew double digits year-over-year, driven by continued video strength as well as healthy growth in nonvideo time on Facebook.” Video time spent on Instagram was more than 30% higher versus last year, while AI ranking optimizations helped drive 10% more time spent on Threads in Q3. This video growth has pushed Reels to a $50 billion annual run rate in Q3, up 5X from its last update in Q2 2023 when it reached a $10 billion run rate.   

Improving ranking models remains a key focus for Meta moving through 2026, with management expecting new model innovations to help “significantly scale up the amount of data and compute we use to train our recommendation models in 2026, yielding more relevant recommendations.”

AI Aiding Meta’s Advertising Growth Flywheel 

On a positive note, Meta is already seeing tailwinds from AI recommendation models driving higher ROI for advertisers and increasing time spent across its family of apps, fueling stronger advertising revenue growth. 

In Q3, advertising revenue grew 25.6% YoY, accelerating more than nine points since Q1 and marking the fastest growth in six quarters. Ad impressions rose 14% YoY in Q3, accelerating from 11% in Q2 and marking a strong inflection from just 5% growth in Q1. Pricing remained steady, rising just one point to 10% YoY.  

However, the dollar growth in advertising stands out more — Meta has delivered its two largest YoY growth quarters on a dollar basis in Q2 and Q3, at $8.23 billion and $10.2 billion, even outperforming Q4 2024’s holiday-boosted growth of $8.08 billion. This high dollar growth is poised to continue in Q4 2025, with guidance pointing to ~$9.3 billion to $10.7 billion in QoQ dollar growth.

Put another way, Meta is delivering larger YoY dollar growth in advertising revenue on a larger base – Q3 grew $10 billion YoY off a $40 billion base, versus $7.5 billion growth in Q3 2024 on a $33.6 billion base. 

ARPP Continues to Accelerate Heading into Q4 

Perhaps the most important metric for Meta’s ads monetization is ARPP (average revenue per person), with the metric continuing to accelerate in Q3 ahead of the seasonally-stronger holiday quarter. ARPP reached $14.46 in Q3, accelerating to 17.7% YoY from 14.8% in Q2. More impressively, this marked a record high for ARPP, surpassing Q4 2024’s seasonally stronger ARPP of $14.25. 

This sets the stage for ARPP to push well beyond $15, potentially to $16 in the upcoming quarter, highlighting that Meta’s AI-driven ad performance improvements and monetization efforts are bearing fruit.  

Meta’s Upcoming Capex Surge and Possible FCF Crunch 

Some of the most important quotes from Q3’s call circled back to Meta’s view on capex and why it believes aggressive expansion of capacity and thus capex is a necessity. CEO Mark Zuckerberg believes it is the “right strategy to aggressively front-load building capacity so that way we're prepared for the most optimistic cases” on when AI superintelligence arrives, so Meta is prepared to capitalize on this opportunity.  

If building superintelligence takes years longer than expected, Zuckerberg says Meta can “use the extra compute to accelerate our core business which continues to be able to profitably use much more compute than we've been able to throw at it. And we're seeing very high demand for additional compute, both internally and externally.  

These comments underscore why Meta is aggressively raising its capex spending this year and next – Zuckerberg believes that the upside potential of superintelligence is so large that it is worth the risk of overbuilding to not fall behind OpenAI or Google (with compute capacity being the main advantage), with Meta able to use extra compute in the meantime to improve core AI ad capabilities and drive growth.  

However, the tradeoff for this is lower free cash flow and potential operating margin headwinds. Meta expects capex dollar growth to be “notably larger in 2026 than 2025,” while total expenses “will grow at a significantly faster percentage rate in 2026” driven by infrastructure costs, incremental cloud costs and depreciation, followed by employee compensation.  

This implies 2026 capex of at least $103 billion, as current guidance for 2025 at $70-72 billion implies a minimum of ~$32 billion YoY growth. However, considering management’s comments for notably larger dollar growth, there is potential for capex to come in at or above $110 billion, up ~55% YoY, above current estimates for $107.9 billion. Put another way, Meta could spend ~$30 billion more in 2025 and 2026 than it did in 2019 through 2024 combined. 

The capex surge will potentially cause another free cash flow crunch similar to 2022, with current consensus estimates pointing to FCF of $19.71 billion in 2026, down nearly (50%) YoY and (63.5%) from 2024. 

Expense Growth to Meaningfully Outpace Revenue Growth in 2026 

Tying into capex is Meta’s expectation for expense growth to be significantly faster in 2026 versus 2025, which means expense growth will outpace revenue growth by a wide margin, potentially as much as a factor of 2x.  

For perspective, Meta is forecasting total operating expenses of $116-118 billion this year, up 22-24% YoY, marginally outpacing estimated revenue growth of 21.3%. This is crucial heading into next year as a lack of meaningful gross margin expansion means this growth will directly pressure operating margin. 

If expenses grow at ~30% YoY in 2026, this would project total expenses in the range of ~$151-153 billion, significantly outpacing expected revenue growth of 17.9% to $234.7 billion. This would also project out to an operating margin of ~35%, marking a relatively sharp contraction back to the lowest levels since early 2023.  

If expenses grow faster, at say 2X estimated revenue growth or ~35% YoY, this would project expenses in the range of ~$157-159 billion. While only slightly higher than the ~30% growth forecast, this would bring operating margin down to 32.5%-33%.   

JP Morgan’s Doug Anmuth questioned management about this capacity expansion strategy and how this spending ties to earnings and cash flow:  

“I appreciate the strategy to front load capacity for superintelligence. Can you just talk about your thought process and kind of triangulating the Capex dollar growth and the significantly faster expense growth next year with core growth in the business and then the impact on earnings and free cash flow? And do you have targets that we should be thinking about for cash on hand or net cash overall?” 

CFO Susan Li offered a lengthy discussion in response that offered no clarification on earnings or cash flow impacts, and hinted that Meta may not be worried about those two line items in this buildout:  

“But to date, we keep on seeing this pattern where we build some amount of infrastructure to what we think is an aggressive assumption. And then we keep on having more demand to be able to use more compute, especially in the core business in ways that we think would be quite profitable than we end up having compute for. 

So I think that, that suggests that being able to make a significantly larger investment here is very likely to be a profitable thing over some period… Now I mean, it's, of course, possible to overshoot that, right? … And then the kind of the very worst case would be that we effectively have just prebuilt for a couple of years, in which case, of course, there would be some loss and depreciation, but we'd grow into that and use it over time.” 

On the point of Meta’s 2026 budget still being put into place, reports surfaced that Meta is considering making budget cuts of up to 30% in its metaverse division, Reality Labs, which is currently burning about ~$20 billion per year. This could save several billion if put into place, though there is potential for that money to simply be reallocated towards data center capex.  

Financials 

Revenue Growth Accelerates to 26% 

Meta reported revenue of $51.24 billion in Q3, accelerating more than 4.5 points to 26.2% YoY, the highest growth since Q1 2024. This also marked an impressive reacceleration from 16% growth in Q1 2025.  

For Q4, management guided to revenue between $56 to $59 billion, up 18.8% YoY at midpoint, driven by expectations for strong ad revenue growth, partially offset by lower YoY revenue in Reality Labs from lapping the Quest 3S introduction. For 2025, revenue is expected to grow 21.3% to $199.5 billion, before decelerating to 17.7% growth to $234.7 billion in 2026. 

Annual Revenue Revisions Seeing Sharp Increase Since July 

What’s notable on the revenue front is the sharp upward revisions to annual revenue estimates, with 2026 and 2027 moving sharply higher since this summer.  

Back in July, prior to Q2’s earnings, Meta was expected to generate $215.1 billion in revenue, with that now sitting at $234.7 billion. On a YoY basis, growth has been revised from 14.0% to 17.7%, a smaller uplift considering 2025 comps have toughened, having risen from 14.7% to 21.3% over the same period.  

For 2027, Meta was expected to generate $240.6 billion in late July, with that now sitting at $271.0 billion, with YoY growth moving from 11.9% to 15.5% on a higher base.  

Key Metrics 

Meta’s key metrics strengthened broadly in Q3. Ad impressions growth accelerated three points to 14% YoY, its fastest growth since Q1 2024. Pricing has remained relatively stable at 10% YoY in Q3, up one point sequentially, driven by increased advertiser demand fueled by improved ad performance.  

Family of apps daily active people (DAP) also accelerated to 7.6% YoY to 3.54 billion, up from 6.4% growth last quarter and marking its fastest growth since Q4 2023. 

Operating Margins Contracts Sequentially 

Despite the topline reacceleration in Q3, Meta’s operating margin contracted as expenses grew 32% YoY, six points faster than revenue growth. 

• Gross margin was 82% in Q3, up 0.2 points YoY but down 0.1 points QoQ. 
• Operating margin was 40%, down 2.7 points YoY and 3.0 points QoQ. Aside from Q4 typically being seasonally stronger, the expense guide for next year suggests operating margin could return to the mid-30% range.  
• Net margin was 5.3%, negatively impacted by a one-time, non-cash income tax charge of $15.93 billion; excluding this charge, net margin would’ve been 36.4%, down 2.3 points YoY and 2.2 points QoQ. 

Earnings 

Due to the income tax charge, Meta reported $1.05 in GAAP EPS; adjusted for this charge, EPS was $7.25, compared to estimates for $6.67. 

Looking ahead, GAAP EPS growth is expected to remain approximately flat for the next three quarters due to the margin pinch from rising expenses: 

• Q4 GAAP EPS estimated at $8.17, up 1.9% YoY. 
• Q1 ’26 GAAP EPS estimated at $6.32, down (1.7%) YoY. 
• Q2 ’26 GAAP EPS estimated at $7.08, down (0.8%) YoY. 

For FY25, Meta is expected to deliver a (2.2%) decline in GAAP EPS to $23.34, with the decline primarily due to Q3’s income tax charge-related miss. Earnings growth is expected to rebound to 27.3% to $29.72 in 2026.  

Cash and Balance Sheet 

Operating and free cash flow margins expanded sequentially, though Meta is expected to see a steep free cash flow crunch moving through 2026 as a result of surging capex (which also does not reflect the company’s true spending on data center infrastructure).  

• Operating cash flow was $30.0 billion in Q3 for a 58.5% margin, down from a 60.9% margin in the year ago quarter but up from a 53.8% margin in Q2. 
• Free cash flow was $10.63 billion for a 20.7% margin, down sharply from a 38.2% margin in the year ago quarter but up from 18% in Q2. 
• Q3 capex rose 110.5% YoY to $19.34 billion, driven by investments in servers, network infrastructure and data centers. Based on 2025’s capex guide, which was raised to $70-72 billion (up 81% YoY at midpoint), Q4 capex is on track to be ~$21 billion, up 41.5% YoY. 

Creative Funding Solutions Not Appearing in Capex, Saving Cash 

Another important point to cover is Meta’s use of creative financing solutions to build out its large scale data centers, such as its joint venture with Blue Owl to fund its Hyperion data center in Louisiana. The additional costs are not appearing in capex but rather in ‘other investing cash flows’.  

Under the JV deal, Blue Owl will own 80% and Meta will retain a 20% stake, overseeing construction and ultimately renting the data center once operational. However, concerns are rising about the deal structure, as the WSJ points out that the facility is “financed with debt, and neither the data center nor the debt will be on [Meta’s] own balance sheet.” 

Truist analysts questioned CFO Susan Li about the JV and how it will appear in and affect capex: 

“And then Susan, how do you see the on-balance sheet versus off-balance sheet financing of your AI initiatives? You've recently struck a deal with Blue Owl for the Louisiana data center. Is that part of the CapEx guide for '26? And if it's not, how significant will that way of funding be for Meta going forward? And basically, would that slow down your CapEx growth past 2026?” 

CFO Susan Li: “So the JV that we announced with Blue Owl is sort of an example of finding a solution that enabled us to partner with external capital providers to codevelop data centers in a way that gives us long-term optionality in supporting our future capacity needs just given both the magnitude, but also uncertainty of what the capacity outlook in future years looks like. 

In terms of how that is recognized as Capex, our prior Capex reflected a portion of the data center build cost prior to the joint venture being established. Going forward, the construction cost of the data center will not be recorded in Capex as the data center is constructed, we will contribute 20% of the remaining construction costs required, which is in line with our ownership stake, and those will be recorded as other investing cash flows.” 

Valuation 

Meta’s shares are trading slightly above its median forward PS valuation since the start of 2024 at 8.1x, though this has compressed from 9.4x in prior to Q3 earnings in late October after shares sold off.  

On the bottom line, Meta is valued at around a 25x forward PE, slightly above its 5-year median of 22.2x, though the company had traded as low as the single-digits in late 2022 and early 2023 when its metaverse spending spree cut into operating and net margins. Looking out to 2026, Meta trades at ~22x current EPS estimates, slightly above its average of 20.7x since the start of 2024.

However, where the valuation gets stretched is on the free cash flow side – looking ahead to 2026 and the projected $19.7 billion in free cash flow (subject to change with capex forecasts), Meta trades at 85x estimated 2026 FCF. This would represent a significant deterioration of this multiple from the current 38.3x and represent the most expensive Meta has traded on an FCF basis since shortly after its IPO.  

Conclusion 

The most recent earnings report proves that Meta is using AI internally to materially move the needle. Meta’s Advantage+ automation tools continue to drive measurable improvement in advertising efficiency with an updated annual run rate that exceeds $60 billion – a substantial run rate considering this was launched only three years ago.  

Across the board, there was a noticeable reacceleration in impressions, and ARPP was at a fresh record despite it not being the company’s seasonally strongest quarter. Plus, Reels is also up 5X in two years, reporting a $50 billion run rate following improvements in the AI-driven video content recommendation system.  

However, the main risks to Meta’s thesis lay within its ambitious capacity expansion plans, with management laying the framework for easily over $100 billion in capex in 2026 and notably stronger expense growth. This is not only expected to create a large FCF crunch similar to 2022, back to <$20 billion, but also a substantial headwind to operating margins and thus EPS.  

There’s an ongoing glass-half-full versus glass-half-empty debate in AI — enormous potential on one side, significant cost on the other. We remain firmly in the glass-half-full camp. But if the market chooses a glass-half-empty view, we also know that lower prices often create some of the best long-term opportunities.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

• Broadcom FQ4 Earnings: $73B AI Backlog with Visibility; $162B Consolidated Backlog
• Coherent: Indium Phosphide Capacity to Double, Data Center to Reaccelerate to 10% QoQ
• Credo Fiscal Q2: Revenue Surges as Reliability Wins in a Crowded Market
• Nvidia Q3: Largest QoQ Growth in 2 Years; Networking up 162%