Author: io-fund

Penguin Solutions has come across our radar for a handful of reasons, with the first being that it is a smaller player in the memory market with a unique approach to CXL-based memory solutions (as we detailed in our early July newsletter series on offload engines and improving inference economics), as well as its strong YTD performance.   Shares in Penguin have risen 159% YTD and 212% since its low at the end of March, far outperforming the market, with a favorable technical setup.  Fundamentally, Penguin reported rapid YoY and QoQ growth in its memory segment, though this growth is dragged down by softer performance in its AI server and LED segments. Topline growth appears to be rather fleeting – management’s raised guidance for FY26 implies growth essentially flatlining sequentially into Q4. Management offered no indication there would be a sharp reacceleration (yet) in FY27.  However, that could change given Penguin is at the forefront of addressing the additional memory and orchestration requirements driven by agentic AI and inference. The company has designed its MemoryAI KV Cache server to offload KV cache into lower-cost, CXL-attached memory, helping support larger context windows during inference tasks. The company states its servers offer 8TB of DDR5 CXL memory and 11TB of total memory. The company also offers ClusterWare AI, which is a control plane for the server components.   Combined with a few other products, such as their core product CXL memory cards, the company is seeking to monetize the memory boom as a platform by combining hardware with operational software. The goal is that by supplying the servers, Penguin can also optimize inference performance by keeping large-context inference data closer to the workloads – accomplished with a five-product suite.  As you’ll also see below, the company is developing a Photonic Memory Appliance through a partnership with Celestial AI, recently acquired by Marvell. The product is intended to extend memory capacity and bandwidth for large-context inference workloads and represents a more forward-looking part to the thesis.   Penguin is certainly in the right place, and its $2.5B market cap means even small wins could help move the stock. However, at this early stage, execution is entirely unknown, and there is a long list of well-established competitors in the space (Dell, Super Micro, HP, among others including a sizable list of Chinese server companies). Therefore, technicals are needed to help offset the risk that Penguin identifies the right bottleneck but does not capture enough of the market. The opposite could also happen to where Penguin executes well; point being, the evidence is not in the financials just yet.  Penguin’s CXL Solutions and Value Add to Data Center Customers  Although Penguin labels itself as the ‘AI Factory Platform’ offering a range production ready AI servers utilizing AMD’s latest MI300-series GPUs and Nvidia’s Blackwell/Blackwell Ultra GPUs, it’s quite a small fish in the server OEM sea with barely $0.5 billion in server revenue while Dell and Super Micro are guiding for 100-140X that.   Penguin’s growth therefore is unlikely to come directly from the AI server space as it simply cannot match the pace and scale of leading vendors, but rather from memory and Penguin’s more unique positioning as it pertains to inference.   On that front, Penguin offers a range of DRAM and NAND modules as well as data center SSDs, though at far lower capacities than what SanDisk is moving to at 30TB for Penguin vs SanDisk approaching 256TB. Penguin’s unique memory product is its high-capacity CXL-based memory servers dedicated entirely to KV-cache offload, directly addressing one of the main memory bottlenecks in inference as our July 9 newsletter emphasized.   Penguin’s MemoryAI KV-cache servers offer up to 8TB of high-capacity DDR5 CXL memory, and up to 11TB of total memory to expand local GPU memory for KV-cache usage in agentic and inference workloads. Penguin says the server can offer up to 2X higher inference performance with up to 8X faster time to first token. The choice to leverage CXL memory makes the server 4-5X more cost effective than expanding memory via HBM, according to Penguin.   The company also offers a complementary ‘photonic memory appliance’ (PMA) leveraging photonic tech from Celestial AI (now Marvell), which could in the future help extend memory bandwidth and capacity across the rack, though this is yet to be seen as Marvell and Celestial have yet to fully ramp the tech. However, these are both new products, with the majority of its memory-related revenue coming from DRAM and NAND (SSDs) memory modules.   As stated, Penguin’s strategy is to become a ‘full-stack’ approach, combining servers with memory solutions and vendor-agnostic operating and monitoring software to facilitate cluster management under one roof. In terms of product lifecycle, Penguin offers a full start-to-finish approach, all the way from design and server integration to full deployment services:   “So as opposed to alternatives where customers may procure hardware and then they have to stitch it together, the value we provide to our customers, we provide full system integration of both our products as well as other products that will include the full AI factories or AI data centers for those customers, including design, build, deploy and manage. So that becomes an advantage for us.  In some cases, we are also managing these factories for the customers for 3 to 5 years, which, again, is an advantage for the customers to be able to get the ROI from solution we deliver.”  While this full-stack approach spanning hardware, software all the way to final deployment could serve as a key lever fueling Penguin’s land-and-expand strategy, the main drawback is that it’s not lending anything to margins, which are marginally better than other server OEMs.  Diversifying Away from Hyperscalers in AI Servers, but not in Memory  There is a bit of an interesting dichotomy between Penguin’s approach to its Advanced Computing (AI servers) and its Integrated Memory segments in terms of customer selection – the company is diversifying away from hyperscalers in AI servers yet leaning on that same group in memory.   Management has been quite blunt about its shift away from hyperscalers in Advanced Computing as it prioritizes engagements with sovereign AI firms, enterprises and neoclouds, all of whom likely are tailored much better to Penguin’s size and scale in terms of rack volumes. For example, Advanced Computing revenue was up 4% YoY to $138 million in Q3, as 81% YoY growth in non-hyperscale revenue to ~$80 million more than offset a (35%) YoY decline in hyperscale revenue to $58 million.  Additionally, that full-year revenue guidance for 22% growth “assumes that we will continue to diversify our customer sales mix and does not include any Advanced Computing AI hardware sales to hyperscale customers,” which implies that hyperscale revenue is essentially a no-growth business moving forward (if any at all come FY27). One of Penguin’s long-time hyperscaler customers has been Meta, with the two working on Meta’s RSC supercomputer platform, though management’s commentary suggests the two are moving in a different direction.  The oddity is that Integrated Memory is reliant on the hyperscale cohort for growth through FY27, with management emphasizing that they are seeing growing engagements from hyperscaler and enterprise computing companies for memory products, with its customer base broadening. Given the nature of diversifying away from hyperscalers in one unit but favoring them in another, analysts questioned what Penguin is supplying (and how this relationship is working out). Management responded that they are supplying memory modules to this hyperscaler, but added that it is just a portion of the overall memory segment and not a multi-million dollar customer.   Overall, it raises some questions about how Penguin will continue to drive growth and expand business at hyperscalers in one segment while simultaneously phasing them out in another.   Preliminary FY27 Guide Points to Acceleration to 30% Growth, from 22% Guided in FY26  While Penguin only recently reported fiscal Q3 earnings in July, management opted to provide an initial view into FY27, which at face value does not offer any shiny glimmers of growth.   To start, management raised FY26 guidance by 10 points in fiscal Q3, now projecting 22% revenue growth for the year versus 12% previously, riding memory’s coattails and sharp acceleration this quarter (as detailed below in the Financials). This would project FY26 revenue at roughly $1.67 billion, up from $1.53 billion previously – Memory’s strength at 90-95% YoY guided is offset by Advanced Computing’s guided (15-20%) YoY decline.   Management also chose to offer a preliminary forecast for 30% revenue growth in FY27 despite still having one quarter left in FY26, basing this on the strength of AI demand and customer demand signals. While Penguin will provide a full FY27 outlook next quarter, it reflects quite positively that management already has the confidence not only to guide for the next fiscal year, but to guide for a growth acceleration. However, it should be noted that this 30% guide falls roughly 18 points below Q3’s YoY growth and Q4’s estimated growth, implying a step-down in growth rates through next year, or a lack of visibility yet into fiscal 2H.  Given the early nature of the guide, analysts poked for a segment breakdown between Integrated Memory, Advanced Computing (AI server products) and Optimized LEDs, with Memory appearing to be the key driver:    Brian Chin, Stifel  I understand you may not want to hone in too much at this stage, but if we kind of look at your implied fiscal 4Q '26 maybe memory revenue could approach, let's say, $300 million quarterly.  If I just kind of flatline that and it probably grows on pricing, et cetera, next year, that could easily grow sort of mid-30%, maybe more percent year-over-year, so above sort of that baseline. But to say LED maybe is not super growthy in your assumption set for next year. Can you put some guardrails maybe on what the Advanced Computing growth could be in fiscal '27?  Nathan Olmstead  Yes. Brian, it is preliminary, right? We're just kind of — we're in the middle of our own planning for next year. We have pretty good visibility, I'd say, into the first half at this point with backlog and customer conversations that we're having. But I would say for Advanced Computing, probably mid-teens, something like that, would be kind of the starting point. Of course, we're still — we'd still have some of the impact year-over-year from the wind down of the Edge business and kind of the transition away from Meta, but a lot of that would have been [ lapped ]. So what you're starting to see next year in Advanced Computing is more of the growth coming through from the non-hyperscale AI Infrastructure business.”  Taking this at face value: mid-teens growth in Advanced Computing, using 16% YoY as a benchmark, along with low-single digit growth in LEDs, would require Integrated Memory revenue to come in at 45.5% YoY, a rather sharp deceleration from the 92.5% guided. This also would plot out barely holding on the double-digit QoQ growth on average through the year.   Memory a Double-Edged Sword: Key Growth Driver with Strong Backlog yet Numerous Headwinds  Memory is Penguin’s main avenue for growth, yet the street is not filled with just green lights ahead but also some larger potholes. Penguin is broadening its memory customer base and seeing solid momentum in landing and expanding new logos, combined with a strong backlog, yet management has previously flagged a handful of headwinds that could easily derail this growth story.  On the customer front, as noted above, engagements are broadening to both hyperscalers and enterprises, while Penguin also stated that a generative AI firm has continued to expand with CXL memory expansion card purchases. This follows Penguin’s land-and-expand strategy, with the company landing 16 new logos over the trailing twelve months in Memory, with five of those 16 already expanding business.   In terms of backlog, Penguin has not put a number on it, but emphasized that they had a “very strong” backlog exiting FQ3, and later said that demand and backlog are increasing despite rising memory prices.   However, while rising prices have yet to impact backlog, management explicitly called it out as a headwind to its outlook: “Our outlook also contemplates the industry-wide higher costs for memory, which may slow customer demand for our products and solutions and may lower our gross margins in our Advanced Computing and Memory businesses.”  Additionally, the current memory dynamics and component shortages could also limit Penguin’s growth and order fulfillment capabilities, with management also stating that extended lead times for certain memory components could impact “how quickly we can ramp existing and new customer projects and fulfill customer orders.”   So while pricing was a key factor in Q3’s Memory growth, and Q4’s guide as well, that same dynamic could impact margins, while further component shortages could also impact Penguin’s ability to ramp new deals, potentially creating revenue headwinds layering in beside those margin headwinds. Additionally, Penguin’s scale and margin profile similar to a server OEM does not lend it much wiggle room to combat said headwinds if they both arrive at once.   Financials  Revenue Growth to Moderate Sharply on a QoQ Basis  Penguin delivered a record FQ3 with revenue up 47.6% YoY and a sharp 39.6% QoQ to $478.7 million, marking a dramatic reacceleration from single-digit YoY growth over the last four quarters and flat QoQ growth in FQ2. Management noted "AI-driven" revenue — Integrated Memory plus the non-hyperscaler portion of Advanced Computing (roughly 58% or $80 million) — comprised 74% of total revenue, up 104% YoY.   For FQ4, revenue is projected to be $505.1 million based on the FY26 guide for 22% YoY growth. This marks a slight acceleration to 49.5% YoY growth, though sequential growth would moderate more than 34 points to just 5.5% QoQ, as pricing tailwinds fade.  As noted before, FY26 revenue growth was boosted 10 points to 22% YoY, implying revenue of $1.67 billion, while the initial FY27 guide for 30% would bring revenue to $2.17 billion. This would imply QoQ growth remaining in the high-single digit range through next year, essentially taking out the possibility of seeing a similar QoQ growth profile as FQ3.   Segments: Memory Drives Growth as Advanced Computing Declines  Integrated Memory was Penguin’s key growth driver and standout in FQ3 with revenue up 111.4% YoY and 60.3% QoQ to $275.1 million, primarily fueled by pricing along with volume. Management added that the majority of revenue comes from data center DRAM and NAND solutions, with CXL a new product line inferred to be a smaller portion of overall segment revenue.   For the full year, management raised Memory growth to 90-95% YoY, driven by pricing and volume, marking a more than 20 point raise from 65-75% guided in FQ2. This would imply FQ4 Memory revenue to be approximately $310.5 million, with YoY accelerating to 134.9% yet QoQ growth moderating sharply to just 12.9%.   Advanced Computing grew a modest 3.8% YoY and rebounded 18.9% QoQ to $137.6 million, following Q2’s (23.6%) QoQ decline as Penguin continues to phase out its high-margin Penguin Edge business and move away hyperscaler customers. Management noted that they added four new AI infrastructure logos in Q3, and across the trailing four quarters, landed 13 new logos with 7 already expanding.   For the full year, management guided for a (15-20%) decline for the segment based on those two factors, a slight improvement from (15-25%) previously. This implies Q4 revenue down (5.4%) QoQ and (5.9%) YoY to $130.2 million. Additionally, management also emphasized that its bookings to revenue lag in the segment remains at about 3 to 6 months, offering strong visibility into 1H FY27.  Optimized LED rose 7.2% YoY and 18.7% QoQ to $66.1 million, and full-year guidance for a (5%) decline implies growth comes in essentially flat in Q4, up 0.7% QoQ but down (1.4%) YoY, to $66.5 million.   Margins Mixed in Q3 and Remain Thin  Despite record dollar profitability, Penguin’s margin quality was mixed as adjusted gross margin fell below 30% for the first time in three years, while operating margins moved higher.    GAAP gross margin was 27.8%, down 1.5 points YoY but up marginally QoQ, while adjusted gross margin contracted 3.6 points YoY and 3.1 points QoQ to 28.1%. FY26 gross margin guidance of 26.5% GAAP/28.5% adjusted implies continued compression versus FY25's 28.8%/31.0%, and reflects fading pricing tailwinds: “Our Q4 outlook assumes less pricing favorability than we experienced in Q3, and we therefore expect some downward pressure on gross margins as we exit the year.”  GAAP operating margin surpassed 10%, coming in at 10.6%, up 7.6 points YoY and 3.1 points QoQ. Adjusted operating margin’s expansion was much more muted, up 1.5 points YoY and less than a point QoQ to 13.4%, a very modest gain set against memory peers riding pricing tailwinds, consistent with Penguin's positioning closer to a server-OEM value profile. This is primarily driven by Integrated Memory, which posted an operating margin of 22.6%, up 13 points YoY and 6.5 points QoQ, more than offsetting Advanced Computing’s 2.8% margin, which declined nearly 16 points YoY and 3.5 points QoQ.   FY guidance implies FQ4 GAAP operating margin to moderate sequentially to 8.6%, though this remains nearly 5 points higher YoY.  GAAP net margin reached 9.3%, up 8.5 points YoY though down from Q2’s 10.9%, aided by its investment in Celestial AI who was acquired by Marvell. Adjusted net margin was 10.9%, up 1.3 points YoY and 1 point QoQ.  EPS GAAP diluted EPS swung to $0.68 from a $(0.01) loss a year ago and improved from $0.58 in Q2. Adjusted diluted EPS was $0.84, up 78.7% YoY, the strongest quarter of the year so far. Penguin raised full-year guidance to $1.97 GAAP and $2.60 adjusted EPS, above its prior guidance for $1.30 GAAP and $2.15 adjusted.   Implied Q4 figures of $0.67 GAAP and $0.75 adjusted) suggest a step-down from Q3's pace, consistent with the flatter revenue guide and softer margin trajectory.  For FY27, similar to revenue, management forecast preliminary 30% YoY growth, which would project adjusted EPS to be approximately $3.38, while also hinting of minimal margin expansion considering EPS growth matches revenue growth at a 1:1 ratio.   Cash Flows Pressured in FQ3 The quarter's clear blemish was cash flow: operating cash flow was ($74.8) million for a (15.6%) margin, reversing from $55 million in FQ2 and $92.8 million a year ago, for margins of 16% and 19.9% respectively. The swing was driven by a sharp build in working capital — inventories nearly doubled QoQ to $498.3 million and accounts receivable rose to $703.0 million from $369.9 million — consistent with ramping Memory production ahead of demand.   Free cash flow was ($77.6 million) for a (16.2%) margin, similarly down from 15.6% last quarter and 19.3% a year ago.  Cash was $440.3 million, while total debt was $443.2 million.   Conclusion  Penguin sits at a unique intersection of the server and memory market, with the company aiming to capture inference demand via its new CXL-based KV cache servers. The company must still execute on the KV cache offloading platform as the majority of its revenue today remains tied to DRAM and NAND modules.   Overall, the fundamental profile for Penguin needs to improve to offer evidence Penguin can swim with the AI Server whales. The 40% QoQ growth was a great start, yet right now, the company is expected to see lumpy growth of 13% next quarter, plus margins and cash flows are not nearly as strong as the major memory players.     For more information on the technical setup the I/O Fund is eyeing, Join Knox Ridley at his 1-hour weekly webinars where he will discuss potential entries into Penguin and many other lesser-known AI stocks. Sign up for Advanced Market Signals here. 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. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Website Migration Notice I/O Fund will soon migrate to a newly redesigned website. Our official domain will remain io-fund.com. If you experience any issues during the transition, please contact premium@io-fund.com. Recommended Reading: Lam Research: Deposition and Etch Intensity Rising with More Complex Chips Seagate Q4: Price per Exabyte to Double While Cost Per TB Falls Marvell: Strong Interconnect Growth, but Really an FY28 Story Corning: Glass Manufacturing Powerhouse Pivoting Hard into AI Networking

As expected, and similar to last quarter, it's hard to find fault with Palantir’s headline numbers. Revenue growth continued to accelerate further on robust momentum in the company’s US Commercial segment, with FY26 raised once again by 14 points to more than 134% YoY growth. This roughly works out to ~21-22% QoQ growth in the back half following Q2’s 28% QoQ growth in the segment, as Palantir retains its crown as one of the strongest AI software companies that we track.    NRR continued to expand with Palantir already quickly closing in on 160%. However, deeper in the report, there were areas to nitpick, such as YoY growth in RPO, remaining deal value (RDV) and total contract value (TCV) all decelerating in unison. The latter two are now lagging revenue growth by 10 to 44 points. As we said last quarter, this may sound nitpicky, but when a company is priced to perfection, these subtle shifts in forward indicators matter.  Interestingly, despite the strong growth rates and strength in margins, Palantir has succumbed to the general malaise across the software sector, with shares underperforming the market the entire year so far, down (3%) YTD versus a 17% gain for the Nasdaq 100.   This quarter, Palantir took a bit of a dig at the leading AI labs this quarter and hinted at an aim to take business from OpenAI and Anthropic. Management also dropped a few tidbits hinting at 150%+ growth in US Commercial through year-end 2027 and a huge disconnect to consensus — we break all this down and the puts and takes for Palantir below.   Palantir Believes it Could Take Business from the AI Labs  As we discussed in the section ‘The Death of Legacy Software’ in our Q1 write-up, Palantir fired shots at software competitors who simply offer ‘slop’ in the form of workflow automation, compared to Palantir’s ability to organize complex or incomplete enterprise data into ontologies that generate actionable business intelligence.  This quarter, Palantir took another dig – but this time at the AI labs, presumably OpenAI and Anthropic. Although the word “slop” is used again, it’s less about AI labs not generating proportional business value and more about the high cost enterprises spend on token usage relative to what they get back:  “Enterprises that are not using Palantir are seeing their token meters spinning endlessly just to get slop without any correlation to value. This token model may be working for the labs, but it is not working for anyone else. It's breaking corporate budgets without results to justify the expense.”   The criticism stems from a fundamental difference in business models. OpenAI and Anthropic monetize through a consumption model, where customers are charged based on token usage. As inference workloads and agentic AI push token volumes higher, the labs are monetizing every prompt, reasoning step and inference into more revenue.   Contrast this to Palantir, with a monetization model that relies on enterprise agreements priced to capture enterprise adoption across workflows. Palantir refers to the business model of monetizing more closely to value rather than raw token usage as “economic value.” Here is how it was described on the earnings call:  “Now you can have a continuous improvement cycle that compounds your alpha into your weights. AIP was built for this. We continue to see that our product is winning head-to-head. The others are focused on productivity. We are laser-focused on turning tokens into real economic value for our customers. The reality is that the market has created far more intelligence than it has converted into value. A more powerful model does not solve this problem. The limiting factor is the rate of AIP deployment. A major Silicon Valley tech company recently ran a bake-off, a frontier lab and its deployment team against AIP and our forward-deployed engineers.”  Therefore, rather than encourage customers to consume more tokens, Palantir’s goal is to help enterprises generate greater business outcomes from every token.   In the call, the CTO also stated that Nvidia’s open-source Nemotron Ultra outperformed a leading frontier model, which leads to the importance of being model agnostic as more LLMs are released. Palantir has consistently stated that large language models will become commoditized, and are essentially interchangeable infrastructure, stating: “We have a product that allows you to switch out models. …  It's not about being beholden to one model. It's about bringing the right models to bear for the right purposes and our contracts, our structures are set up to do that to support and compound their alpha in the organization.”  This view is self-serving, of course, given commoditized LLMs shift the competitive advantage to the enterprise software layer that orchestrates them. In this case, it does not matter who has the best model, as Palantir seeks to monetize all of them.  Lastly, a bonus is that because Palantir does not rely on datasets and training, the company asserts it keeps enterprise data safer than the R&D Labs.  Karp Hints at US Commercial Growth Goal of 150%+ for Six Quarters  Before diving into the nitty-gritty world of Palantir’s key metrics, it’s worth diving into the US Commercial segment, as CEO Alex Karp laid out quite the positive goal, essentially implying the segment surpasses a $12 billion run rate by the end of 2027 – or 4X higher than its current run rate today.  Getting away from the long-winded metaphors, Karp hinted that he is aiming to drive “the business to grow at a rate equal or above to what we have in U.S. commercial for the next 18 months, which is a very high goal, but it is one we can actually get to because we are fully aligned with what's right and what's good and what actually works well in an enterprise.”  Notably, Karp is known for ambitious and aspirational statements, and this was not formal guidance. However, if we take the comment at face value, Karp is laying out a floor of ~150% YoY growth for US Commercial for the next six quarters, extending the 149% growth reported in Q2. This comes as a noticeably higher growth rate than what the raised guidance for 134% growth for the year suggests, as our current modeling points to 135% and 123% YoY growth in Q3 and Q4.   Plotting out this 150% growth floor for US Commercial, should it materialize, shows quite a large disconnect between the segment’s trajectory and overall consensus estimates for Palantir.   For example, maintaining that 150% YoY trajectory through Q4 2027 would project US Commercial revenue to reach $3.17 billion, or a ~$12.7 billion run rate. It would also project Palantir maintaining 28-30% QoQ growth each quarter (aside from a softer 17% QoQ estimate in Q1 2027), which could prove to be quite a challenging feat as revenue scales.   Comparing this to consensus estimates shows quite a profound disconnect, as the Street is currently modeling $3.54 billion in total revenue by Q4 2027 – meaning Karp’s commentary for 150% YoY growth takes US Commercial revenue to ~90% of that revenue estimate.   Assuming a 60/40 split for total Commercial revenue and Government revenue by the end of 2027 – which is a fair assumption considering Government is growing at a healthy 80% and still remains Palantir’s largest segment as of Q2 – this would project total Commercial revenue to be $2.12 billion. This means that the 150% trajectory is likely around ~70% higher than the Street is expecting, assuming ~$250-300 million in contribution from International Commercial.  If Karp can truly achieve what he is hinting at with the 150% growth for six quarters, then current consensus estimates for Palantir are far too low. For now, we view this as an upside scenario rather than something investors should bake into estimates.  Palantir Believes NRR Has More Room to Run Beyond 157%  Within the key metrics, Palantir has been a standout when it comes to expanding its net retention rate (NRR), which now hit 157% in Q2, up seven points QoQ, 29 points YoY, and a whopping 50 point expansion since bottoming at 107% in Q3 2023.   Despite the accelerated sequential step-ups in NRR in Q1 and Q2, management commented that there is room for more upside ahead as older customers are returning and expanding product adoption across Palantir’s suite:    NRR “is anomalously strong. That is also going to shift and, as hard as to believe, become even more positive because some of our older partners we haven't really interacted with, they also showed up. They're like, ‘Oh, okay, now we get why we would need you.’ And not just Foundry. They're migrating across our stack.  So customers that were only using Foundry, now want Ontology, now want to be part of the sovereign AI stack.”  To not just take Palantir’s word for it, evidence in quarterly deals signed supports this view, as Palantir is now beginning to lap quarters with much higher (and larger) deal counts. Palantir has noted that NRR “does not include revenue from new customers that were acquired in the past 12 months,” so as these quarters are lapped, and should these customers begin expanding deals, the runway for NRR continues.   As seen in the chart below, deals signed over the last four quarters took a step-up from the prior four – Palantir signed no less than 180 in each of the last four quarters, whereas the prior four quarters never surpassed 157. Deal sizes also have grown significantly, with Q2 seeing 73 $10M+ deals and the trailing four quarters having 234 $10M+, a nearly 2X increase from the 121 $10M+ deals in the prior four-quarter period.  TCV, RDV Remain on Watch (and Maybe RPO Soon Too)  The pieces to nitpick in Palantir’s report lie within TCV, RDV and potentially RPO, as all three of the key metrics decelerated in unison in Q2, with both TCV and RDV notably growing much slower than revenue on a YoY basis. Typically, this sort of deceleration can lead to lower future growth. Previously, decelerations in RPO historically preceded decelerating revenue growth in cloud’s slowdown through 2022 and 2023.  For TCV, we noted last quarter that timing issues led to a soft QoQ print for booked TCV growth, though the bigger issue at play was that booked TCV was essentially flat for four quarters when adjusting for the contract timing issue. While Q2 did see booked TCV improve on a dollar basis to $3.37 billion from Q1’s $2.41 billion, the larger issue now visible is a clear deceleration in TCV YoY growth.   TCV growth was 49% YoY in Q2, slowing from 61% in Q2 and a notable deceleration from the 138-151% growth reported from Q2 2025 through Q4 2025. Given the deal momentum, TCV theoretically should have a smooth path to growth, yet we’re seeing a ~90 point deceleration over two quarters. Additionally, this growth rate in TCV lags revenue growth by 43 points, widening from 24 points in Q1, with this serving as an early point of concern for revenue growth should this gap persist or widen further. RDV is also seeing a similar dynamic, up 83% YoY to $13.1 billion, decelerating 15 points from 98% YoY in Q1 and 22 points from 105% in Q4.  RPO growth came in at 102% YoY in Q2, and while this is still very robust to have RPO doubling from last year for a third consecutive quarter with triple-digit growth, the point to nitpick is that this marks a 32 point deceleration from Q2, and if this dynamic continues, there is a chance for RPO growth to then decelerate below revenue growth next quarter.    Financials  Revenue Growth Accelerates to 93% YoY, Guidance Raised Again  Palantir’s revenue accelerated to a record 92.8% YoY to $1.94 billion, accelerating more than eight points from Q1’s 84.7% and marking an exceptional 80-point acceleration since growth bottomed at 12.8% three years ago. Sequential growth also accelerated despite the larger revenue base, coming in at 18.6% QoQ versus the 16% QoQ growth from Q1. This also marked a solid 7.9% beat to Palantir’s guidance for the quarter for roughly $1.8 billion, and a 6.8% beat to consensus.   For Q3, Palantir guided for revenue of $2.162 billion at the midpoint, implying growth decelerating to 83.1% YoY; however, it should be noted that this dynamic was also present in Q2’s guide from last quarter, which pointed to growth decelerating more than five points to 79%. At the midpoint, sequential growth would also decelerate to 11.7%.   Management also raised FY26 guidance to $8.154 billion at midpoint, a meaningful raise from the $7.656 billion guide from Q1. This points to 82.2% YoY growth, raised more than 11 points from 71.1% from the Q1 guidance above. However, working backward, the new guidance implies a sharper deceleration into Q4 – a 20 point decel from Q2 toward the 72% level. Based on current momentum, and considering Karp’s comments for US Commercial sustaining 150% growth, there could be a degree of conservatism to the FY guide.   Commercial Growth Accelerates, 30 Points Faster than Government  Palantir’s Commercial segment drove growth in Q2, with US Commercial in the driver’s seat with nearly 150% YoY growth as outlined above. Notably, Commercial growth now outpaces Government growth by 30 points, up from the 20-point level in both the prior two quarters.  Total Commercial revenue hit $945 million, accelerating nearly 15 points to 109.5% YoY, while sequential growth reaccelerated nearly 8 points to 22.1%. Despite the strong growth, Commercial revenue has yet to overtake Government as Palantir’s largest segment, though it is on the cusp at 48.8% of revenue.  Total Government revenue accelerated less than 3 points to 79% YoY to $990 million, with sequential growth moderated slightly but remained healthy at 15.4% QoQ versus 17.5% QoQ in Q1. US government revenue, which makes up the bulk of the segment, grew 89.9% YoY to $809 million, accelerating more than five points from 84.2% in Q1. Sequential growth was similarly healthy at 17.8% QoQ, decelerating modestly from 20.5% QoQ in Q1.  International remains a soft spot for Palantir for both segments: International Commercial revenue grew just 26% YoY and 2% QoQ to $182 million, while International Government revenue grew 42% YoY and 5% QoQ to $181 million – both emphasizing how critical US momentum remains for Palantir’s growth.  Margins Continue to Expand, GAAP Operating Margin Approaching 50%  When looking into the software universe, and primarily enterprise software with high AI exposure, you would be hard-pressed to find a fundamental profile anywhere in the same realm as Palantir, not just in growth but also in margins.   Just take a look at Palantir’s Rule of 40, which expanded 61 points YoY to 155%.   GAAP gross margin dropped nearly 2 points QoQ to 84.7%, though this still marked a roughly 4 point YoY expansion. Adjusted gross margin contracted 1.6 points QoQ to 86.3%, though again this remained up 4 points YoY. Management said the sequential softness was due to “taking on cloud hosting for one of our government customers” which led to a higher cost of revenue in the quarter.  Operating leverage remained a highlight, as GAAP operating margin reached 47.1% in Q2, up less than a point QoQ despite the gross margin contraction, and up 20.3 points YoY, closing in on the 50% threshold. Adjusted operating margin came in at 61.7%, up 1.5 points QoQ and 15.4 points YoY, marking the second consecutive quarter above 60%. For Q3, guidance points to adjusted operating margin easing modestly to 59.9%.   GAAP net margin was 54.9%, up 1.6 points QoQ and 22.3 points YoY, marking the second straight quarter in the low-to-mid 50% range; adjusted net margin was similarly strong at 54.1%, up 1.6 points QoQ and 13.8 points YoY.  EPS Growth Remains Exceptional, Deceleration Guided for Q3  Palantir’s EPS continued to grow at triple-digit rates, driven by margin expansion down the line. GAAP EPS was $0.41 in Q2, up 215% YoY, while adjusted EPS also reached $0.41, up 156% YoY and beating estimates for $0.35. Management said that unrealized gains from its stake in SpaceX created a $0.03 tailwind to GAAP and a $0.02 tailwind to adjusted EPS.   For Q3, guidance implies GAAP EPS of $0.38, decelerating more than 100 points to 111% YoY, while adjusted EPS is expected to remain steady at $0.41, similarly decelerating 65 points to 91% YoY.  Cash Flows Robust with 60%+ Margins  Last quarter we called Palantir’s cash flows “exceptionally strong,” yet this quarter cash flows were even better with cash flow margins moving into the low-60% range.  Operating cash flow was nearly $1.22 billion for a 62.8% margin, up 7.7 points QoQ and 9.1 points YoY.   Adjusted free cash flow was $1.22 billion for a 63.1% margin, up 6.5 points QoQ and 6.6 points YoY YoY. For FY26, Palantir raised its adjusted FCF forecast to $4.5-$4.7 billion from the $4.2-4.4 billion guided in Q1, representing a 56.4% margin and implying FCF generation remains strong in 2H.  Palantir ended the quarter with $9.4 billion of cash and marketable securities, while debt remained zero.  Conclusion  Palantir’s revenue accelerated to 93% YoY while adjusted operating margin expanded to almost 62%, a combination of growth and profitability that is unmatched anywhere in software. US Commercial reaccelerated to 149%, its fastest print at scale, and NRR pushed to 157% with management signaling still more room to run as older customers migrate across the stack.  The bull case hinges on Palantir sustaining momentum. The CEO hinted at 150% US Commercial growth for the next six quarters, if realized, it would carry the segment to a $12.7B run rate exiting 2027 – or nearly 70% above what the Street is modeling. However, Alex Karp is often sensational, and this can’t be taken as formal guidance. For now, we are treating this an upside scenario to track over the coming quarters.   It’s been well known for quite a while that Palantir’s valuation is not cheap at 51X forward revenue, thus the softness in the key metrics across TCV and RDV decelerating remains on watch. Along with RPO, these metrics serve as leading indicators for forward growth, and any blips or hiccups in the story could drive a re-rating for its rather steep valuation.   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. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Recommended Reading: Lumentum FQ4: The AI Optics Winner Keeps Winning Coherent Q4: AI Optics to Drive $3B in Quarterly Revenue Western Digital FQ4: HAMR to Ramp in CY27; LTAs Need More Clarity Cloudflare Q2: The Agentic AI Inflection is Beginning

Lumentum is a tough act to follow with revenue doubling year-over-year and four consecutive quarters of 20%+ growth. However, Coherent only recently inflected from 6%-7% QoQ growth to now 13.3% QoQ growth this quarter and 12.4% expected next quarter. The YoY basis is also clearly in Lumentum’s shadow at 34% YoY growth this quarter, and 45.4% next quarter. If we look at Coherent on an individual basis, some important milestones were reached this quarter, such as reaching $2 billion in quarterly revenue for the first time. Management stated they will reach $3 billion in quarterly revenue by YE27 whereas analyst consensus has this at $2.63B. Similar to other networking stocks we’ve covered this past week, 800G provides a nice base while 1.6T is ramping faster than expected. For Coherent, 6-inch indium phosphide should relieve a critical supply constraint, ahead of OCS scaling and CPO/NPO beginning to contribute in 2027. Data Center Growth Guided to 80% as 6-inch InP Removes Bottleneck  Data center grew 66% YoY and 24% QoQ and is expected to “exceed 80%” in the upcoming quarter. This is being largely driven by more InP production as Coherent stated they produced approximately 80% more InP lasers YoY during the June quarter, stating they currently have the lasers to support the 80% YoY growth rate. This marks an important acceleration for Coherent that will be further supported as more InP comes online with management stating: “Our six-inch indium phosphide capacity expansion is a key driver of revenue growth and margin expansion. We remain on track to double our internal indium phosphide output capacity year-over-year by the end of the current quarter, one quarter ahead of our original plan.” Texas and Sweden are ramping 6-inch production today, while Zurich is expected to become the company’s third 6-inch manufacturing location in 1H 27. According to the call, this is 4X the output as 3-inch but at half the cost: “Because if you remember, we've talked about the fact that 6-inch wafers are, we get four times the amount of output from that wafer versus 3-inch, right? Four times, but it's at half the cost. So it's that cost structure that's beneficial to us.” This ties back to the CFO discussing gross margin expansion: “Along with strong revenue growth, we expect continued gross margin expansion and operating leverage, enabling us to grow EPS significantly faster than revenue.” Coherent’s Road Map Over the 1-2 Years is Loaded  Coherent’s product road map looks a lot like Lumentum’s road map – and that’s a good thing. The investor presentation lays out the following timeline:  Optical circuit switching (OCS) – ramping now with a $4B TAM CPO/NPO arriving in 1H 27  Multi-rail in 1H 27  Thermal Solutions in 2H 27 Coherent believes this represents $20 billion of incremental 2030 SAM compared to $50 billion for its current SAM. Of these, OCS carried the more near-term bullish tone with management doubling the addressable market estimate from $2 billion to $4 billion.   This is because OCS has shifted from primarily scale-out to now scale-across and eventually scale-up: “Originally when we started working on OCS, we were thinking about it mostly in the context of scale-out. But now, clearly, we think we'll see adoption in scale-across, and then a clear path to scale-up as well. We have active customer engagements in scale-up applications. That's really what led us to double the size of our market outlook at OFC earlier this year. We doubled it from 2 to over 4 billion. We may have even been conservative on that $4 billion number in terms of the addressable market over time. I think that was for a 2030 timeframe”  Regarding CPO/NPO, it was stated CPO is seeing a pull forward and NPO is seeing a new leg up in demand over the past 3-6 months. The content opportunity could be significant for players like Coherent and Lumentum, by supplying the laser, module, and many other optical components. Coherent stated every major strategic customer is engaged on CPO, NPO or both. This is leading Coherent to launch Photon Link in September, which is a new integrated-optics platform.  “At the ECOC industry event in September, we plan to unveil Coherent PhotonLink, our new platform for integrated optics. PhotonLink spans the complete optical signal chain from light generation and beam shaping through transmission, detection, and conversion back to an electric signal for the XPU or switch chip. The platform supports CPO, NPO, and other forms of optical integration […] We expect initial revenue from PhotonLink-related products to begin in our December quarter.”  Quick Note on China:  Management stated on the call that Coherent would likely benefit from any China restrictions as they are the largest U.S. supplier of transceivers with 20 U.S. production facilities. This could provide Coherent a considerable tailwind. “As the largest U.S. supplier of transceivers, something like that would certainly be beneficial […] One of the things that we are really proud about on our manufacturing is, of course, we are a global manufacturer with locations all over the world, which gives us resiliency. Of course, we are also vertically integrated. We build a number of the very important components ourselves. Simon, as you mentioned, we have an outstanding footprint in the U.S. […] We were founded over 50 years ago as a U.S. manufacturing company. We have a great footprint, we could certainly build off of that.” Financials  Revenue Growth Accelerates as Datacenter & Communications Scales  Coherent reported FQ4 revenue of $2.05 billion, up 33.7% YoY and 13.3% QoQ, beating estimates by 3.3%. This marked a solid acceleration from 20.5% YoY growth in FQ3 as the Datacenter & Communications segment inflected higher on strong optical component demand from AI buildouts. Management also flagged 42% YoY growth on a pro forma (organic) basis, a more than 14 point acceleration from 27.6% reported in Q3.   Sequential growth was a highlight in FQ4, as Coherent reported its first double-digit QoQ growth since September 2022. This also marked a more than 6 point acceleration from 7.1% QoQ in FQ3 and a roughly 10 point acceleration since the start of FY26 at 3.4% QoQ. On a dollar basis, sequential growth was ~$240 million, doubling from the $120 million added in FQ3.   For FQ1 FY27, Coherent guided revenue to $2.2–$2.4 billion, implying YoY growth would accelerate further to 45% YoY and coming in 8.1% ahead of estimates for $2.13 billion. At the $2.3 billion midpoint, QoQ growth would hover around 12.4%, roughly steady with the current acceleration trend.  For FY2026, revenue was $7.12 billion, up 22.5% YoY, as Coherent exited the year nearly 2X faster than the 17% growth rate it held entering FY26. For FY27, initial estimates point to revenue of $9.66 billion, accelerating to 35.7% YoY, but the raise to Q1 suggests FY27 estimates could soon follow.  Segments: Data Center Accelerates to 66% YoY, Q1 Guided 80%+  Datacenter & Communications remains Coherent’s main growth driver, contributing nearly 80% of revenue, yet the segment’s growth continues to be weighed down by declines in Coherent’s Industrial segment. Revenue in the Data Center & Communications segment was nearly $1.62 billion in FQ4, up 58.6% YoY and 18.6% QoQ, with both figures accelerating from 40.6% YoY and 12.7% QoQ growth in FQ3.   In Data Center, revenue increased 66% YoY and 24% QoQ, a rather sharp acceleration from 37% YoY and 13% QoQ growth reported in Q3. Coherent is continuing to ramp capacity, reiterating from Q3 that it remains on track to double its InP capacity by year-end 2026 and double it again by year-end 2027. Management also noted that its 6-inch wafer lines are producing EMLs, CW lasers and photodiodes, with better yields than 3-inch lines.   For Q1, management stated that they would expect the 80% growth in lasers in Q4 to translate almost directly over to Data Center growth in Q1, or YoY growth accelerating beyond 80%: “You know, that 80% growth in in lasers in the June quarter? We use those — those basically go into transceiver shipments in the current quarter. And so we would expect our data center growth. For instance, this quarter on a YoY basis to exceed 80%.”  For Communications, revenue increased 56% YoY and 11% QoQ in Q4 on strong DCI demand, though this moderated a bit from 60% YoY and 16% QoQ in Q3.  The offset is the Industrial segment, where revenue declined (15.8%) YoY and (3%) QoQ to $430.5 million, the third straight quarter of YoY declines and the sixth straight quarter of QoQ declines. This continual softness means Industrial is acting as a bit of a drag on Coherent’s overall growth rate.  Margins  While margins did expand across the board in FQ4 on both a YoY and QoQ basis, the pace of expansion and level of margins remained quite a bit below peer Lumentum.   GAAP gross margin was 38.5%, up 2.8 points YoY and less than a point QoQ, while adjusted gross margin was 40.2%, up 2.1 points YoY and less than a point QoQ. For FQ1, adjusted gross margin was guided to improve marginally sequentially to 40.5%. This compares to GAAP and adjusted gross margins of 47.4% and 50.4% respectively for Lumentum in FQ4, with both expanding 12-14 points YoY and 2-3 points QoQ.  GAAP operating margin was 12.4%, up 12 points YoY against an impairment charge-impacted quarter and up 1.3 points QoQ. Adjusted operating margin was 21.8%, up 3.8 points YoY and 1.5 points QoQ, and FQ1 was guided to expand nearly 1 point sequentially to 22.7%. For comparison, Lumentum’s operating margins were 15 points higher at 27.8% for GAAP and 36.6% adjusted.   GAAP net margin was 11.8%, swinging from (6.3%) a year ago and improving from 10.6% in Q3. Adjusted net margin reached 17.2%, up 4.6 points YoY and 1.9 points QoQ.  For FY2026, Coherent saw slight margin expansion with operating leverage more visible: GAAP gross margin expanded 2.3 points to 37.5%, and adjusted gross margin expanded 1.5 points to 39.4%. GAAP operating margin expanded 7.6 points to 12.6%, while adjusted operating margin rose 2.7 points to 20.5%. GAAP net margin expanded 10.4 points to 11.3%, while adjusted net margin rose 3.5 points to 15.4%.  Adjusted EPS Growth Accelerates to 74%  Coherent reported GAAP EPS of $1.19 in Q4, up from an ($0.83) loss in the year-ago quarter tied to restructuring and impairment charges. Adjusted EPS was $1.74, up 74% YoY and beating estimates by 7.4%. This also marked a 19 point acceleration from 54.9% YoY growth in Q3 and a swift rebound from 35.8% growth in Q2.  For FQ1, Coherent guided for adjusted EPS to be $1.85-$2.05, representing a slight moderation to 68% YoY growth at the $1.95 midpoint; similar to revenue, this came in 10.2% above the consensus estimate for $1.77.  For FY25, GAAP EPS was $4.12, swinging positive from a ($0.52) loss in FY25, while adjusted EPS grew 59% to $5.61. Initial estimates for FY27 point to more than 52% growth in GAAP EPS to $6.28, while adjusted EPS is projected to grow more than 49% to $8.37, signaling more margin expansion and operating leverage is expected.   Cash Flows and Balance Sheet  Cash flow was the major soft spot in Q4’s report – despite operating cash flow returning to positive territory after dipping negative in Q3, Coherent’s already-large negative FCF widened further in the quarter.  Operating cash flow was $69 million in FQ4 for a 3.4% margin, rebounding from a (5.2%) margin in Q3 but remaining below the 8.5% margin from the year ago quarter. For FY26, operating cash flow was a thin $79.5 million or 1.1% margin, down from 10.9% in FY25.  Free cash flow was approximately ($486.3) million in the quarter for a (23.8% margin), the third straight quarter of negative free cash flow and a further deterioration from ($383.5) million in FQ3 for a (21.2%) margin, and just ($1.1) million a year ago. For FY26, free cash flow was ($1.02 billion) or a (14.4%) margin, down sharply from 3.3% in FY25. The primary driver of this was capex more than doubling YoY from $441 million to $1.1 billion as Coherent prioritizes capacity expansion to meet accelerating customer demand. Commentary from management about doubling InP capacity by year-end and doubling again in 2027 suggests capex will likely remain elevated and pressure FCF into next year.  Cash and equivalents thinned sequentially, dropping from $1.59 billion in Q3 to $1.16 billion in Q4, while debt was $3.22 billion.   Inventories rose more than 21% QoQ to $2.58 billion, outpacing revenue growth, while accounts receivable rose 13% QoQ to $1.34 billion, both consistent with scaling production ahead of demand. Conclusion: Coherent is not only in Lumentum’s shadow, but they also report one day apart from each other. Where Coherent could set itself apart is two ways – the first is the 6-inch InP capacity could help narrow the gap with Lumentum on growth rates. As you’ll recall from previous analyses, this transition is fairly new. Secondly, the large manufacturing footprint could become a tailwind should there be any China restrictions. There are also many growth opportunities for both companies with optics only in its first inning. That last point is probably the most important one. 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 own shares in COHR at the time of writing and may own stocks pictured in the charts. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Recommended Reading: Lumentum FQ4: The AI Optics Winner Keeps Winning Western Digital FQ4: HAMR to Ramp in CY27; LTAs Need More Clarity Cloudflare Q2: The Agentic AI Inflection is Beginning Applied Optoelectronics Q2: Demand is 20% to 40% Above an Already-Aggressive Ramp

In July, the I/O Fund team wrote a deep dive on networking entitled: “AI Networking in 2026: What’s in Motion Tends to Stay in Motion.” Lumentum embodied the thesis of that report last night, which is that networking is evolving to solve crucial bottlenecks for training, and more importantly, inference workloads, to where the demand driving this transition shows no signs of slowing down.  From the core driver EMLs to the incoming CW lasers, from the newer opportunity in OCS, to the incoming NPO/CPO opportunity – if it’s crucial to increasing speeds and lower latency in AI clusters, then Lumentum is there.   This quarter, revenue increased 109% YoY and 24.5% QoQ for the third consecutive quarter of 20%+ sequential growth. Adjusted gross margin crossed 50% much earlier than anticipated and the guide to $1.25B at the midpoint is being reached one quarter earlier the company’s previous target model. The guide implies a fourth consecutive quarter of over 20% sequential growth with YoY accelerating to 134.2%.  Below, we look at how the acceleration is broad-based and why Lumentum keeps winning.  EML Lasers Remain Strong while CW Lasers Prepare to Layer-In  Components revenue reached $649 million, increasing 22% QoQ and 103% YoY. Within that segment, Lumentum delivered another record quarter for EMLs, driven by 100G per lane. Meanwhile, the transition to 200G per lane now accounts for 25% of total EML revenue, with the company expecting 200G EMLs to represent the majority of shipments in 2027.   Here is what was stated on the call: “Even as we allocate additional capacity to CW lasers, we remain on track to deliver over 50% EML unit growth by the December 2026 quarter compared to the year- ago quarter.”  As noted in the statement above, CW lasers are expected to be the major growth driver soon, to where EMLs are more constrained by InP capacity. On one hand, this could create a roadbump for Lumentum as they prepare to pass the baton from one leading component to another, with CW lasers having more competition. CW lasers also do not contain modulators like EMLs when they ship, presumably leading to lower sales prices and margins.   However, management pushed back on this in the Q&A. This is a long quote but a necessary one as it explains Lumentum’s strategy to keep momentum as the market transitions to CW lasers for the 1.6T transition.  “ Michael Hurlston 
President and CEO  Chris, good question, and I will give some commentary and then have Wupen comment as well. One, no slowdown in our EML demand. I think it was Papa that asked a minute ago, we see still a significant supply-demand imbalance on our EMLs. As Kathy put in the prepared remarks, we are still very much on track to increase our EML output year- over- year, but even at the end of year, we would expect to be significantly behind demand.   So no slowdown at all in the demand that we are seeing from EML. We acknowledge, and we have said this before, that we would expect and are seeing CW lasers take a significant portion of transceiver output. Silicon photonics is a viable solution at 1.6T.  […]  What we are doing as part of that is allocating some of this excess output to CW lasers, and we are now shipping CW lasers in a relatively meaningful way in this 200 G- per-lane silicon photonics opportunity. What has also changed for us is we have significantly reduced the die size of our CW laser.  So before when you and I talked, we said, hey, probably from a margin standpoint, EML is better. Now, we have brought things much more in line. CW lasers are smaller. They are better performing. We are commanding a significant price premium, as I said to you in a previous question, against what we view as the market price because of the performance of these lasers. As such, the margin opportunity is still better on EMLs. But that gap has closed considerably, I think, since the last time we talked.”  It was also discussed by their C-suite that EMLs will remain strong as they are today on 200G per lane for 800G; with a mix of EMLs and CW lasers, and that ultimately EMLs will remain oversubscribed:  “Number one, if you look at the application today, there is a lot of 200G-per-lane volume that is on 800G. In that, we actually see a much higher EML share versus CW laser share. Michael talked about dynamics. I think this will both continue. EML will remain a very important player in 200G-per-lane. As 1.6T ramps up, I think we will see more scheduled lasers. As Michael talked about, our new design of 200G CW laser is much more efficient, therefore smaller. It gave us a much better gross margin profile.”  OCS Grows Triple Digits; Pump Lasers and Narrow-Linewidth Lasers Surge  As we look beyond the core products of EMLs and CW lasers, the next current driver of growth is optical circuit switches, which we’ve covered in the past here. Lumentum doubled OCS QoQ and expects fiscal Q1 to be the first quarter with more than $100 million in OCS revenue. This puts Lumentum on track to reach $400 million by YE CY26, with demand for FY27 said to be incredibly strong:  “After doubling shipments from fiscal Q3 to Q4, our guidance includes our first triple-digit OCS revenue quarter. The demand signal for 2027 continues to be incredibly strong, and we have started the initial work to add capacity with contract manufacturers, as well as continuing to increase output in our internal factories. Since our last call, the roadmap for OCS has also come into better view.”   Later, in the Q&A, it was stated OCS is “doing a bit better than expected.”  After OCS, another strong driver in the current earnings report was pump lasers and narrow-linewidth lasers for the scale-across opportunity. This was a welcome surprise last quarter and growth continued to surge this quarter (albeit on small numbers). Pump lasers are said to be sold out and grew 80% YoY while narrow linewidth lasers grew 130% YoY and reported their 10th quarter of growth.  
Here was a killer quote on how connecting network capacity compares to previous buildouts:  “Expanding, inferencing, and training applications are driving full-rate connectivity between data centers, while political and regulatory constraints favor smaller, more modular builds.  These two factors, among others, are substantially increasing the demand for our pump laser solutions. To put this in perspective, for one major hyperscaler, the network capacity connecting just two AI data center sites is double the total global backbone capacity they built over the entirety of the last decade”   NPO and CPO Opportunity Sees Initial Revenue   Looking beyond these products, the foundation for CPO and NPO is being laid to where Lumentum will participate with their high-powered lasers. Management stated they will see their ultra-high-power laser revenue (UHP) reach $50M this year and are on track to report their first $100M quarter by fiscal Q3.   This may seem like low revenue, but the race to supply AI systems for scale-up will be intense, with many vying for this coveted spot (i.e., with Nvidia).   Read more on the NPO and CPO opportunity here.  According to Lumentum’s management team, CPO production plans are on schedule and demand has actually strengthened despite rumors that CPO was pushed out another year, Lumentum said the opposite – that UHP laser demand will begin ramping in 2H 2027 with scale-up deployments in 2028 (which is right on time, for now at least).  Here is why Lumentum may continue to win as NPO and CPO offers an equal, if not greater opportunity, as the scale-out market.   “These architectural shifts represent a major market inflection that plays directly to our core strengths as a premier laser chip manufacturer, benefiting us as optics begin to penetrate the copper domain. NPO offers a faster time-to-market option by placing optical engines on the board right next to the XPU accelerator, trading power and cost for simplicity in optical scale-up applications. Customers are evaluating two types of laser chips for NPO: a mid-power laser integrated directly with the optical engine, and a high-power laser used in an external light source module. Our mid-power lasers inherit the reliability and engineering of our flagship high-power platform. Our family of NPO and CPO lasers utilizes common design and process know-how to achieve industry-leading efficiency across 120 mW, 150 mW, and 400 mW output levels.  Looking ahead, CPO continues to be viewed as the natural end state on the technology roadmap, placing optics directly on the substrate or interposer for maximum power efficiency through foundry-level advanced packaging.”  Lumentum has also received their first purchase order for an external light source (ELS) module. Delivery is expected in 2H27, which helps Lumentum to move beyond supplying individual laser chips into a higher-ASP module. This is likely connected to some of the China rumors, given Lumentum is deeply integrated with Chinese module makers, to where Lumentum is bringing the module manufacturing in-house.  Financials  Revenue Growth Guided to Accelerate 25 Points to 134% in Q1  Q4 capped off a strong year for Lumentum, with revenue up 24.5% QoQ and 109.3% YoY to $1.01 billion, its first ever quarter above the $1 billion mark and beating estimates by 1.9%. YoY growth accelerated 19 points from 90.1% in Q3 and 51 points from the start of the year, while sequentially Lumentum recorded its third consecutive quarter with QoQ growth above 21%.   For Q1 FY27, Lumentum guided for revenue of $1.225 to $1.275 billion, or $1.25 billion at midpoint, implying a 25 point acceleration to 134.2% YoY and QoQ growth holding steady at 24.2% despite the larger revenue base. Management also stated that Q1’s guide signaled the company reaching its target model for a $5 billion annualized run rate more than a quarter earlier than expected. For FY26, revenue totaled $3.01 billion, up 83.2% YoY, while initial estimates point to growth accelerating further to 88.9% to $5.69 billion in FY27. Considering that Q1’s guide beat estimates by nearly $100 million, there is a higher likelihood that FY27 revenue estimates are revised higher in the coming weeks.  Components Accelerates to 103% YoY on Strong Scale-out, Scale-Across Demand  Components remained a key driver for Lumentum, accounting for 64.5% of revenue in Q4 and benefitting from robust demand for scale-out and scale-across components. Components revenue was $649.4 million, accelerating more than 25 points to 102.7% YoY and up 21.8% QoQ, a slight acceleration from 20.2% QoQ in Q3. For the full year, Components revenue was nearly $2.01 billion, up 79.7% YoY.  Lumentum said that it set new records for 100G and 200G EML shipments, with 200G EMLs accounting for more than 25% of total EML revenue in Q4. For CW lasers, Lumentum said it expanded 200G shipments across multiple transceiver customers.   On the scale-across front, growth remained robust – narrow linewidth lasers for DCI and scale-across solutions recorded a tenth consecutive quarter of sequential growth while YoY growth accelerated 10 points to 130% YoY. Pump lasers maintained the same growth as in Q3 at 80% YoY, with Lumentum signing multiple LTAs with customers for scale-across deployments.  Lumentum provided a brief update on CPO, noting that ultra-high-power (UHP) laser shipments for CPO are growing and on track to generate “meaningful” revenue exiting calendar 2026. Systems Grows 30% QoQ with First 1.6T Shipments Systems revenue grew faster than components this quarter, but that was largely due to its smaller base accounting for just 35.5% of revenue. Q4 revenue was $356.9 million, up 122.5% YoY (roughly in line with 121.1% in Q3) and 29.7% QoQ, accelerating from 24% QoQ growth in Q3. For the full year, Systems revenue was $1.01 billion, up 90.7% YoY.  Lumentum said Q4’s growth was led by record 800G cloud transceiver shipments, adding that it began initial 1.6T shipments in the quarter with a portion using its internal CW lasers. We covered this dynamic in Q3, Lumentum FQ3: Firing on All Cylinders Despite Stiff Supply Constraints Across EMLs, Pump Lasers, explaining that insourcing CW lasers is expected to augment transceiver margins as 1.6T ramps, while offering better yields and lower scrap rates.    Management added that its OCS ramp remains on track, implying Lumentum remains confident in meeting its $400 million target in calendar 2H 2026 and ramping to >$1 billion in calendar 2027 with demand strengthening from customers. Margins Continue Expanding, With Adjusted Operating Margin Guided to 40%  Notably, Lumentum is operating above its $5 billion target model on the margin front, with multiple levers to pull moving through FY27 as higher-margin products ramp.  In Q4, GAAP gross margin was 47.4%, up 3.2 points QoQ and 14.1 points YoY. Adjusted gross margin was 50.4%, up 2.5 points QoQ and 12.6 points YoY. This sits above management’s $5 billion annualized revenue model, which calls for adjusted gross margin to be 45-48%.  GAAP operating margin was 27.8%, up 6.2 points QoQ and 29.5 points YoY, moving further past the operating losses that characterized FY24 and FY25. Adjusted operating margin was 36.6%, up 4.4 points QoQ and 21.6 points YoY. For Q1 FY27, management guided for adjusted operating margin of 40.0%, up 3.4 points QoQ and more than 21 points YoY; again, this comes in above the $5 billion model which forecast adjusted operating margin to be 33-37%.  GAAP net margin was (711.7%), though this was driven by a $7.76 billion non-cash loss on debt extinguishment tied to the equitization of Lumentum's convertible notes. Excluding this charge, adjusted net margin was 32.4%, up 4.5 points QoQ and 19.2 points YoY. For FY26, margins showed strong expansion across the board (with the exception of GAAP net margin due to Q4’s debt extinguishment). GAAP gross margin expanded 13.7 points to 41.7%, while adjusted gross margin expanded 11.3 points to 46%. GAAP operating margin expanded 28.3 points to 17.4%, and adjusted operating margin expanded 20.1 points to 29.8%. GAAP net margin was (230.1%), while adjusted net margin expanded 17.1 points to 26%.  Adjusted EPS Beats and Guided Higher  Driven by the debt extinguishment, GAAP EPS was ($84.65) in Q4. Adjusted EPS, which excludes the charge, rose 267% YoY to $3.23, beating the consensus estimate of $2.97 by 8.8%.   For Q1 FY27, Lumentum guided adjusted EPS of $4.20, accelerating to 281.8% YoY and coming in roughly 15.7% ahead of the consensus estimate of $3.63. For FY26, GAAP EPS was a loss of ($92.96) versus EPS of $0.37 in FY25, again reflecting the extinguishment charge; adjusted EPS was $8.67, up 320.9% YoY. For FY27, GAAP EPS is estimated to be $16.04, while adjusted EPS is estimated to increase 116% to $18.73.  Debt Cut in Half  Cash and equivalents were $2.74 billion, while total debt was $1.64 billion, as the convertible note equitization referenced above roughly cut Lumentum's debt load in half sequentially.  Accounts receivable rose nearly 18% QoQ to $520.3 million while inventories rose a more modest 9.3% QoQ to $691.6 million, both continuing to build at a dollar pace similar to Q3 as the company scales to support demand.   Conclusion:  Lumentum is our top position, thus, we are pleased to see the company keep pace with others we’ve given a high allocation. What is most impressive about Lumentum is not only the growth but the breadth of the growth. EMLs are supply constrained, CW lasers are preparing to layer-in, OCS is now a strong contributor, and NPO/CPO are what could help the stock continue to earn its keep in our portfolio through 2027 and 2028.  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 own shares in LITE at the time of writing and may own stocks pictured in the charts. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Recommended Reading: SanDisk FQ4: Inference to Drive Demand and Visibility for Years Applied Optoelectronics Q2: Demand is 20% to 40% Above an Already-Aggressive Ramp Cloudflare Q2: The Agentic AI Inflection is Beginning Western Digital FQ4: HAMR to Ramp in CY27; LTAs Need More Clarity

Western Digital’s report was not as strong as its peer, given Seagate’s HAMR products are ramping now whereas Western Digital will be ramping HAMR in CY27. All-in, Western Digital will trail Seagate by about a year.  Also contributing to weak price action following the earnings report last week was the lack of clarity regarding the long-term agreements. There were at least three clear attempts by analysts to get visibility on pricing terms, yet management did not provide enough specifics to assign a premium valuation (detailed below).  Regardless of any one quarter, HDD stocks are bound to be winners once inference truly takes off. There is simply no way around the fact that agentic AI will drive explosive growth in data repositories and storage requirements. Below, we weigh the puts and takes of this earnings report, with the understanding this is a stock with a strong, long-term horizon.  Long-Term Agreements Need More Clarity  This quarter, Western Digital’s management team stated a significant portion of the cloud business is now under LTAs, with discussions extending into 2029, 2030 and 2031. On the surface, this could be a positive as it seems to support more than 25% annual exabyte growth. However, despite many attempts to understand pricing flexibility, it remains an unanswered question if the LTAs are capped or uncapped.   Here’ s an example of how management answered one attempt for more clarity, although all attempts were answered similarly:  “And to reiterate our whole pricing strategy there, is to make sure it's predictable and we give a lot of visibility to our customer. Where we've seen a bit more ability to drive more rapid price increases has been in the non-nearline space. If we fast forward that to what we are looking at in terms of what we are discussing with customers on '29, '30, '31, we're working through the specific commercial constructs of the LTAs, but we have a good line of sight and visibility to the volume requirements of our customers that sort of further strengthen our conviction in that 25%-plus exabyte growth. What we're working through is the pricing regime of how we would affect that for those years going forward.”  The most I could extract from the Q&A was a comment that LTAs have a “base price associated to a base volume” with incremental upside subject to a different pricing structure. This is key because although long-term agreements reduce uncertainty around demand, they also reduce upside if pricing becomes increasingly locked-in.  HAMR is a 2027 Story and Nearline Exabyte Growth Decelerates   HAMR is coming in the first half of 2027 according to management, yet this will be nearly a year after Seagate by the time the product line ships in volume. This was called out on the earnings call with an analyst stating: “It's hard not to compare your results with your main competitor where they're seeing better sequential top line growth and targeting gross margins nearly 200 bps higher than your September guide. So curious what you make of this? Is that due to the earlier ramp of HAMR? Is it a difference in timing of contracts, perhaps you're selling fewer bits into the open market?”  The overall update on HAMR is that 2027 should be a year when Western Digital more fully competes with Seagate: “Then obviously, we have the 44-terabyte HAMR coming out in the first half of calendar year '27, which will ramp. And then obviously, we've indicated in our road map, we will have 50-terabyte products coming out towards the second half of calendar year '27.”  Nearline exabytes shipments also came in materially slower than peer Seagate. WDC’s nearline shipments were 209 EB in FQ4, up 5% QoQ and 22.9% YoY, marking a more than 14 point deceleration from 37.2% YoY. The QoQ growth represents just 10 incremental EB shipped sequentially. While this was an increase from the 7 incremental EB shipped in FQ3, this was dwarfed by Seagate’s 20 incremental EB added in the quarter as it grew from 175 EB in FQ3 to 195 EB in FQ4.   This EB shipment growth also stands out as notably softer than Seagate, which reported a more than 5 point acceleration to 11.4% QoQ growth while YoY growth decelerated slightly but remained rather robust at 42.3% YoY.   This raises a legitimate question of whether WD is ceding some nearline share to Seagate in the current cycle as Seagate moves quicker with higher-capacity HAMR, ramping its 44TB Mozaic4+ platform with 50% of HAMR exabytes expected to be Mozaic4+ by the end of CY26, whereas WDC is not expected to ramp 44TB HAMR until 1H CY27.  On the call, analysts questioned this, stating that Western Digital is slipping in terms of growth compared to previous quarters. Here was a question from Evercore:  “I guess maybe the first one, is exabyte shipments were up about 21%, I think, year-over-year to 23%. That's below the 30% exabyte growth you folks have had in the last several quarters. Can you just talk about, is this a temporary pause related to timing or supply? Or does this represent a more normalized growth rate? And how should we think about exabyte growth through fiscal '27?”  Management responded saying that exabytes will see demand growing above 25% moving forward, “And as we move into the second half of the year, as we really ramp up the 40 terabyte shipments, as I mentioned, that will be over 50% of the nearline exabytes we ship. We expect exabyte growth rate to accelerate. And then obviously, we have HAMR coming on in the second half of the year and into calendar '27 as well at the 44 terabyte level.”  AI Inference Remains the Bigger Picture  If we step back from this quarter’s results, then it’s important to not forget the broader investment thesis is centered around AI inference. Western Digital already sees about 90% of its HDD bits from cloud customers, and yet the AI industry is at the very beginning of a multi-year expansion for exabyte demand.   Inference creates additional data to not only support reinforcement learning, but also because storing the context generated by AI workloads can be more economical than repeating the data through compute and memory resources. As explained in the CEO’s opening remarks, economics favor retaining more data rather than continuously regenerating it.  Today, the largest AI platforms process tens of billions of tokens per minute, and per the CEO, this is creating volumes of data that must be stored and managed. Agentic AI will make storage requirements even more pronounced.   Here is what was stated:  “Recent disclosures show token volumes growing several fold year-over-year, underscoring the pace at which inference is scaling. Meanwhile, AI is moving rapidly from merely answering questions to Agentic AI that does the work, coordinating tasks, accessing data and operating continuously across multistep workflows. This transition creates a fundamentally more data-intensive workload and one that is increasingly persistent rather than transient.  The storage implications are significant. Agents generate data at every step of the workflow, increasing both the volume of data created and the amount that must be stored over time. This is why we continue to view Agentic AI as a structural and step function driver of capacity-orientated storage demand.”  Financials  Revenue Growth Mixed as QoQ Accelerates While YoY Decelerates  Western Digital reported FQ4 revenue of $3.75 billion, up 12.3% QoQ and 43.8% YoY and only beating estimates by 1.4%. QoQ growth accelerated marginally from 10.6% in FQ3, though YoY growth decelerated slightly from 45.5% in FQ3. This stands in contrast to peer Seagate, which saw QoQ growth accelerate six points to a faster 16.6% QoQ pace with YoY also accelerating more than four points to 48.5%.  For fiscal Q1 2027, Western Digital guided for revenue of $4.1 billion, +/- $100 million, coming in 2% above estimates for $4.02 billion and representing growth of 9.4% QoQ and 45.5% YoY. This points to a roughly 3 point deceleration in sequential growth and a 1.7 point acceleration in YoY growth, again softer than peer Seagate’s guide for 13% QoQ growth in the quarter.   For FY26, WDC’s revenue increased 35.7% YoY to $12.92 billion, with current estimates for FY27 pointing to revenue of $19.13 billion, accelerating to 48.1% YoY. Early estimates for FY28 project revenue of $25.96 billion, decelerating back to the mid-30% range at 35.7% YoY.  Data Center (Cloud) Pointed to Decelerate in Q1   Data Center (Cloud) remains WDC’s main growth driver contributing 89% of revenue, with the forecasted QoQ deceleration on the topline driven by the segment, while soft growth in nearline exabyte (EB) shipments raises some concerns over WDC’s competitive positioning against Seagate.    Data Center (Cloud) revenue was $3.34 billion in FQ4, up 12.3% QoQ and 43.2% YoY, marginally accelerating from 10.6% QoQ in FQ3 but decelerating from 48% YoY. Assuming Cloud maintains its 89% revenue mix again in FQ1, revenue would project out to $3.65 billion, decelerating roughly 3 points to 9.4% QoQ while YoY growth would accelerate slightly to 45.5%.  Client represented 6% of total revenue in FQ4 but posted standout growth of 34.7% QoQ and 60.6% YoY to $225 million, which is likely to be driven primarily by pricing rather than unit growth, consistent with broader component pricing dynamics this year.   Consumer revenue declined (6.4%) QoQ to $187 million, though it remained up 37.8% YoY; the sequential softness may reflect the same pricing actions weighing on consumer product demand.   Gross and Operating Margins Expand Sharply on Higher-Capacity Drive Mix  WDC displayed strong margin expansion in FQ4, aided by pricing and improving mix of higher capacity drives, with this expansion expected to continue into FQ1 though at a slower pace. Net margin continued to benefit from gains from WDC’s prior stake in SanDisk.   GAAP gross margin was 54.1% in FQ4 while adjusted gross margin was 54.4%, both up 3.9 points QoQ and 13.1 points YoY, driven by pricing and a larger mix of higher-capacity nearline drives, including the ramp of next-generation ePMR drives (up to 40TB/drive) that began shipping this quarter. For FQ1, WDC guided for adjusted gross margin to be 55.5%, slowing to a sequential expansion of 1.1 points QoQ but still up 11.6 points YoY.  GAAP operating margin was 41.7%, up 6 points QoQ and 15.6 points YoY, while adjusted operating margin reached a record 44.2%, up 5.6 points QoQ and 16.1 points YoY, signaling tailwinds from operating leverage as revenue growth continues to outpace opex growth. For FQ1, WDC guided for adjusted operating margin to be 45.9%, up 1.7 points QoQ and 15.5 points YoY.  GAAP net margin benefitted again from gains in WDC’s SanDisk stake, which totaled $2.05 billion in FQ4, pushing GAAP net margin to 85.3%. Adjusted net margin was 36.9%, up 5.1 points QoQ and 13.2 points YoY.   For FY26, GAAP gross margin was 48.9%, up 10.1 points YoY, and adjusted gross margin was 49.1%, up 9.7 points YoY. FY26 GAAP operating margin was 34.5%, expanding 10 points, while adjusted operating margin was 37.3%, up 12.9 points. GAAP net margin was 72.9%, benefitting from the SanDisk stake, while adjusted net margin was 30.5%, up 11.4 points.  Adjusted EPS Growth Accelerates to Triple Digits  As mentioned above, GAAP net income was distorted by the SanDisk stake, though adjusted net income, the cleaner read for WDC’s operations, is showing growth accelerating in FQ4 and into FQ1.   GAAP diluted EPS came in at $8.21 in FQ4, essentially flat QoQ versus $8.20 in FQ3 which carried a similarly outsized $2.73 billion gain; YoY growth comparisons off of these prints are not reflective of WDC’s true earnings power.   Adjusted EPS, which excludes the SanDisk gains, offers a clearer view for earnings growth, which is accelerating into FQ1 driven by margin expansion. FQ4 adjusted EPS was $3.56, beating estimates by 7.9% YoY and accelerating more than 12 points to 109.4% YoY. For FQ1, WDC guided for adjusted EPS to be $4.00, +/- $0.15, implying a further acceleration to 124.7% and marking the third consecutive quarter of accelerating growth. Current estimates call for adjusted EPS growth of roughly 120% YoY to continue into FQ2 2027 as well.  For FY26, adjusted EPS was $10.22, up 103.6% YoY from $5.02 in FY25, while GAAP EPS was $24.28, up 445.6% YoY from $4.45 — again, a figure almost entirely a function of SanDisk stake gains rather than core operating performance. For FY27, adjusted EPS is expected to nearly double again to $19.96, up 95.3% YoY, while GAAP EPS is forecast to decline (21.9%) YoY to $18.96 as FQ3 and FQ4’s stake-impacted quarters are lapped.   Cash Flow Margins Reach the High-30s While Aggressive Deleveraging Continues  Cash flows were a strong point in FQ4’s report as cash flow margins expanded to the mid to high-30% range, helping WDC continue quickly deleveraging its balance sheet, with debt now down to just $1 billion versus nearly $7.5 billion entering FY25.  Operating cash flow was $1.39 billion in FQ4 for a 37.1% margin, up 3.4 points QoQ and 8.5 points YoY, representing its highest OCF margin on record since early 2012. For FY26, operating cash flow was $3.93 billion, up more than 132% YoY for a 30.4% margin, versus a 17.8% margin in FY25.   Free cash flow was $1.28 billion for a 34.2% margin, up 4.9 points QoQ and 8.3 points YoY. For FY26, FCF was $3.51 billion, up 145% YoY and representing a 27.2% margin, up from 15% in FY25.   On the balance sheet, WD continued to aggressively pay down debt, ending FQ4 with just $1.05 billion in total debt, down from $1.58 billion in the prior quarter and $4.7 billion a year ago. Cash and equivalents were $1.58 billion, down slightly from $2.06 billion last quarter as WDC worked to pay down debt.   Inventories rose 11.3% QoQ and 17% YoY to $1.51 billion, a healthy build with sequential growth just ahead of revenue growth and supporting potential double-digit growth at the upside of guidance in FQ1.  Conclusion:  Regarding this quarter’s results, the long-term agreements need more clarity, and Seagate appears to have the upper hand on HAMR impacting results in 2026. For at least two more quarters, Western Digital is likely to be in Seagate’s shadow.  However, is this about Western Digital versus Seagate, or is this about the inference tsunami that is gathering more strength every quarter? If our thesis on inference and Agentic AI proves correct, both companies should find themselves on the right side of one a large secular storage expansion, one that HDD industry has not experienced in decades (if ever).  We tend to exit stocks that don’t compete with peers yet are bookmarking this one for 2027.  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 own shares in WDC at the time of writing and may own stocks pictured in the charts. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Recommended Reading: AMD Q2: Venice Meets Helios in 2H26 and 2027 SanDisk FQ4: Inference to Drive Demand and Visibility for Years Applied Optoelectronics Q2: Demand is 20% to 40% Above an Already-Aggressive Ramp Cloudflare Q2: The Agentic AI Inflection is Beginning

Shots have been fired, with Cloudflare’s management team calling hyperscaler businesses commoditized and, overall, an “unattractive” business. Given I sit in on many earnings calls, I have to say, this was one of the more exciting ones this quarter.   For a company with about $3 billion in annual revenue to call into question the Big 3, whose cloud businesses generate a combined $400 billion, is bold. But perhaps we should hear it out.   Regarding this quarter, the results were “good,” but not spectacular like many of our hardware stocks. Instead, what caught my attention was the compelling argument that Cloudflare’s business model is better suited than the hyperscalers for the era of agentic AI. That’s why I want to get directly to the point.   Also, I think NET could be right.  Cloudflare’s Business Model is Fundamentally Different  I was keenly interested in this earnings call, and couldn’t help but smile when these shots were fired: “So I think not all revenue is created equal. And if you're selling what is just commodity compute, if you're basically letting an AI company use your balance sheet,and your credit rating in order to buy servers that are the same as everybody else's servers. Then like that's just not an attractive business for us.”  You have to love the David and Goliath dynamic here. What Cloudflare is describing is that AWS, Google Cloud and Azure as having the less desirable business model as they buy machines with the goal of renting them back for 5X what they paid for them rather than optimizing the hardware for maximum utilization and selling the output: “The traditional first-generation clouds are in the business of buying a server and then trying to sell it back, lease it back and get 5 turns of revenue off of it. We’re in a very different business, where we’re selling actually work getting done.”  The difference is that hyperscalers offer heavyweight compute as they build around virtual machines and containers, with customers provisioning blocks of compute with low utilization rates. To best describe this, it was stated: “As Rita on our team walked through at Investor Day, if every knowledge worker on Earth ran an agent that was running in a container, we don't have enough CPU to actually power that. We need to increase the amount of CPU that's accessed by many orders — many, many times.”  Meanwhile, Cloudflare’s business model is lightweight, offering an isolates architecture (we’ve covered this in detail here). Rather than reserve a container or virtual machine for every task, Cloudflare uses lightweight isolates to run small pieces of code across its global, distributed network.   The distributed network fits the way that agents retrieve data with agents retrieving data, storage and APIs across many locations. Here is how it was described on the earnings call:  “The agents need a new generation of the cloud. And that new generation of cloud, even things like containers are too heavy” […] “And so what we've done, which is a new version, which is a much lighter weight in terms of our sandboxing technology, which we called isolates, that is — that gives you the ability and the scale to actually deploy and run code in a way that is going to keep up with the demands that agents have.”  Perhaps where Cloudflare is most unique from the hyperscalers is its lower capital intensity. Call it good timing, or call it good planning, to have Cloudflare report a slight acceleration in revenue growth combined with lower capex as a percentage of revenue could not have come at a better time given hyperscalers greatly disappointed on this line item last week.  Here are the brief numbers with a more detailed financial analysis below  Last quarter, Cloudflare reported growth of 4.1% QoQ and 33.5% YoY. Capex as a percentage of revenue was 9%  This quarter, Cloudflare reported growth of 8.8% QoQ and 35.9% YoY. Capex as a percentage of revenue was 7%  Capex for the year is expected to be 14% to 15% of revenue, an attractive number given the backdrop being set by Big Tech, with Google at about 40% of revenue right now.   Non-Human Internet Traffic Surpasses Human Traffic  Agentic AI is driving an explosion in internet traffic with Cloudflare stating that non-human internet traffic surpassed human traffic for the first time ever: “For the first time in human history, in Q2, more than 50% of the traffic flowing across Cloudflare's network was not human. The number of requests on our network from AI agents continues to grow unabated.”  There are key reasons that Cloudflare is uniquely positioned to benefit from machine-to-machine traffic over the internet. Specifically, Cloudflare manages 20% of the internet and that number is in the mid-30% when considering the top 10,000 sites. By managing a leading percentage of internet traffic, agentic AI will route through Cloudflare’s reverse proxy and CDN to fetch data and trigger APIs.   More traffic leads to greater demand for Cloudflare’s core products, such as its CDN, reverse proxy and Workers Platform. To have this much traffic, some potentially malicious, also can lead to more revenue for Cloudfare’s security products.   Additionally, although AI agent internet traffic only now surpassed human traffic for the first time, that traffic will eventually grow by orders of magnitude. Although Cloudflare’s management team is biased, here is how they framed where this is headed: “if the current trends continue, we think in 5 years, nonhuman traffic will be as much as 1,000x as much as human traffic. In other words, humans will be a rounding error on the Internet, not because human traffic goes down, but that's just how fast we're seeing nonhuman traffic grow.”  By sitting perfectly in the middle, Cloudflare also aims to become a payment and monetization layer to where AI agents can pay a fraction of a penny for content, APIs and digital resources.   I said in my introduction the management team is bold, here’s another example – although the real takeaway is related to just how perfectly Cloudflare is positioned:  “But we handle, let's say, about 0.5 billion requests per second through Cloudflare's network. We roughly estimate that somewhere between 1% and 10% of those, you could monetize through some sort of a micro transaction. Again, these will be tiny fractions of pennies. But that means that you day 1 on launching, something like this, you need to be able to support, call it, 10 million transaction — financial transactions per second and be able to scale up to, call it, 100 million financial transactions per second.  To give you some sense, Visa — and again, these are from memory, but Visa, which is the largest payment network in the world at peak during the holidays, handles about 20,000 transactions per second.”  Developers are Flocking to NET; Cloudflare OS  The number one key metric that supports the idea Cloudflare could be permanently inflecting ahead of the Agentic AI boom is its developer numbers. The company now has 7.4 million developers on its platform after adding nearly 2 million during the quarter alone.  To compare, Nvidia has 7.5 million developers on its CUDA platform and Google has about 9 million developers. In fact, the number was so surprising that Cloudflare’s CEO had his team “triple check the developer numbers.”  Here is what was stated as to why Cloudflare is seeing a spike in developers:  “The first is that Cloudflare Workers is turning out to just be the perfect platform for building agents and agentic workloads. It's extremely lightweight, it's — you only get charged for when it's actually doing work. You can spin things up and spin them down very, very quickly. And so it has become the go-to place for sophisticated developers to be able to launch code.”  For more information on Cloudflare Workers and WebAssembly, read our analysis here.  Recently, Cloudflare also open-sourced Cloudflare OS, an internal platform that is used to deploy AI across departments. This allows employees to build AI-powered workflows securely across an organization. According to management, the interest has been very strong, and even though many AI startups have done something similar, their approach is unique and more desirable because NET is, at its core, a security company.   Financials  Revenue Growth Accelerates to 35.9% YoY, Fastest Pace Since Q1 2023  Cloudflare reported Q2 revenue of $696.1 million, beating estimates by 4.5%. Revenue growth came in at 35.9% YoY, a 2.4 point acceleration from 33.5% YoY in Q1 and marking Cloudflare’s fastest growth print since Q1 2023. Sequential growth similarly accelerated to 8.8% QoQ from 4.1% in Q1.   For Q3, management guided for revenue of $736-$737 million, which at face value would represent a deceleration to 31% YoY and 5.8% QoQ growth. Although this points to some moderation from the pace set in the first half of the year, the guide still implies a healthy sequential dollar build of roughly $43 million, only modestly below the $56 million added in Q2.  For the full year, Cloudflare guided to revenue of $2.864 to $2.87 billion, or 32.2% YoY growth. As mentioned above, this figure sits noticeably below the 33.5%/35.9% YoY growth delivered in the first half of the year, suggesting management's full-year guide may carry a degree of conservatism baked into 2H growth.   Key Metrics Remain Relatively Healthy, NRR Rebounds, $100K ARR Customer Growth Accelerates  Cloudflare’s key metrics were par for the course, remaining healthy with some modest accelerations in $100K ARR customer growth and RPO – nothing exceptionally outstanding and no major red flags.   NRR rebounded to 120% in Q2, up 2 points QoQ from 118% in Q1 and up 6 points from 114% a year ago, though it has yet to meaningfully break above the 120% level it also touched in Q4 2025.   Cloudflare’s $100K ARR customers hit 4,698, accelerating from 25.2% growth in Q1 to 26.6% YoY this quarter. Sequential growth similarly accelerated from 2.7% QoQ to 6.4% QoQ this quarter, representing a healthy addition of 282 customers in the cohort this quarter.   RPO accelerated from 36.4% to 38.2% YoY to $2.73 billion; sequential growth also rebounded 5.5 points to 7.4% QoQ. Current RPO, the portion recognizable within 12 months, grew similarly and represented 64% of total RPO at $1.75 billion, consistent with recent quarters. Current RPO growth was roughly 7% QoQ, accelerating from 4% QoQ, and 35% YoY, roughly in line with Q1’s 34% growth.    Billings were more mixed, coming in at $753.5 million as YoY growth decelerated 3 points to 34.8% YoY while QoQ growth rebounded more than 4 points to 6.2%.  Restructuring Charge Distorts GAAP Margins; Underlying Trends Still Mixed  GAAP operating and net margins both showed a sharp sequential contraction in the quarter, though this was driven by $150.7 million restructuring and severance charges; on the flip side, adjusted operating margin expanded with the pace of acceleration picking up into Q3.   Gross margins showed a modest rebound in Q2 but remained lower on a YoY basis. GAAP gross margin was 71.8%, down 3.1 points YoY but rebounding less than a point QoQ. Adjusted gross margin followed the same pattern at 73.1%. The YoY contraction has been a multi-quarter trend and bears watching to see whether this rebound can continue into Q3.  The severance and restructuring charge was the primary driver of GAAP operating loss widening to $(205.7) million, or a (29.6%) margin, down sharply from (9.7%) in Q1 and (13.1%) in the year ago quarter. Excluding this charge, adjusted operating margin was 13.8%, up 2.4 points QoQ from 11.4% in Q1 but down slightly from 14.1% in the year ago quarter.   For Q3, management guided adjusted operating margin to expand further toward 17.6%, a slight pick up in the pace of acceleration to 3.8 points QoQ. Management also guided for FY?26 adjusted operating margin to be 15.5%, implying a strong finish with solid expansion in 2H, as margins were 11.4% and 13.8% in 1H.  GAAP net margin was (24.4%), weighed down by the restructuring charge, versus (3.6%) in Q1 and (9.8%) in Q2 2025. Adjusted net margin was 15.5%, up nearly one point QoQ and YoY from 14.7% in those respective quarters.  Adjusted EPS Growth Decelerates   GAAP EPS was ($0.48) in Q2, versus ($0.07) in Q1 and ($0.15) in Q2 2025, a function of the restructuring charge rather than any deterioration in underlying earnings power. Adjusted EPS was $0.29, up 38.1% YoY, a deceleration from 56.3% YoY growth in Q1 and slightly beating estimates for $0.27.   For Q3, management guided adjusted EPS to $0.34, which implies YoY growth decelerating further to 25.9%. Full-year adjusted EPS was guided to be $1.25-$1.26, implying ~35% growth at midpoint with room to improve should Q3 and/or Q4 beat guidance.   Cash Flow Margins Compress in Q2  Cash flow margins compressed in Q2, with operating cash flow margin falling to its lowest level since Q1 2023.   Operating cash flow was $117.6 million in Q2 for a 16.9% margin, down 7.8 points QoQ from 24.7% in Q1 and down 2.6 points YoY from 19.5% in the year ago quarter.   Free cash flow was $56.4 million, falling into single-digit territory once again at an 8.1% margin, down about 5 points QoQ from 13.1% in Q1, though up 1.6 points YoY from 6.5%.  The sequential compression in cash flow margins is worth monitoring as network capex steps up in 2H, with management maintaining a guide for 14-15% of revenue despite Q1 and Q2 capex being just 7% and 9% respectively.  Cash and cash equivalents were $1.66 billion, while convertible senior notes totaled $3.27 billion.  Conclusion:  The I/O Fund has a 7-year history of covering Cloudflare as a company starting in the Cloud Era, and we are going on 3-years of covering the stock’s AI angle with the first analysis on this thesis called “Bringing AI Inference to the Edge”.   Finally, after a few years, if you look closely enough, you will see the quiet and subtle beginning of an important inflection. The developer surge is the most obvious sign but accelerating revenue while keeping capex reasonable is a nice beginning as it communicates the company has strong product-market fit, without needing to participate in the same capital-intensive arms race as the hyperscalers.   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 own shares in NET at the time of writing and may own stocks pictured in the charts. Recommended Reading: Astera Labs Q2: Scorpio is Locked and Loaded AMD Q2: Venice Meets Helios in 2H26 and 2027 SanDisk FQ4: Inference to Drive Demand and Visibility for Years Applied Optoelectronics Q2: Demand is 20% to 40% Above an Already-Aggressive Ramp

This quarter, Macom’s Data Center segment officially pulled ahead of its Defense segment, and this lead is expected to widen as we move into Q4 and beyond. This isn’t too surprising given data center revenue was up 40% sequentially and up 85% year-over-year for revenue of $137.6 million.   Management also provided an update to its indium phosphide (InP) portfolio, which consists of 200G photodectors (PDs), 25G DFB lasers, and the incoming CW laser opportunity. However, Macom is a very diversified company with a portfolio spanning many products, best stated by the CEO in his opening remarks: “As a reminder, our portfolio is highly diversified supporting NRZ, PAM4 and coherent modulations across EML, silicon photonics and VCSEL based architectures. Our connectivity solutions include IC and photonic semiconductors.”  Also important, Macom produces its photonic products in its own fabs and has internal InP manufacturing capabilities. This offers a distinct competitive advantage that supports the supply for the company’s product portfolio.   Macom is expected to benefit from the need for higher density interconnects and the copper to optical transition, with a customer list for $10M+ customers has doubled over the past two years with a few in the $50 million to $100 million.   Proliferation of Optical Links in Data Centers  Our research site has covered many times the migration from passive copper toward fiber in scale-up networks. Where Macom sits, is deeper in the stack than AAOI or Lumentum, to where they supply the components inside the transceivers. As we covered following AAOI, one of the components their shipments depend on is TIAs, and they are currently highly constrained. TIAs, drivers, along with many other components for optical transceivers, is where Macom’s product portfolio resides. The company’s indium phosphide (InP) portfolio is gaining traction across a few key products, with 200G photodetectors (PDs) being a product where management has high confidence: “Not only do we have what we believe to be one of the industry's best 200G PDs, but we've also launched higher data rate PDs including a 400G PD which is now in the hands of perhaps a dozen customers as they think about the next-generation interconnects. So we are on the very front edge of the market as it relates to photodiodes.” It was also specifically called out that PDs are ramping in volume and becoming a meaningful contributor to data center growth. Another product called out in the earnings call is the 25G DFB lasers, which are experiencing an industry-wide shortage, with management stating: “Customers are coming to us with urgency due to the general supply shortage of indium phosphide DFB lasers.”  While PDs and 25G DFB lasers are driving revenue today, the opportunity as the industry moves toward near-packaged optics and co-packaged optics is expected to drive demand for Macom’s multi-channel TIAs and drivers, and other components. The company is sampling 200G and 400G-per lane TIAs and driver products, with the understanding it may not have a major revenue impact until 2028. Another key product to keep an eye-on down the line is the 75-milliwatt CW lasers, with management foreseeing these reaching production in 2H FY27: “Customers have been providing us with positive feedback on our product's performance and there is intense interest and customer pull to get us into production.” Management stated CW lasers could lead to a “watershed moment” in 2028.  For more information on Macom’s extensive product portfolio, read our previous analysis: “Macom: Data Center Revenue Accelerating to 35% QoQ” Record Bookings from 800G and 1.6T  Current data center revenue is being driven by 200G PAM4 products used in optical modules, plus ZR/ZR Lite 100G-per-lane products. In contrast, record bookings are being reported due to the incoming next-generation 800G and 1.6T deployments. This has led to the book-to-bill rising from 1.3 in Q1 to 1.6 in Q3, with backlog at an all-time high. Management also stated Data Center revenue can grow approximately 50% in fiscal year 2027. It’s also important to note that management sees both TIAs/drivers and photonic content representing separate multi-billion dollar opportunities: “At the end of the day, both classes of products are in our mind multi billion dollar product areas that we can service.” Financials  Revenue Accelerates to 35.7% YoY; Q4 Inflecting Higher to 60.8%  Macom’s FQ3 revenue came in at $342.2 million, inching ahead of estimates for $335.6 million. YoY growth accelerated more than 13 points from 22.5% in FQ2 to 35.7%, while QoQ growth accelerated a notable 12 points to 18.4% QoQ, Macom’s fastest sequential growth in more than nine years.   For FQ4, Macom guided for this strong momentum to continue, projecting revenue to be between $415 to $425 million, landing nearly 15% above estimates for $365.6 million. At the $420 million midpoint, this points to sequential growth accelerating further to 22.7% QoQ, and a sharp 25 point acceleration to 60.8% YoY.   Data Center Maintaining 35% QoQ in FQ4, Becomes Largest Segment  This revenue acceleration in both Q3 and Q4 is being driven by Macom’s Data Center segment, which surpassed Industrial & Defense this quarter to become Macom’s largest segment at 40% of total revenue.   Data Center revenue was $137.6 million in Q3, accelerating 45.5 points to 81.5% YoY while QoQ growth inflected 25.6 points to 40.1% QoQ, beating guidance for 35% QoQ growth. FQ4’s guide was arguably more impressive, with management guiding for a second consecutive quarter of 35% QoQ growth, implying revenue of ~$185.7 million. This would project YoY growth to accelerate nearly 52 points to 133.3% YoY, an incredible trajectory from Q1’s 31.4% growth; it also led to an increased FY guide for Data Center to grow 74%, up from 60% previously.   Industrial & Defense growth was more modest at 10.5% QoQ and 23.3% YoY to $133.4 million. However, the guide implies a similar acceleration pattern, with Q4 growth guided to reach 20.0% QoQ and 38.5% YoY. This more than 15-point YoY and 10-point QoQ acceleration is likely aided by Macom’s ongoing fab capacity expansion, including the RF business fab in Research Triangle Park.  Telecom remained the laggard of the three end markets, growing just 1.7% QoQ and decelerating from 7.5% to 4.7% to $71.3 million. Even so, the Q4 guide points to some reacceleration here as well, to 4.1% QoQ and 12.4% YoY.  GAAP Operating Margin Hits Record High in Public History  You might be tired of hearing the words ‘operating leverage’ this quarter, but that was evident once again with Macom as the company delivered its highest GAAP operating margin in its public history this quarter.  GAAP gross margin was 58.3% in Q3, expanding 1.4 points QoQ and 3 points YoY, while adjusted gross margin reached 59.7%, up 1.2 points QoQ and 2.1 points YoY, with both likely benefiting from the larger Data Center contribution. For Q4, Macom guided adjusted gross margin to improve further to 60-61%, up less than a point QoQ and 3.4 points YoY.  GAAP operating margin reached a record 22.5%, up 4.9 points QoQ and 7.5 points YoY. Adjusted operating margin surpassed the 30% level at 31.5%, up 3.6 points QoQ and 6.3 points YoY.   GAAP net margin jumped to 29.4%, up 13.4 points QoQ and 14.9 points YoY, though this was aided by a $41.5 million gain on investment tied to Macom’s stake in IQE plc. Adjusted net margin still expanded a healthy 2.9 points QoQ and 5.0 points YoY to 32.1%.  Adjusted EPS Guide Implies Sharp Sequential Growth  Driven by the investment gain, GAAP diluted EPS was $1.28, up 167% YoY. Adjusted diluted EPS came in at $1.40, up 55.6% YoY and a narrow 3.7% beat.   For Q4, MACOM guided adjusted EPS to $1.97-$2.03, well above the $1.57 estimate heading into the print, implying adjusted EPS growth accelerating more than 57 points to 112.8% YoY.   Cash Flows and Balance Sheet  Cash flow margins dipped sequentially but otherwise remained relatively healthy, while inventories rose at a moderate 12% QoQ pace.   Operating cash flow was $80.0 million in Q3 for a 23.4% margin, down from a 27.2% margin in Q2 and 24.0% a year ago. Free cash flow was $59.2 million for a 17.3% margin, also down from 22.7% in Q2 and from 20.5% a year ago.   Cash and short-term investments totaled $663.0 million, while debt stood at $340.5 million.  Accounts receivable rose more than 12% QoQ and 38% YoY to $179.1 million, while inventories increased nearly 12% QoQ and 31% YoY to $281.5 million, a still-measured pace of inventory build that, combined with the strength in Data Center bookings, suggests this growth trajectory has room to continue. Conclusion:  If I were to coin a phrase for Macom, it would be “Highly Diversified,” as this company is not dependent on a single product, architecture or end market. Instead, its breadth is central to the thesis of optical content proliferating across AI clusters.  With current growth supported by PAM4 products and photodetectors, and future growth potentially coming from CW lasers and 400G-per-lane products, Macom is attacking the opportunity from all angles rather than relying on a single product bet. 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 own shares in MTSI at the time of writing and may own stocks pictured in the charts. Recommended Reading: Seagate Q4: Price per Exabyte to Double While Cost Per TB Falls SiTime Q2: Blowout Growth from Timing; Rising Density Applied Optoelectronics Q2: Demand is 20% to 40% Above an Already-Aggressive Ramp SanDisk FQ4: Inference to Drive Demand and Visibility for Years Positions Report – July 2026