Author: io-fund

Seagate’s earnings report ticked a lot of boxes with accelerating revenue growth of 49% YoY and 17% QoQ, margins that are expanding with a significant step-up on GAAP operating margin, and a free cash flow margin that is leading to an important reduction in debt.  On the technology roadmap side, Seagate's Mozaic 3+ platform is now qualified and operating in production environments across all major cloud customers. Mozaic 4+ is ramping with the two largest global CSPs, with additional customer qualifications underway, while Mozaic 5+ qualification shipments are set to begin in late calendar 2027 — positioning the areal-density roadmap to support exabyte demand growth into the back half of the decade.  In the more near-term, management stated that price per exabyte grew approximately 10% in the June quarter, yet an analyst asserted the guidance implies this will reach 20% YoY growth by the September quarter. Meanwhile, the transition to higher-density HAMR products is simultaneously lowering cost per terabyte and expanding margins.  Seagate is Seeing Better Unit Economics from HAMR  Seagate reported price per exabyte growth of about 10% yet an analyst challenged management to provide more visibility into the guide as it implies 20% as soon as next quarter.   Here was the question on the call: “You just reported 10% year-over-year price per exabyte growth in June. I think the September quarter guide implies pricing growth closer to maybe 20% year-over-year or even above that. Can you maybe just provide an update for us on how we should be thinking about pricing looking forward? Why this trend we're seeing in the September quarter shouldn't sustain or maybe even accelerate, just given supply-demand imbalance, customer demand strength, delivering more value to customers, et cetera.”  Management did not confirm the exact percentage, but the response was favorable, with the CFO expecting “every quarter revenue to improve and every quarter gross margin and profitability in general to increase. Of course, pricing is a part of this sequential improvement through the fiscal year.”  Meanwhile, the shift to 3TB to 4TB is helping to improve margins with the CFO stating: “Of course, moving the mix from 3 TB to 4 TB per disk is, of course, giving us another boost in terms of profitability” – referring to the increase in storage that Seagate can sell from each drive, as the disc count can carry about 1/3 more capacity. According to an analyst on the call, the reduction is in the mid-teens: “Two, how do you think about the cost down execution as we move through Mozaic 3+ to Mozaic 4+, you've been operating at a mid-teens kind of cost down per year on a per terabyte basis.”  When combining both sides of the equations with increasing capacity while driving down costs, it ultimately results in more revenue per drive and lower cost per terabyte (TB). HAMR is also increasing in product mix with Seagate expecting 50% of its HAMR exabytes to come from Moziac 4+ by end of calendar year 2026.  To further support better unit economics, the company is driving more exabyte growth with 218 exabytes shipped in Q4, up 34% YoY and of this, 195 exabytes were from data centers. Meanwhile, it was pointed out on the call that drive-unit output was essentially flat, with the growth instead coming from more heads, disks per drive and more capacity per disk: “Now, for example, if you look our last year, and if you look at the number of disk and the number of heads inside the box, they probably grew between 15% and 20%, and the units were absolutely flat” and it was also stated: “These investments enable us to maintain relatively stable drive unit output as customers mix up to higher capacity drives and manufacturing cycle time increase.”  The strategy described in the discussion on drive-unit output is centered on the increasing areal density rather than expanding hard-drive unit capacity. Management discussed that increasing the amount of data stored on every disk is the fastest and most capital-efficient path to achieve exabyte growth while maintaining stable unit output.  We discussed more on heat-assisted magnetic recording (HAMR) in our write-up on Seagate in March of 2026 stating: “Seagate has developed heat-assisted magnetic recording (HAMR) tech for substantial areal density gains, which refers to how many bits can be packed onto each square inch of disk platter (where data is stored). HAMR uses a laser diode to heat a small spot on the disk, enabling polarity of a single bit to be flipped to allow data to be written.”  We also discussed on our Discovery tier additional information on peer Western Digital, stating: “The first is to increase areal density from 32TB to eventually 100TB as we end the decade. By packing more capacity into the same footprint, Western Digital delivers improved economics to alleviate surging capex. The company’s UltraSMR-enabled JBOD platforms offer TB per drive, lower cost per TB and lower power (and space) per TB, delivering not only increased capacity but also lower total cost of ownership.”  Financials  Revenue Rises 48% YoY to a Record $3.63 Billion   Seagate reported FQ4 revenue of a record $3.63 billion, coming in above the upper end of guidance for $3.55 billion. YoY growth accelerated slightly more than 4 points to 48.5% in the quarter, capping off FY26 with a 27 point acceleration since FQ1’s 21.3% growth.   QoQ growth accelerated 6.4 points to 16.6% QoQ, marking Seagate’s fastest sequential growth print since 2012, as cloud and enterprise data center demand continues to outstrip available nearline supply.  For fiscal Q1 2027, Seagate guided for revenue of $4.1 billion, +/- $100 million, a level that sits well above the pre-print estimate of $3.75 billion for the quarter. Q1’s guide implies YoY growth accelerating further to approximately 56%, while sequential growth would moderate slightly to 13% QoQ at the midpoint.   Data Center Accelerates to 17.3% QoQ, Pricing Power More Evident:  Seagate’s Data Center revenue was $2.93 billion, up 57.4% YoY and 17.3% QoQ. Growth accelerated slightly from 54.8% YoY in Q3, while QoQ stepped up from 12.4% in Q3, a nearly five point acceleration.   The pace of growth on a YoY and QoQ basis for revenue versus exabyte shipments suggests that Seagate is capturing more pricing power this quarter. For comparison, nearline exabyte shipments rose 11.4% QoQ and decelerated slightly to 42.3%, creating a 15-point delta for YoY revenue growth over exabyte growth. This expanded from a roughly 9-point delta for YoY revenue and exabyte growth in Q3, at 54.8% versus 45.8%, implying pricing tailwinds likely strengthened during the quarter.   Management noted that nearline exabyte supply is now largely allocated into calendar 2028, with customers extending their planning horizons into calendar 2029 and beyond, a strong signal of forward visibility into the demand backlog. Cloud remains the primary nearline demand driver, now with three full years of sequential quarterly exabyte growth, while Enterprise/OEM data center customers also posted strong double-digit YoY revenue and exabyte growth, per Seagate, pointing to a broadening customer base beyond the largest hyperscalers.  Edge IoT revenue was $697 million, up 20% YoY and 13.9% QoQ, a notable acceleration from 1.8% QoQ in FQ3. However, Edge IoT represents 19% of total revenue, down from 24% a year ago, as Data Center growth continues to outpace the legacy business. This is reflected in non-nearline exabyte shipments, which declined (11.5%) YoY and (4.2%) QoQ.    Rough back of napkin math for FQ1’s guide suggests Data Center growth could moderate slightly – assuming a slight increase in mix to 81.5% of total revenue, Data Center would project to $3.34 billion, maintaining 58% YoY growth while decelerating slightly to 13.9% QoQ. This would project roughly $759 million in Edge IoT revenue, also marking a deceleration to 8.8% QoQ though YoY growth would accelerate 27 points to 47.3% YoY against a soft comp.   Margins Expand to Record Levels on Mix and Pricing  One of the key highlights of the report was margins expanding to record levels, with operating and net margins showing strong YoY and QoQ expansion as Seagate saw opex decline sequentially.   GAAP gross margin reached a company record of 52.3%, up 14.9 points YoY and 5.8 points QoQ.   GAAP operating margin also hit a record 43%, up 19.8 points YoY and 10.9 points QoQ. Adjusted operating margin was 44.6%, up 18.4 points YoY and 7.1 points QoQ. The magnitude of sequential expansion points to continued favorable mix shift toward higher-value nearline capacity alongside disciplined cost control, as opex actually declined (2%) QoQ even as revenue grew 17%. Seagate guided for FQ1 2027 adjusted operating margin of 50% at midpoint, implying further sequential expansion of 5.4 points from Q4, an aggressive guide that underscores management's confidence that favorable pricing and mix will persist into fiscal 2027.  GAAP net margin was 35.7%, up 15.7 points YoY and 13.7 points QoQ. Adjusted net margin was 36.3%, up 13.6 points YoY and 6.3 points QoQ.  For fiscal 2026, GAAP gross margin reached 45.6%, up 10.4 points YoY. GAAP operating margin hit 33.6%, up 12.8 points YoY, while adjusted operating margin was 36.5%, up 13.1 points, offering a clear demonstration of operating leverage as revenue grew 34% YoY while operating expenses rose less than 12% for the year.   Adj EPS More Than Doubles YoY to a Record $5.71  GAAP EPS rose 149% YoY to a record $5.58, accelerating from 108% growth in Q3 and 68% in Q2. Adjusted EPS was a record $5.71, up 121% YoY and accelerating from 115% growth in Q3 and 53% growth in Q2; this also handily beat estimates for $5.09 by more than 12%.    For FQ1 2027, Seagate guided for adjusted EPS of $7.30 +/- $0.20, representing a further acceleration to 179% YoY growth. This also came in meaningfully above the $5.85 estimate for the quarter heading in to the report.   For fiscal 2026, GAAP EPS more than doubled to $13.90, up 105% YoY, while adjusted EPS was $15.58, up 92% YoY.   Free Cash Flow Rises More than 2.5X in Q4 as Balance Sheet Strengthens  Cash flow generation was robust in the quarter, but perhaps more important is the progress Seagate has made in strengthening its balance sheet, with net leverage substantially improving throughout the year as debt has been paid off.   Operating cash flow increased 157% YoY to $1.3 billion for a 36.0% margin, up 15.2 points YoY and roughly flat QoQ. For FY26, operating cash flow more than tripled to $3.67 billion for a 30.1% margin, expanding substantially from 11.9% in FY25.  Free cash flow rose more than 2.5X in Q4 to $1.1 billion for a 30.8% margin, up 13.4 points YoY and roughly flat QoQ. For FY26, free cash flow rose nearly 4X to $3.1 billion for a 25.5% margin, a significant improvement from 9% in FY25.   Seagate continued to strengthen its balance sheet, ending the quarter with $1.7 billion in cash and $3.6 billion in debt, for net debt of $1.9 billion. In Q4, Seagate reduced its gross debt by $302 million and $1.4 billion for the full year, bringing its net leverage down to 0.4X from 1.8X a year ago. Subsequent to quarter-end, Seagate also extinguished $1 billion of high-yield senior notes in July 2026 and plans to retire the remaining balance on its convertible notes in September 2026, continuing its deleveraging trajectory.   Conclusion:  In addition to reporting strong across nearly every financial metric, Seagate’s future growth is being supported by better underlying unit economics rather than only a rebound in hard-drive demand.  The takeaway is that exabytes are growing faster than physical drive volumes, and the pricing is favorable. HAMR is increasing areal density for higher revenue while resulting in improved cost per terabyte.   Although lesser-known Seagate was in SK Hynix’s shadow yesterday evening, this earnings report has captured our attention. As AI investors brace for more Big Tech capex surprises today and tomorrow, this report proves that lesser-known suppliers are quietly crushing it.  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 STX at the time of writing. Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis. Recommended Reading: Marvell: Strong Interconnect Growth, but Really an FY28 Story Corning: Glass Manufacturing Powerhouse Pivoting Hard into AI Networking Credo: Reliability Leader Aggressively Moves into Optics Macom: Data Center Revenue Accelerating to 35% QoQ in FQ3

Meta’s Q2 2026 did not pass baseline criteria for a Big Tech company with revenue guidance for Q3 missing analyst estimates, an operating margin contraction and the company is on the razor’s edge of becoming free cash flow negative. Despite revenue growing 28% YoY in the current quarter, Meta’s margins and cash came up short.  Although selling compute is a noble effort to absorb some of the capex weakness, it also points to Meta’s core business not being able to sustain the company’s AI spend on its own. Meta framed this as a sign that AI capacity is tight, stating “we're getting a lot of offers for compute at a significant premium over what we paid for it." However, something doesn’t quite add up as Meta is selling compute yet also buying from third parties.   Given the string of strong earnings reports we have already seen this quarter, including from a lesser-known memory stock we covered on the Discovery tier today, this isn’t a report from a company we want to own right now. Overall, the tide is beginning to turn with Big Tech providing the weakest reports in our universe right now – although we saw this coming, it’s also something the market will have to get used to.  Compute Constrained or Excess Compute – which one is it? Earlier this month, Bloomberg and others reported that Meta was preparing to sell access to its AI computing power and models, similar to AWS, Google Cloud and Azure. The idea is that Meta will sell excess compute with Bloomberg stating: “the company is also considering selling access to “raw” computing capacity, akin to other so-called neocloud businesses like CoreWeave.”   The new business venture will be called Meta Compute, and will offer excess computing infrastructure and also an API service where customers pay for AI usage, with more details in the commentary this evening: “Two additional revenue streams are subscriptions and monetizing our competitive models through an API. Meta One is an evolution of our subscription portfolio to create more value for everyday users, businesses, and creators, so they get more features and AI tools to create, connect, and stand out […] We recently made Muse Spark available on OpenRouter for U.S.-based developers, broadening its distribution and making it easier for developers to adopt the model.”  However, selling compute runs directly counter to other statements the company has made. Last quarter, it was stated the company was signing multi-year cloud deals for 2026-2027, and alongside infrastructure purchase orders, contributed to a $107 billion step-up in contractual commitments in just one quarter alone.  An analyst called this out on the call, asking “Mark, just in terms of the number of offers to monetize your compute externally, you also at the same time are purchasing capacity from a number of third parties. Just hoping you can help us understand some of the differences here. Is it just timing and stopgap issues, or is it training versus inference and leveraging the chips that are best suited for each? Thanks.”  Because the response is a head-scratcher, I’m quoting it at length below.   Mark Zuckerberg CEO  "I can answer the second part of that. I think your question was about how we think about offers that we're getting to sell the compute, we're also buying the compute. Look, the high-level observation is that there's just nowhere near enough compute for all the demand. That is why we see that basically, we are getting a large number of offers for the compute that we have, we have a lot of internal uses that we think are going to be quite valuable. Now, in terms of running the business, obviously, a common trade-off that we need to make is around how much do you monetize something today versus develop future assets for the future? I think that it's always a portfolio, right?  You don't want to only do long-term things and not kind of prove the markets out that exist in the near term. I also think it would be foolish to basically just sell all of the compute and take a short-term profit. When you have the opportunity to build intelligence on top of it, which will be a kind of multiple and that compounds the value of the compute on top of that.  I think the answer is what we're doing, which is to basically use a lot of our capital to build out compute, having confidence that we have the ability to monetize the compute directly when that makes sense, but also knowing that we have quite a number of different use cases to monetize the intelligence on top of the compute, including the enterprise cases that we talked about and including some of the consumer and cases that we talked about, and including just the core business, which is not even necessarily new products that we haven't talked about, but just in terms of using that to be able to further add intelligence and improve the ranking and recommendations and ads in the core services." Unfortunately that response is not able to explain why a company is spending $137.5B on capex, but missing revenue estimates next quarter, while renting cloud capacity, and simultaneously offering excess cloud capacity. Financials  Revenue Growth Decelerates to 28%, Guidance Points to Further Cooling in Q3  Meta reported revenue of $60.80 billion in Q2, as growth decelerated from 33.1% YoY in Q1 to 28% YoY this quarter. Sequentially, revenue rose 8.0% QoQ following Q1's seasonal (6.0%) QoQ decline.   Despite a slight 0.8% beat to estimates in Q2, Meta’s guide for Q3 was quite underwhelming, not only missing estimates by a larger margin but also pointing to a further deceleration on both a YoY and QoQ basis. For Q3, Meta guided to revenue of $61–64 billion, missing consensus estimates for $63.24 billion by 1.2% at the $62.5 billion midpoint, and pointing to a 6 point deceleration to 22% YoY and 2.8% QoQ.  Advertising Growth Decelerates, Ad Key Metrics Mixed  Advertising revenue was $59.36 billion, up 27.5% YoY (26.0% on a constant currency basis) and 7.9% QoQ, decelerating from 32.9% YoY growth in Q1. Family of Apps revenue, which includes advertising plus the smaller "other revenue" line (payments, business messaging, etc.), was $60.37 billion, up 28.0% YoY and 8.0% QoQ.  Below the surface, advertising key metrics were mixed: ARPP reached a fresh record high, though growth decelerated, while ad impressions growth also decelerated. This suggests that Meta could be hitting a ceiling in how much they can squeeze out of their ad platform via AI optimizations in the near term.   ARPP rose 22.9% YoY and 7.1% QoQ to a record $16.77 (above Q4 25’s seasonally-strong $16.56), though growth decelerating from 26.7% in Q1. Ad impressions decelerated five points to 14% YoY, while ad pricing remained stable at 12% YoY. The ARPP strength and ad pricing stability likely lends itself to a six point acceleration in pricing growth in Meta’s highest-monetizing market, US & Canada, to 20%.  Outside of advertising, Reality Labs remains deep in the red – revenue rose 16.5% YoY and 7.2% QoQ to $431 million, yet operating losses for the segment widened to ($4.62 billion) or a (1072%) margin.  Margins Crunch in Q2, Underlying Cost Growth Also Accelerating  One of the larger red flags within Meta’s report stemmed from the cost side, as opex growth outpaced revenue growth at >2:1 in the quarter, driving operating income down (8%) YoY and operating margin down 10 points sequentially. While Meta tried to chalk up the YoY decrease in operating income to increased legal and severance expenses, the fact of the matter is that even when stripping that out, operating margin would still be declining both YoY and QoQ.  GAAP gross margin felt a slight pinch in Q2, coming in at 81.4% versus 82.1% a year ago and 81.9% in Q1.   GAAP operating margin was 30.9%, down 12.1 points YoY and 9.7 points QoQ, as opex rose 65% YoY versus 28% growth for revenue. This was primarily driven by a 67% increase in R&D to $21.7 billion. Even when stripping out the ~$3.6 billion in legal and severance charges, which Meta said would have led to a 9% YoY increase for operating income, operating margin would come in at 36.8%, still down roughly 6.2 points YoY and 3.8 points QoQ. This suggests that the elevated R&D and infrastructure-related cost growth could be creating margin pressure independent of the one-time items.  Despite this, management reiterated that it expects 2026 operating income to come in above 2025's $83.28 billion. With H1 2026 operating income already at $41.65 billion, that guidance implies H2 operating income of at least $41.6 billion, essentially flat versus 1H and indicating that there may not be a sharp V-shaped recovery for operating margin.  GAAP net margin was 26.1%, following operating margin with a 12.5 points YoY and 21.4 point QoQ decline.   EPS Declines for the Second Time in Nine Quarters  GAAP diluted EPS was $6.18, down (13.4%) YoY from $7.14 and marking the second YoY EPS decline since Q1 2024 (the other being Q3 2025's litigation-charge-driven drop to $1.05). Notably, this also was a substantial (14.4%) or ($1.04) miss to estimates for $7.22  For Q3, estimates pre-print sat at $7.36, though the margin pressure arising in Q2 could shift the outlook for next quarter should it persist.   Capex Ramp Nearly Wipes Out FCF, Debt Load Jumps 42% Sequentially  Outside of the margin headwinds, another glaring flag for Meta’s Q2 report comes down to capex and FCF, as the company nearly followed Big Tech peer Alphabet in going FCF negative – and likely will do so next quarter.   Operating cash flow was $31.86 billion for a 52.4% margin, down 1.4 points YoY and 4.8 points QoQ.  Free cash flow collapsed (90.8%) YoY to just $784 million for a margin of only 1.3%, down from 18% a year ago and 22% in Q1. This was driven by a surge in capex to $31.08 billion, up 82.7% YoY and accounting for 97.5% of OCF, up from 66.6% a year ago.    With Meta tightening its capex guide to $137.5 billion at midpoint ($130-145 billion), up from $135 billion previously, there is a high likelihood that capex heads negative in Q3. At the midpoint, 2H capex is projected to be $86.6 billion, more than 70% higher than what Meta spent in 1H and implying an average spend of $43.3 billion split across both quarters.   On the balance sheet, cash, equivalents, and marketable securities rose to $90.26 billion, aided by $24.91 billion of net proceeds from a long-term debt issuance in the quarter. Long-term debt jumped to $83.66 billion from $58.75 billion in Q1, a 42.4% QoQ increase and nearly 3x the year-ago balance of $28.83 billion. The net cash position narrowed sharply to $6.6 billion from $22.4 billion in Q1, as Meta increasingly leans on debt issuance to fund its infrastructure build-out.  Similar to Alphabet, Meta’s off-balance sheet liabilities are surging – while the 10-Q is needed to confirm Q2’s numbers, Meta’s lease obligations for leases not yet commenced and contractual commitments totaled $420.6 billion in Q1, up from $234.8 billion in Q4 and $67 billion at the end of 2024. When including operating lease liabilities listed on the balance sheet, Meta’s obligations reached $448.6 billion in Q1, up from $260 billion in Q4 and $87.2 billion at the end of 2024.  Conclusion:  Big Tech management teams are beginning to contradict themselves regarding the scarcity in compute in ways that are becoming difficult to reconcile. Meanwhile, Meta escalated capex from $20B last quarter to $31B this quarter, with a clear indication there will be more increases to come with a fiscal year guide of $130B to $145B. To speculate on Meta Compute is a tough ask, especially given the other incumbents aren’t able to offset their capex costs yet, such as Google, Amazon or Microsoft. Meta’s mixed-messaging around leasing third-party cloud capacity while simultaneously launching Meta Compute to sell excess compute and API usage leads me to the same conclusion as I wrote earlier this week following Google’s ballooning purchase commitments, which is that Big Tech is using its formerly good reputation to squeak by despite capital intensity and circular deals that could pose a future liquidity risk. Our portfolio will remain concentrated in the clear beneficiaries that are downwind from the rising capex, and who win regardless of which balance sheet ultimately carries the compute costs. As for Big Tech, these leasing structures, balance sheets and cash flow profiles are too messy to meet our criteria. 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 META 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: The I/O Fund’s Top 20 Stocks for Q3 2026 Bloom Q2: Backlog is Growing Faster than Revenue at 166% YoY Growth Google Earnings Q2: Free Cash Flow in the Red; Purchase Commitments Balloon MaxLinear Beats on All Fronts with Keystone; Rushmore Ramping in 2027

Bloom Energy reported revenue of $1.065 billion, beating revenue estimates by 29% with analysts expecting revenue of $827 million going into the print. This represents growth of 166% YoY, yet Bloom stated their backlog is growing even faster.   The company has not updated backlog since the Q4 call when it was stated they had a $20 billion backlog with $6 billion from product and $14 billion from services. However, the update this evening was that “[…] customers are now placing longer-term orders, leading to our backlog growing at a faster pace than revenue. Let me repeat, leading to our backlog growing at a faster pace than revenue.”  Within the management commentary it was also stated that “all the major US hyperscalers and over a dozen US neoclouds, AI labs, and colocation data center operators have validated and approved our power solutions for their AI factories.” This is a significant statement as it was only last quarter that Bloom began to first discuss hyperscalers as customers, given Oracle, Brookfield and AEP were the only disclosed large customers. At the time, last quarter, it was stated “more than half of our current data center backlog comes from other hyperscalers, neo clouds and colocation providers.”  The company raised full year guidance to $4.05B at the midpoint, up from $3.6B. Adjusted operating income was also raised to $850 million at the midpoint, and adjusted EPS was raised to $2.70 at the midpoint compared to analyst estimates calling for $2.17 for FY26.  On the call, Bloom discussed the rather dire reality that hyperscalers plan to build 30GW to 40GWs next year, yet encouraged analysts to “take a guess on how much of those projects will get delayed because the power provider is not able to provide power on time.”   If projects do face delays, this will help intensify the need for Bloom’s on-site power solutions. Notably, the evidence is already visible in the expanding backlog with “all the major” hyperscalers and neoclouds rushing to validate and approve Bloom’s solid oxide fuel cells.  “4-Year Backlog is not a Trophy”  Bloom’s management does an exceptional job of describing their value proposition in short, pithy descriptions. To describe Bloom’s positioning in no uncertain terms, the following was shared tonight: “Chips without power are inventory, not intelligence. Grid operators quote years, timelines on which billion-dollar compute clusters go obsolete in a warehouse. Legacy suppliers celebrate backlogs stretching to 2029 and beyond. We think a four-year backlog is not a trophy.”  Management is referring to the four-year backlogs that many large-scale utilities and nuclear power companies provide on their earnings calls. The statement is challenging the other power solutions that lead with multi-year backlogs. Meanwhile, customers are in a queue waiting with compute they can’t power.   Management further quantified the opportunity cost of waiting for 4-year backlogs to materialize in the earnings call – putting real numbers to the difference between Bloom and its competitors:  “A full stack AI provider that is responsible for everything in the data center financially, the 1 gigawatt data center in one single year, depending on the nature of the AI customer, will deliver between $12 billion-$24 billion in revenue per year. You pull in power for them within a month, which is the tall pole, that is $1 billion-$2 billion of revenue that they would not have had, on a 40%-50% gross margin and a 20%-25% net margin.”  To contrast with the multi-year backlogs, management pointed out that even with a growing backlog, they can deliver system to a new customer within the year: “We have booked major customers as of today who were not in our reported backlog at the end of last year, for whom we will ship some of our systems this year.  Brookfield 5X’s Partnership, Becoming a Key Enabler for Bloom to Quickly Scale  The expansion of the Brookfield partnership last month is more important than this quarter’s results as it lays a path for Bloom’s scale to multiply by removing a critical financing bottleneck.   Given many data center customers purchase power or capacity under long-term agreements rather than own the generation equipment directly. Under the Bloom-Brookfield partnership, the financing partner (Brookfield) owns the energy servers, and the end customers pays over time through “a power purchase agreement priced per kilowatt-hour, capacity agreement or equipment lease priced on installed capacity.” The goal is for the customer to not own the asset, but rather pay over time to avoid a large capex sale.   Brookfield’s increased confidence of providing project capital grew 5-fold over the past nine months from $5 billion to now offering commitments of up to $25 billion. According to management commentary, more financing partners committed capital this past quarter, and the pool of infrastructure investors is expected to grow:  “One of the largest and most experienced infrastructure investors in the world evaluated our technology, our delivery record, and our pipeline, backed us with $5 billion, watched us execute, and then multiplied that backing by 500%. Capital of that quality and quantity does not follow letters of intent, MOUs, or press releases. It follows performance, happy customers, and firm bankable orders. Brookfield is not alone. This quarter, Industrial Development Funding, who has previously funded Bloom deployments, partnered with Oaktree, MUFG Bank, and Morgan Stanley to fund Bloom deployments, cumulatively bringing their total commitment to $2.6 billion. More financing partners are in the wings. Gigawatt demand needs gigadollars of capital.”  Distributed Inference Will Become Bloom’s Next Growth Driver  We’ve gone over the training market versus inference market from many different angles as it pertains to investment opportunities, such as this Memory overview. However, Bloom has an important physical advantage in powering data centers for inference, even more so than training. While training is concentrated in massive data centers, inference will need to move to closer to end users, such as cities and local data centers.   As Bloom stated on the call, its modular, onsite fuel cells are an attractive option because gas turbines cannot be placed in dense urban areas such as Manhattan. As lower costs per token drive significantly higher token consumption, more inference capacity will be needed in metropolitan areas where air pollution and water consumption are top of mind when evaluating competing power solutions. This sets up an important Stage 2 catalyst for Bloom.  “Now, as inference comes along, if the transmission distribution infrastructure in the country is having difficulty doing transmission, building highways, imagine how difficult it'll be for them to upgrade distribution, which is surface streets. That's where inference power is going to be needed. Bloom is ideally suited for that. You cannot put a gas turbine in the middle of Manhattan.”  Financials Revenue Surpasses $1 Billion for the First Time, 29% Beat to Estimates Bloom Energy's Q2 revenue accelerated sharply to $1.065 billion, up 165.5% YoY and 41.8% QoQ, crossing the $1 billion mark for the first time and marking a notable 28.8% beat versus consensus estimates for $827 million. YoY growth reaccelerated by more than 35 points from 130.4% in Q1, marking Bloom's strongest print since going public. QoQ growth swung from a (3.4%) sequential decline in Q1 to a 41.8% increase, reflecting the lumpy but rapidly compounding nature of Bloom's project-based revenue recognition.  Management noted that all major U.S. hyperscalers and more than a dozen neoclouds, AI labs, and colocation operators have now validated Bloom's fuel cell platform for AI data center power, supporting the durability of demand across the AI landscape rather than being tied to a  single large one-off order.  For the full year, Bloom boosted its revenue guidance once again, now projecting ~100% growth to $3.9-$4.2 billion, a raise from its prior guidance for 80% growth and 40 points above its original FY26 guide for 60% growth.   Segment Breakdown: Product Accelerates to 215% YoY Growth, Electricity the Lone Laggard  Q2 growth was overwhelmingly driven by the Product segment, which saw revenue increase 215.4% YoY and 43.2% QoQ to $935.4 million, now representing 88% of total revenue. To put in perspective how quickly Bloom has ramped over the past four quarters, Product revenue was $296.6 million in Q2 2025, meaning Q2 2026’s sequential growth of $282.1 million is nearly equivalent to its entirely quarterly scale last year.   Install revenue was $51.0 million, up 36.4% YoY and 96.6% QoQ, following Product with a sharp QoQ acceleration. However, Install remains structurally low/negative-margin as Bloom continues shifting toward a consult-only installation model for large-load sites, pushing more installation work (and margin) onto third parties and channel partners.   Service revenue grew 26.8% YoY and 11.5% QoQ to $69.0 million, benefiting from a larger installed base of fielded systems as growth accelerated from 16% YoY in Q1.   Electricity revenue, the smallest and most volatile segment, declined (22.3%) YoY and was up marginally QoQ to $10.0 million, though it remains an immaterial share of the mix (1% of total revenue).   Margins Expand Sharply as Operating Leverage Kicks In  Operating leverage is becoming more visible as Bloom’s revenue scales, with GAAP operating margin showing a sharp YoY expansion to the high-teens, while GAAP net margin expanded nearly 30 points YoY. Put another way, Bloom’s profitability picture is quickly improving as Product (and total) revenue ramps higher.   GAAP gross margin expanded 6.7 points YoY and 3.4 points QoQ to 33.4%, while adjusted gross margin expanded 6 points YoY and 2.8 points QoQ to 34.3%, both accelerating from the roughly 3-4 points of YoY expansion seen in the prior two quarters. Product gross margin was 36.5% on a GAAP and 37.2% on an adjusted basis in Q2, well ahead of corporate average, and should continue to be the primary lever for further margin expansion as revenue scales.   Operating leverage was a core theme evident in Q2’s report: GAAP operating margin was 17.1%, expanding 18 points from a (0.9%) operating loss margin a year ago, while adjusted operating margin reached 22.5%, up 15.4 points YoY and 5.2 points QoQ.  Adjusted operating expenses grew just 21.7% YoY versus 165.5% revenue growth, illustrating the scale benefits now appearing as Bloom laps its cost base against a larger, faster-growing top line.   Reflecting this leverage, Bloom raised full-year adjusted operating income guidance to $800-900 million (or 4X YoY growth from $221 million in FY25), implying a 21% margin, up more than 10 points YoY.  GAAP net margin was 18.4%, up 29 points YoY and 9 points QoQ, while adjusted net margin was 23.3%, up 17.8 points YoY and 4.9 points QoQ. Adjusted EBITDA was $253.4 million, up roughly 6.1x YoY and representing a 23.8% margin, improving from a 10.3% margin a year ago and 19% in Q1.  GAAP EPS Beats by Triple-Digits, Adjusted EPS Growth Nearly 8x  Driven by the revenue beat, Bloom delivered significant beats to consensus estimates on the bottom line on both a GAAP and adjusted basis – GAAP EPS beat estimates by a wild 148%, while adjusted EPS beat by 90.2%.   GAAP diluted EPS was $0.62, swinging from a ($0.18) loss per share a year ago and beating estimates for $0.25. This was aided by both the operating margin expansion and $14.1 million of non-operating income (versus a $39.1 million non-operating expense in the prior-year period. For Q3, GAAP EPS estimates sat at $0.37 prior to Q2’s report, but the magnitude of the beat combined with Bloom’s margin strength suggest this will move higher.   Adjusted EPS was $0.78, up 680% from $0.10 a year ago and beating estimates for $0.41. Similar to GAAP EPS, estimates for Q3 sat at $0.52 for 248% growth prior to the report, and are likely to move higher in the coming days and weeks as updated FY guidance gets accounted for.   For the full year, Bloom raised adjusted EPS guidance to $2.55-2.85 from its prior view for $1.85-$2.25, implying further sequential acceleration in the back half as the operating leverage story continues to play out. Reaching the midpoint of guidance would require $1.48 in adjusted EPS in 2H.   Operating Cash Flow Positive for Fourth Consecutive Quarter  While Bloom had seen some lumpiness with cash flows through Q1 2024 to Q2 2025 with only one quarter seeing positive operating cash flow, that story has shifted in early 2026 with Q2 seeing Bloom report its fourth consecutive quarter of positive OCF.   Operating cash flow was $226.4 million for a 21.3% margin, a swing of more than $439 million from ($213.1 million) a year ago for a (53%) margin. OCF also tripled sequentially from $73.6 million in Q1. Free cash flow was $174.8 million for a 16.4% margin, a sharp improvement from ($220.4 million) a year ago for a (55%) margin.   Cash and restricted cash totaled $2.69 billion while debt totaled $2.48 billion, leaving Bloom in a modest net cash position for the first time in recent memory.   Conclusion  Bloom surpassed the $1 billion quarterly revenue milestone in fashion, with revenue up 41.8% QoQ and beating estimates by nearly 29%. Q2’s report was similarly impressive down the line with operating leverage becoming more visible as Bloom scales, with GAAP operating margin up 18 points YoY to 17.1% as revenue growth far outpaced opex growth. GAAP EPS beat by triple digits while management provided a substantial upward revision to adjusted EPS for the full-year, implying margin strength and earnings power will carry on through year end.  Given we’ve tracked Bloom very closely for about six quarters now, there were not any major reveals outside of the financial strength – and that is exactly the way we like it because it translates to the management executing as promised. As I stated in my Top 20 report: “The importance of Bloom Energy’s time-to-power thesis has only intensified over the past quarter (and I wonder how long I will be saying this to you).”    That is a reference to how consistent this company has been in delivering strong earnings reports, laying out an expanding customer list and a growing backlog – but not the same kind of multi-year backlogs we will hear in the many earnings reports to follow this quarter. There is something special about this company, and it is evident in every report I've covered going back to Q4 2024. 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 BE at the time of writing and may own stocks pictured in the charts. Recommended Reading: The I/O Fund’s Top 20 Stocks for Q3 2026 Google Earnings Q2: Free Cash Flow in the Red; Purchase Commitments Balloon MaxLinear Beats on All Fronts with Keystone; Rushmore Ramping in 2027 GE Vernova Q2: Laying a Strong Foundation for Years to Come

Alphabet delivered a strong quarter with revenue rising 24% and Google Cloud accelerating 82% to $24.8 billion. However, quarterly capex ballooned to $44.9 billion, which exceeds operating cash flow of $39.1 billion, which pushed free cash flow into the red at ($5.9 billion).  Last week, in an article highlighting the risk that Big Tech’s free cash flow turn negative, we stated: “Across Google, Microsoft, Meta and Amazon, capex is rising much faster than operating cash flow as Big Tech races to build out AI infrastructure. While this is well understood, what may come as a surprise to investors is these highly liquid companies with strong earnings could soon, one-by-one, turn free cash flow negative as we move into 2027.”  Where things intensified is that Google reported negative FCF yet analyst were not modeling for this to happen until around 2028. You can see below that Google was expected to report FCF of $22.7 billion this year, yet raised capex from $185B at the midpoint to $200B at the midpoint, which quickly eroded the FCF margin.  Source: Big Tech’s Free Cash Flow is Turning Negative – Who's Next?  Google turning FCF negative has poured fuel onto the fire for the bearish narrative on AI. However, to remain balanced, Google is moving TPUs beyond its own cloud and there is a material (nearly historic) re-acceleration in the company’s growth rates.   Google’s Merchant TPU Sales are a Significant Expansion Opportunity  This quarter, Google confirmed it’s delivering TPUs directly to data centers, bringing its highly competitive silicon program external for the first time. Merchant TPUs were delivered to customer data centers in the second quarter, but will take time to ramp.   As we had covered in our Q3 Top 20 Stocks report, seeing Google compete head-on with Nvidia is a major turning point for the AI supply chain. In my view, as hyperscalers deploy more of their own accelerators, it will not only disrupt Nvidia’s GPUs but also Nvidia’s proprietary stack, ultimately opening up more demand for third-party components that connect heterogeneous AI systems.   The earnings call indicated that 2027 would be the standout year for this:  “We start building inventory to be able to sell those systems. So you see that impact on the cash from operations because we built ahead obviously, as we’re building that business and ramping up. And then once we start delivering the sales, generally, that’s when we start recognizing revenue. This quarter was a small amount of the total—that total agreement will continue to ramp up throughout 2026, but then you’ll see the vast majority of the revenue from that agreement comes through in 2027.”  Google Cloud Crushes on Growth and Margins; But It’s Complicated  Google Cloud offered strong headline numbers with revenue increasing 82% to $24.8 billion and cloud operating income tripling to $8.8 billion with operating margin expanding nearly 15 points to 35.7%. Backlog rose more than $50 billion QoQ to $514 billion with about half converting to revenue during the next 24 months.  While this would typically be reason for celebration, the Google Cloud segment is complicated as cloud margins are expected to be pressured as management discussed using third-party cloud providers to bridge demand.   “And given the supply constrained environment, we plan to expand the use of third-party capacity in Q3 as a bridging strategy while we build up more internal capacity. This strategy allows us to keep growing our customer base and capture greater overall value. However, it will create modest margin pressure in the near term as we utilize this capacity.”  Management framed this as a short-term hit for long-term leverage:   “Sundar Pichai 
CEO & Director  […] I think on the bridge deal, the main thing I would say is, look, there are, on the margin, very, very large customers of ours on cloud, who we are trying to support them through this extraordinary moment. And the incremental opportunities they are bringing to us while a short-term cost over a few months may be very high in the lifetime of the deal as we bring more capacity on, is highly ROI positive, right? So those are factors we are taking into account. So are you willing to take upfront 6-month deal to be able to serve that customer in what is a multiyear opportunity where the margins and the returns are very, very attractive over that multiyear horizon? So hopefully, that gives some color on how we thought about those opportunities.”  However, if one takes a step back, it does feel unhinged from reality that Google spent $45 billion on capex this quarter for $25 billion in revenue (so expanding a business that runs at a deficit), yet lacks enough internal capacity and now needs to pursue capacity through a third-party (while selling their internal silicon externally now). That one is hard to reconcile.   To recap, it’s this broader picture that goes beyond only free cash flow being negative, to also include external capacity commitments as a bridge, while also selling Merchant TPUs externally that creates a very complicated picture.  We wrote a deep dive on this issue in the article: “Nvidia, CoreWeave and Nebius: Inside the Circular Financing of the GPU Boom” where we stated:  “By leasing compute capacity from neoclouds, hyperscalers shift their cost timeline from being a large upfront capex outflow to an operational expense outflow spread over long-term contracts. The need to spread costs is becoming increasingly evident due to the massive spending hyperscalers are engaged in.    Although this is the “bear” case on why hyperscalers work with neoclouds—contrasting this with the rationale behind GPU access and utilization is key because one could argue that hyperscalers are quite capable of software optimizations and GPU utilization on their own (in fact, they are the longstanding incumbent here with deep expertise in cloud operations and workload optimizations).”  The article is worth a read for more information on why balance sheets woes are complicated compared to tech booms in the past.  Double-Clicking on Purchase Commitments, up Nearly $500B to $811B  You may be familiar with our discussions on Nvidia’s surging supply-related commitments as a key signal that Nvidia is securing its supply chain to meet its $1 trillion in cumulative revenue forecast, and the same goes for Alphabet with its purchase commitments.   Take a look at the trend below for purchase commitments, which reached $811 billion in Q2, up more than 11X YoY and nearly 1.5X higher QoQ – a nearly $500 billion QoQ rise. Within this, $200.7 billion of the commitments were classified as short-term, up from $138 billion in Q1.   The vast majority of the purchase commitments are primarily tied to long-term supply agreements for infrastructure and inventory components, with these long-term supply agreements generally lasting through 2030, as well as open purchase orders. Long-term take-or-pay energy agreements and software content licenses are also reflected in the commitments, but represent a smaller portion of that $811 billion figure.   Assuming roughly $700 billion of these commitments are tied to infrastructure and inventory (as Alphabet lists the vast majority of $707 billion in fixed or guaranteed commitments as tied to LT supply agreements), it could have north of $170 billion per year earmarked for data center infrastructure. There is a chance that a chunk of the recent increases from Q4 through Q2 are tied to rising memory prices, especially if Alphabet is securing memory supply through the end of the decade for its TPU roadmap. Given that TPU shipments for 2026 and 2027 are estimated to be around 14 million units, and assuming 10 million shipments each year in 2028 through 2030, total HBM commitments (not including DRAM or NAND) could be upwards of $130 billion — even assuming no change in HBM content in future chips and pricing in line with HBM3e’s ~$10/GB, both of which are likely to be higher.   For reference, Nvidia’s supply related commitments totaled $119 billion last quarter, or roughly a $700 billion differential between the two companies. The size of the difference likely boils down to their respective positions in the AI ecosystem – Nvidia is simply building the GPUs and is securing the necessary memory, wafers and other components to support that, while Alphabet has to build its TPUs as well as the physical data center infrastructure and everything else in between.   All told, the sheer pace of this increase towards nearly $1 trillion in purchase commitments suggests that Alphabet’s capex cycle will remain higher for longer, as these purchase commitments convert over to capex and as cash moves out the door.   Financials  Revenue Growth Accelerates to 24% YoY, Sixth Straight Quarter of Acceleration   Alphabet reported Q2 revenue up 24.2% YoY (23% in constant currency) to $119.8 billion, extending an acceleration that has now run for six consecutive quarters from 12% YoY in Q1 2025. This also marked Alphabet's 12th consecutive quarter of double-digit revenue growth. On a QoQ basis, revenue grew 9%, a slightly stronger sequential print than the 7% and 5% seen in the comparable quarters the last two years.   Q3 revenue is currently estimated to be $127 billion, maintaining Q2’s pace at 24% YoY while QoQ growth would moderate to 6%. Consensus estimates also project Q4 to maintain this 24% YoY pace to $140.9 billion, primarily driven by Cloud strength.   Google Cloud Growth Accelerates Sharply to 82% YoY and 24% QoQ  Google Cloud stood out in Q2 as growth accelerated on both a YoY and QoQ basis, a notable feat considering the segment’s run rate has now reached $100 billion, just five quarters after hitting $50 billion. YoY growth accelerated 19 points from Q1 to 82% YoY, or 50 points faster than Q2 2025’s growth of 32% YoY, emphasizing how enterprise AI infrastructure demand has quickly reshaped Cloud’s growth profile. QoQ growth also accelerated double-digits, up 11 points to 24% QoQ.   Cloud continues to show solid signs of AI monetization, with tokens per minute up 6 billion QoQ to 22 billion per minute (or 11.5 quadrillion annualized), matching Q1’s pace. Gemini App MAUs reached 950 million, up nearly 27% from 750 million in Q4.   Also supporting continuation of this strong growth ramp is Cloud RPO, which reached $513.9 billion in Q2, up roughly 375% YoY and 11% QoQ; current RPO was approximately $260 billion. The scale of current RPO at 2.6X of Cloud’s current run rate, and total RPO at 5X+, will likely require rapid capacity expansion to convert this over to revenue, suggesting Alphabet could be preparing for a higher and faster capex cycle – with one other data point likely confirming this.   Cloud operating margin also expanded meaningfully, up roughly 15 points YoY to 35.6% (from 20.7% in Q2'25), and up from 32.9% in Q1'26, offering evidence that the increased pace of infrastructure investments to keep pace with demand are not weighing on the cost side. Management also hinted that its internal infrastructure buildout is not quick enough to meet this near-term demand, explaining that it plans to “expand the use of third-party capacity in Q3 as a bridging strategy while we build up more internal capacity.” This is expected to create some near-term margin pressure as utilizing third-party providers comes with a higher cost, such as Alphabet’s deal with SpaceX for ~110K GPUs for $920 million per month (or a total of $29.4 billion for the entire 32 month term).  Search Grows a Solid 17% YoY as Comps Begin to Toughen  Search & Other revenue was $63.3 billion in Q2, up 17% YoY, moderating slightly from 19% in Q1'26 which came against an easier comp in the single-digits. QoQ growth of 4.8% was broadly in line with typical seasonal patterns. This moderation is not much of a red flag as Search is growing faster on a larger base, when compared to 2024’s growth rates, though it should be noted that comps are getting tougher into the next three quarters.  YouTube revenue increased a healthy 12% QoQ and 13% YoY to $11.06 billion, marking a second consecutive quarter of accelerating growth. Google Network remained a soft spot, with revenue down (1%) YoY but up 5% QoQ to $7.3 billion.   Total Google Advertising revenue increased 14% YoY and 6% QoQ to $81.63 billion, following Search with a 2 point deceleration from 16% YoY growth in Q1. Total Google Services revenue increased 15% YoY and 5% QoQ to $94.54 billion.  Margins Expand YoY, but Contract Sequentially   Gross and operating margin expanded modestly YoY, but both also showed sequential contractions from Q1 as cost of revenue and operating expense growth accelerated.   Gross margin was 61.6%, up 2.1 points from 59.5% a year ago but contracting slightly from 62.4% in Q1. Consolidated operating margin was 34%, up 2 points YoY but down 2.1 points from 36.1% in Q1. The QoQ decline stemmed from Alphabet seeing just a 2.7% QoQ increase in operating income, weighed down by Google Services where operating income declined ($1 billion) from Q1.   One other factor to watch as Alphabet continues to scale investments in Google Cloud and other AI initiatives in Search is operating expenses, as opex growth accelerated 3 points to 27%, outpacing revenue growth for the third consecutive quarter and in five of the last six quarters. Further acceleration in opex could present more headwinds to operating margin moving forward.  EPS is Overwhelmingly a Function of Investment Gains, Not Operations  Alphabet reported a net margin of 93.6% in Q2, which corresponded to GAAP EPS of $9.11, up 294% YoY from $2.31. However, it is important to note that this is not a reflection of Alphabet’s operations, but rather a function of Alphabet’s equity investments in firms including Anthropic and SpaceX, which drove a $99 billion net gain on equity investments in the quarter.   Per Alphabet’s own disclosure, this gain increased the tax provision by $21.9 billion, net income by $77.1 billion, and diluted EPS by $6.26. Backing that out, core EPS was roughly $2.85, up about 23% YoY from $2.31.    Free Cash Flow Turns Negative for the First Time since IPO   Perhaps the most important piece of Q2’s report was not only that free cash flow turned negative for the first time since Alphabet’s IPO in 2004, but that capex guidance suggests quarterly FCF could remain negative through year-end.   Operating cash flow rose 41% YoY to $39.1 billion for a 32.6% margin, though capex grew significantly faster at 100% YoY to $44.9 billion. This pushed Q2’s free cash flow down to ($5.9) billion, or a (4.9%) margin, down from 9.2% in Q1 and 21.6% in Q4. As a result, TTM free cash flow fell (20%) YoY to $53.3 billion.   Driven by strong demand trends in Cloud, Alphabet raised its full-year 2026 capex outlook by $15 billion at the midpoint to a new range of $195–205 billion, up from the prior $180–190 billion guide, pointing to YoY growth of ~119% YoY. The updated capex guidance signals an acceleration in capex intensity into the back half of the year – 1H capex totaled just $80.6 billion, leaving roughly $120 billion in 2H to reach the midpoint of the guide. Assuming a similar ramp profile as 1H, this would project Q3 capex at roughly $55 billion and Q4 capex at $65 billion. For OCF to fully cover this estimated capex and prevent FCF from going negative, Alphabet would have to drive a 45%+ OCF margin in both Q3 and Q4 based on current revenue estimates, a feat it has done only once over the past decade — in Q3 and Q4 2025.   On the cash side, Alphabet bolstered its balance sheet meaningfully during the quarter to help fund this investment: cash and marketable securities rose to $242.5 billion from $126.8 billion in Q1, aided by $30.5 billion in net proceeds from a common stock issuance, $19.1 billion from mandatory convertible preferred stock, and $24.8 billion in net proceeds from debt issuance. Total long-term debt increased to $98.2 billion from $77.5 billion in Q1.   Conclusion  Alphabet spent nearly $45 billion on capex during the quarter, pushed free cash flow into negative territory, raised its 2026 outlook, and said capex will increase again in 2027. Yet despite these outsized investments, Google has stated they do not have enough internal capacity and must rely on more expensive third-party infrastructure to “bridge” the gap.    Big Tech is using its formerly good reputation to squeak by despite capital intensity and purchase commitments that could pose future liquidity risks. I will be blunt – I am an AI bull but not to the point where I will overlook serious risks.   Our portfolio will remain concentrated in the clear beneficiaries that are downwind from the rising capex and ballooning purchase commitments. As for Google, the balance sheet and cash flow profile is too messy to meet our criteria.  Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund do not own shares in (GOOG) at the time of writing and may own stocks pictured in the charts. Recommended Reading: MaxLinear Beats on All Fronts with Keystone; Rushmore Ramping in 2027 GE Vernova Q2: Laying a Strong Foundation for Years to Come AI Networking in 2026: What’s in Motion Tends to Stay in Motion Broadcom Offers Strong AI Growth at Scale; Yet Enters Circular Investing

Since exiting crypto in late 2025, our portfolio has been oriented entirely toward AI. Every name we hold has come through our own proprietary of screening for the best fundamentals, technicals, and product positioning within the tech sector. Our research guidelines allow us to invest anywhere in tech, but we are drawn to where the growth is greatest right now – AI.  We do intend to reallocate to Bitcoin in the coming months, as prices approach our lower targets. Until we see a material shift in the AI story, one that shows up in both the fundamentals and the technicals, we expect to remain heavily exposed to this microtrend.  Within AI, our focus is hardware: specifically, Networking, Memory, and Energy. These are the bottlenecks AI must clear to move forward, and we believe they are where the growth in this microtrend will be concentrated.  The portfolio currently breaks down as follows:  56.7% AI Networking  12.4% AI Energy  21.8% AI Memory  9.2% AI Accelerators  I/O Fund Technical Analysis  The below section dives into the technical analysis surrounding the positions in our portfolio, as well as some that are new ideas, discussed in depth within our Discovery Tier. The technical analysis outlines the two most likely paths that the price action is suggesting, at this time, with levels that will confirm or negate each scenario. We will continue to disuses these setups as they manifest in the weekly webinars.  SiTime Corp. (SITM)  Primary – We are in wave 4 of 3. We should trend toward the buy zone at $598 – $525, then hold $502. We’ll then head toward $1061 – $1215 in wave 5 of 3. Below $502 signals that we are in a larger 4th wave.  Alt – We do not trend to the buy zone, and instead breakout over $901.81.  Breakout: $757.80 – $901.81  Breakdown: $502  Lower Target: $598 – $525  Upper Target: $1061 – $2031  Lumentum (LITE)  Primary – We are in the consolidation of the larger 4th wave. We’ll break below $767 – $690 and then head toward $504 – $324. The 5th wave target will be $2210 – $3623. This will complete a larger 3rd wave.  Alt – We hold $690 and then breakout over $1083.18. We’ll then head toward $1322 – $1718 in an extension of the current 3rd wave.   Breakout: $998 – $1083.18  Breakdown: $690  Lower Target: $504 – $324  Upper Target: $2210 – $3623 Applied Optoelectronics (AAOI)  Primary – We are in wave 4 of a larger 3rd wave. The Lower Target is $130 – $89. We’ll then breakout over $233.87 and head toward the upper target at $298 – $622. A bounce should be happening soon, as the momentum indicators have 3 different time frames at extreme lows – a rare instance that usually does not last long.  Alt – The bounce is a 3-wave bounce that fails to breakout. We then turn lower to the lower boundary of the Lower Target. We must hold $66.   Breakout: $192 – $233.87  Breakdown: $144.97  Lower Target: $130 – $89  Upper Target: $298 – $622  Silicon Motion Technology Corp. (SIMO)  Primary – SIMO broke out in a vertical fashion on volume. This is wave 3, and we are well above that spot. We are in a minor 4th wave and should target the buy zone of $269 – $222.65, hold $216 and then push higher toward $433 – $549. If we breakdown $208, we are in a deeper 4th wave.   Alt – We do not drop into the Lower Target, and instead breakout over $355  Break Out: $319 – $355  Breakdown: $208  Lower Target: $269 – $222.65  Upper Target: $769 – $1762 Micron (MU)  Primary – We are completing wave 3. We will break below $794 and head toward $609 – $392 in a 4th wave. The 5th wave should target $1702 – 2976.  Alt – We’ll break over $1,255, and head toward $1700 next. As long as momentum and volume fade with each breakout, this move higher will be viewed as a 5th wave extension of the larger 3rd wave. Note how the momentum indicator is making. Anew low, while price is making a significantly higher low. This is signaling a large bounce, at least, is likely, and supports the alt count.  Breakout: $1092 – $1255  Breakdown: $794  Lower Target: $609 – $392  Upper Target: $1702 – $2976  MACOM Technologies (MTSI)  Primary – We are in wave 4 of 3. We should hold $216 and then head toward $468 – $700 to complete wave 5 of 3. The current 4th wave Lower Target is $311 – $253.  Alt – We break $216 confirming that we are not going to see a 5th wave swing, and that we have completed the 5th wave. We will be in a larger 4th wave toward $246 – $205.  Breakout: $418.90  Breakdown: $216  Lower Target: $311 – $253  Upper Target: $468 – $567  Advanced Micro Devices (AMD)  Primary – AMD’s pattern appears to be incomplete to the upside. So, we should see one more push higher, at least. This count has us in a very large 3rd wave. We need to hold $452, ideally, on any additional weakness. However, we can drop as low as $347 and still hold the pattern. We would then breakout over $584.73, and head toward $950 – $1827 next.  Alt – The next overhead targets mark a large 3rd wave top in very large ending diagonal.   Breakout: $584.73  Breakdown: $347  Lower Target: $450 – $369  Upper Targets: $950 – $1827  Bloom Energy (BE)  Primary – We are completing a very large 3rd wave. Below $167 and this is confirmed. The Lower Target will be $132 – $77.  Alt – We breakout over $350 and head toward $443 – $660 next.   Breakout: $304 – $350  Breakdown: $167  Lower Target: $132 – $77  Upper Target: $443 – $660  MaxLinear (MXL)  Blue – We are in a 4th wave. This bounce will fail under and push under $77, then turn lower toward the buy zone at $63 – $50.75.  We’ll then turn higher in a 5th wave toward $114 – $170.  Green – The 4th wave made a double bottom and is not going to give us a lower low. We’ll instead breakout over $128 and head directly higher in a 5th wave.  Breakout: $128.30  Breakdown: $49.60  Lower Target: $63 – $50.75  Upper Target: $168 – $245  GE Vernova (GEV)  Blue – We are in the final 5th wave, which is playing out as a diagonal. The b wave should hit $992 – $940 and hold $940. We’ll then heat toward $1370 – $1655 to complete the large 3rd wave.  Green – We’ll instead breakout over $1182 – $1195.94 in a shallow b wave, and then head toward the overhead target.   Breakout: $1182 – $1195.94  Breakdown: $855  Lower Target: $992 – $940  Upper Target: $1370 – $1655  Coherent (COHR)  Primary – We are in the B wave of 5. We should hold $255.14 and then turn higher for a breakout over $399.45 – $438.68. The 5th wave target is generally between $498 – $630  Alt – We break below $250. This signals that we are in a larger 4th wave. The targets will be $331 – $156.  Breakout: $399.45 – $438.68  Breakdown: $255.14  Lower Target: $300 -$250  Upper Target: $498 – $630  SanDisk (SNDK)  Primary – We are completing wave 2 of 3. We dropped into the upper range of this count at $1326, but could go as low as in a minor 2nd wave. We’ll hold $719 and then breakout over $2348 in a large 3rd wave.   Alt – We instead break below $719, signaling that we are in a much larger 2nd wave, which would take months to play out.   Breakout: $2348  Breakdown: $719  Lower Target: $1274 – $893  Upper Target: $3074 – $4455  Corning Incorporated (GLW)  Primary – We are in wave 5 of 3 and should fail soon. We’ll break below $170 confirming this, setting up a drop into the $142 – $113 buy zone for wave 4.  Alt – This is a minor 4th wave low, which will give way to a breakout over $271.78. This would require a 3rd wave extension, which is not as probable as the primary count.   Breakout: $271.78  Breakdown: $170  Lower Target: $127 – $101  Upper Target: $700 – $1032  Astera Labs (ALAB)  Primary – We are completing wave 5 of a very large leading diagonal. We have hit the Lower Target that was laid out months ago. Volume and momentum are lower than they were around the 3rd wave, supporting this count. We should hold under $480 and then break below $315 to confirm this. This will start a large 2nd wave pullback.   Alt – We breakout over $499.55 will suggesting a move to $560 – $776 next. If any break higher is on weaker volume/momentum, it suggests an extension in the 5th wave count.  Breakout: $499.55  Breakdown: $315  Lower Target: $248 – $126  Upper Targets: $364 – $480  Broadcom (AVGO)  Primary – We topped in a large 3rd wave and are now in the early stages of wave 4. We will break through $332, which will confirm this count. We should then trend toward $248 – $193.  Alt – We are in wave 1 of a large 5th wave. We will hold $332 and then breakout over $494.78. This breakout will be a 3rd wave, so volume and momentum will expand with price. We’ll then head toward $868 – $1,362.  Breakout: $494.78  Breakdown: $332  Lower Target: $886 – $13,62, or a strong breakout over $494.78  Upper Targets: $868 – $1362 for the Blue count.   Nvidia (NVDA)  Primary – NVDA has broken through the breakdown warning line $198. It will eventually break the $176.19 support, signaling a bigger correction is underway. The first target will be around $155 – $135. This will be a large 4th wave in a very large uptrend.  Alt – We breakout over $236.55. This breakout would be coming off of 2 overlapping moves, so it would imply an ending diagonal. The targets would be $345 – $370.   Breakout: $236.55  Breakdown: $176.19  Lower Target: $155 – $135  Upper Target: $345 – $370. In conclusion, investors have plenty to worry about: a hawkish Fed, escalating conflict with Iran, and now more cost-efficient Chinese LLMs that could disrupt the AI capex boom. But as we've noted in prior webinars, markets often correct first, and the narrative arrives later to explain the move. Having anticipated an AI correction since mid-May and positioned accordingly, we now think the market is set up for at least one more push higher.  The setups above map these scenarios and the breakout levels that confirm them. Some names are better positioned than others, but once price reaches the overhead targets, our focus shifts back to risk. And if we're wrong, and this is the start of a prolonged decline, we've listed the support levels that would confirm that outcome as well.  For years, one of our key edges at the I/O Fund has come down to a simple, disciplined process where we systematically take profits and raise cash into perceived highs, then redeploy that capital at our predetermined buy targets. This time is no different.  This exact playbook has driven some of our largest outsized winners over the past several years. However, markets are dynamic, and eventually, every systematic approach faces a regime shift where a setup fails.  When that moment comes, our risk management protocol will pivot to aggressively stop out of select positions, build a cash cushion, and lean heavily into our hedge to protect capital on the downside.  However, there is currently minimal technical evidence suggesting this dip is that breakdown point. Until the charts tell us otherwise, we will not let market noise dictate our strategy and instead let the market dictate how we position. The final section of this report addresses setups in 14 stocks that we are currently considering for inclusion into our portfolio. Most of these stocks have been discussed in detail in our Discovery Tier and are reserved for our Discovery Members. They range across software, energy, and memory.   Software led the AI trade in June and is riding renewed momentum. Get Knox's favorite AI software names. Plus, new memory stocks the crowd hasn't flooded into yet, along with two AI-energy stocks riding regulatory tailwinds.  To subscribe to Discovery with 40% off, click here to email us or email premium@io-fund.com and mention code DISCOVERY40 Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Recommended Reading: Positions Report – October 2025 Corning: Glass Manufacturing Powerhouse Pivoting Hard into AI Networking Macom: Data Center Revenue Accelerating to 35% QoQ in FQ3 Arm: Computex Update, CPU Core Demand Hinted at Being Higher

The strength of MaxLinear’s Q2 results were driven by Keystone, its 800G PAM4 DSP platform, with management raising guidance for data center revenue by $50 million to $220 million at the midpoint. Despite Keystone driving this quarter’s results, Rushmore is already sampling with initial revenue expected in 2027.   MaxLinear’s momentum is palpable as total revenue grew 23% QoQ and 55% YoY. Its AI segment, Infrastructure, grew 35% QoQ and 145% YoY. The guidance for Q3 showed unusual strength at 27% QoQ growth and 70% YoY growth for revenue of $215 million.  However, the market may be selling the stock on the lack of updated full-year guidance, but that’s a bit hasty given management doesn’t guide more than one quarter out and we will likely see estimates go up in the coming days/weeks.  When asked about the optical revenue raise and if the run rate will continue into 2H 2026 and next year, management replied: “I mean, we started out with a great run rate going into the year. I think that's just continued to improve. — obviously, raising this number here kind of set expectations for '27 as well. So you would expect that there's not a stair step. I mean, we continue to see as more customers qualifications get completed, move into production volumes you're seeing those numbers go up, and I would expect that to continue into next year.” In other words, management expects a smooth transition between Keystone and Rushmore. 
While the market takes it times ironing that out, an equally important story may be the strengthening bottom line as MXL is now GAAP profitable – although somewhat thin; still an important milestone.   Keystone Product Drives $50M Optical Revenue Raise  In a previous article on MaxLinear for our Discovery Members, we discussed that MaxLinear’s Keystone family third-gen DSPs offer ‘best-in-class’ power consumption, enabling 7W 400G optical modules and 13W 800G designs.   It was stated last quarter that Keystone is ramping at multiple major hyperscale customers across the US and Asia for both 400G and 800G scale-up and scale-out applications. Last quarter, management made it clear the incoming step-function from the Keystone products would sustain: We also expect a step function data center revenue increase beginning in Q2 with expected strong upside as run rates expand into 2027. At the center of this data center momentum is our Keystone PAM4 DSP optical transceiver platform. Keystone is now ramping at multiple major hyperscale customers across both the U.S. and Asia, supporting 400G and 800G — 800G PAM4 deployments for scale-up and scale-out applications.”  This quarter, management emphasized the 40% lower power consumption compared to competing 100G PAM4 DSPs as a primary reason for Keystone’s strong adoption across hyperscalers, stating: “Keystone, our 100 gigabit per lane, 5 nanometer CMOS PAM4 DSP and SerDes technology, continues to ramp into high-volume production at major hyperscale customers across the U.S. and Asia for 400 gig and 800 gig deployments, delivering almost 40% lower consumption of power than competition. 
When pressed if Keystone is simply tracking the market growth or if its gaining market share, management stated the latter: “What we are confident in and what we are seeing is more market share gains. The market is growing nicely, but our share is going up.” It was also stated that Keystone was the first 5m to ship in volume for 100G speeds.   In terms of Keystone remaining strong even as Rushmore ramps, management sounded definitive, stating: “Both 800 gigabit and 1.6 terabit are expected to be workhorse speed nodes for a long time to come, so even as Rushmore ramps, Keystone will still be a growth engine and capacity moving forward.”    Rushmore 1.6T PAM4 DSP is a Major Catalyst  As MaxLinear moves from 800G to 1.6T for its product cycle, the Rushmore 1.6T PAM4 DSP will double the lane speed compared to Keystone. Management expects Rushmore to contribute to revenue in 2027 and stated it will “layer on top of Keystone’s successful ongoing ramp.” Here was another statement to that effect: “Growth is happening through both TAM growth and market-share growth, with performance differentiation and increasing traction from the successful rollout of products to various customers having a knock-on effect of more acceleration in the ramps we are seeing with Rushmore.”  MaxLinear is 1 of 3 major players in the DSP market, yet management feels they offer substantial market differentiation through perforrmance and power advantages, combined with a more diversified supply chain. All of the above is causing strong cross-sell opportunities for Rushmore:   “Keystone is a foundational product for MaxLinear and was the first major one that went to mass rollout, while Rushmore at 1.6 terabit is now sampling with performance and power advantages that are very substantial and supply-chain diversification that is very unique versus competition. The same customers using Keystone are eagerly working towards deploying the 1.6 terabit, and we feel very well positioned to be successful with 1.6 terabit Rushmore as a successor to the Keystone offering.”  Wafer Prepayments Lead to Higher Purchase Obligations  MaxLinear is reporting rising purchase obligations due to wafer prepayments, and this is one reason the stock may be under pressure. Inventory purchase obligations were at $82.3 million in Q1 2025 yet rose to $180.3 million in Q1. Management stated this would rise by another $40 million to $220 million, although the 10-Q is needed to confirm the exact figure. On the positive side, management stated the latest commitments support growth in Q4 2026 and Q1 2027.  During the Q&A, it was further discussed that MaxLinear is not only entering purchase commitments but rather prepaying for wafers and other supply. These prepayments are expected to continue into Q3.  Tore Svanberg:  Yeah, thank you. Just had a follow up and I'll ask a question that has not been asked. So looking at the filing, looks like your purchase obligations went up about $40 million.   But then you also have another obligations item that I think went up Even more than that, $45 million. Can you just explain a little bit, you know, the difference between those two? You talked about obviously the wafer prepays and so on and so forth. I'm sure there's stuff you can do on the back end as well, Any more color on the difference in those two? Because obviously it's pretty important increase in.   Steve Litchfield:  Yeah, yeah. So obviously the purchase obligations are probably the bigger takeaway. We did have some prepayments. I mean with the stock price increase that we saw in the quarter, there were A handful of payroll accruals that had to be done as well. And so that's a portion of it with around. stock comp.  But again the majority is the prepayments and as we had talked about a little bit earlier, that portion obviously supporting growth in Q4 and into Q1 as those lead times we're starting to place orders now for Q1 and that's the majority of those numbers and those commitments.”  The increase is likely a positive given the strong backlog for Keystone, yet it also introduces working-capital risk. While purchase obligations are contractual commitments for future inventory, wafer prepayments can drain cash. MaxLinear ended Q2 with $93.7 million in cash and cash equivalents, yet more than 2X that is committed to future supply prior to revenue being recognized. The concern is not that demand is weak, and probability favors Keystone and Rushmore absorbing the prepayments, yet it raises risk around customer qualifications and shipment schedules.   Financials  Revenue Reaccelerates to 55% YoY on Continued Infrastructure Strength  MaxLinear reported $168.8 million in revenue in Q2, above the high-end of the guided range for $150 to $165 million and marking a reacceleration to 55.2% YoY from 43.0% YoY in Q1. QoQ growth was notable as it accelerated sharply to 23.1% QoQ from just 0.6% in Q1.   For Q3, MaxLinear guided for this revenue momentum to persist, forecasting revenue of $210–$220 million, which at the $215 million midpoint would represent 27.3% QoQ growth and 70.0% YoY growth, both a further acceleration from Q2.   Notably, the $215 million guide for Q3 comes six quarters ahead of consensus estimates, which had MaxLinear reaching this revenue scale by Q4 2027. While management did not guide for the full year, assuming a similar $40 million QoQ step-up in Q4 on the Keystone ramp would roughly estimate FY26 revenue to be $776 million, up 66% YoY and roughly $120 million above the $657 million estimate heading into the report.  Infrastructure Maintains 35% QoQ Growth in Q2  Infrastructure revenue was $85.0 million, up 35.3% QoQ and 145% YoY, as MaxLinear’s  Keystone 800G PAM4 DSP continue to ramp while its Rushmore 1.6T PAM4 DSP is progressing through qualifications. This maintained Q1’s robust sequential growth rate and marked a nine point acceleration from 136% YoY. Additionally, for the first time, Infrastructure represented the majority of MaxLinear's business at 50.4% of total revenue, up from 45.8% in Q1 and just 31.9% a year ago.   Assuming a similar increase in share mix to 54% in Q3 on the continued ramp in Keystone and broadband weakness, this would roughly project Infrastructure revenue to be $116.1 million, or an acceleration to 37% QoQ and 188% YoY.   Management also raised its full-year 2026 Keystone revenue outlook to $190–$210 million, citing robust customer orders and rising visibility into program ramps extending into 2027. This was a notable raise from MaxLinear’s prior view for $150-$170 million in revenue, and based on implied revenue from 2025 around $60-70 million, the new forecast would represent 208% YoY growth.   Other Segments: Broadband Declines (6%), while Connectivity and Industrial Growth Strong  Broadband revenue was $44.9 million, up marginally QoQ at 2.9%, but down (5.6%) YoY, as cable data shipments continue to decline. Broadband growth has plummeted over the last five quarters, from 118% YoY in Q2 2025 to 7% in Q1 and now into the negatives. Broadband’s share of total revenue continued to shrink, now at 26.6% from 44.0% a year ago.   Outside of Broadband’s weakness, MaxLinear’s remaining two segments showed strong sequential growth with 23%+ QoQ in the quarter. Connectivity revenue grew 28.9% QoQ and 15.6% YoY to $24.0 million on higher Wi-Fi shipments, accelerating from 3% QoQ and (8%) YoY in Q1.   Industrial & Multi-Market revenue rose 23.3% QoQ and 158.0% YoY to $15.0 million on higher component and analog shipments, accelerating from (14%) QoQ and 47% YoY in Q1. Overall revenue contribution from the two remains smaller, at 14% and 9% respectively.   Margins Expand Sharply on Infrastructure Mix Shift  Gross margin expansion is rather minimal, but the more important piece for margins lay within Q3’s guide, as it implied strong operating leverage arising in the quarter as operating expenses were forecast to be flat QoQ. This dynamic pushed Q3’s GAAP operating margin guide to the low double-digit range, a substantial improvement from (12.5%) in Q1 and (32.7%) a year ago.   GAAP gross margin was 57.8%, up marginally QoQ and 1.3 points YoY, while non-GAAP gross margin was 59.5%, flat QoQ and up marginally YoY.   GAAP operating margin showed a strong improvement and nearly broke even in the quarter,  improving 10 points QoQ and 20.1 points YoY to (2.5%). Adjusted operating margin expanded to 22.3%, up 6.4 points QoQ and 15.1 points YoY — the highest level since the start of 2023 — as higher-margin Infrastructure mix and moderating opex growth (opex up just 6% QoQ against 23% revenue growth) drove leverage. The 24-point gap between GAAP and adjusted margins boils down to high SBC and performance-based equity costs at ~$36.5 million this quarter.   GAAP net margin broke into positive territory, coming in at 1%, up from (24.4%) a year ago and (32.9%) in Q1. Adjusted net margin improved to 20.3%, up from 1.6% a year ago and 14.2% in Q1.   Q3’s guide for operating margin was an impressive piece of the report. Despite minimal gross margin expansion with GAAP/adjusted gross margin guided to 58.5%/60% respectively, MaxLinear forecast opex flat QoQ at $101 million.   This implies that the ~$30 million guided QoQ increase in gross margin will flow directly down the line, with Q3 GAAP operating income forecast at $24.8 million for an 11.5% margin, up from (2.5%) in Q2 and (32.7%) a year ago. Adjusted operating margin was guided to be 28.1%, up from 22.3% in Q1 and 12% a year ago.  EPS Turns GAAP-Profitable for the First Time Since the 2023 Downturn  GAAP diluted EPS was $0.02, MaxLinear's first positive GAAP quarter following twelve consecutive quarters of GAAP losses, comparing to $(0.52) in Q1 and $(0.31) a year ago. The swing to profitability was aided by an $8.3 million income tax benefit in the quarter, in addition to the operating improvement described above. Adjusted EPS was $0.35, up 59% QoQ from $0.22 and up 1,650% YoY from $0.02, marking the fourth consecutive quarter of accelerating growth.  Cash Flows Thin with Single-Digit Margins  Cash flows swung positive in Q2 after being negative in Q1, though cash flow margins were quite thin in the low single-digit range.   Operating cash flow was $4.8 million for a 2.8% margin, swinging positive from ($8.9 million in Q1) but down from Q2 2025’s $10.5 million at a 9.6% margin.   Free cash flow was $2.5 million for a 1.5% margin, improving from ($10.3 million) in Q1 but down from Q2 2025’s $9.3 million or an 8.6% margin. The YoY decline in cash generation was driven by a substantial working-capital build — inventory rose 22.9% QoQ to $105.5 million as MaxLinear prepaid for wafers to support rising demand for optical data center products.  Cash, cash equivalents and restricted cash totaled $93.7 million, up from $89.9 million in Q1 but down from $110.3 million a year ago, while long-term debt was largely unchanged at $123.9 million. In April 2026, MaxLinear amended its credit agreement to extend the revolving facility's maturity to March 2028 and added $30 million of incremental revolving commitments, providing access to additional liquidity although the revolver remained fully undrawn as of quarter-end. Conclusion:  MaxLinear is seeing negative price action after hours, but this is not a report we are concerned with. The report ticked a lot of boxes with Keystone driving an acceleration in optical data center revenue, management raising 2026 optical outlook, plus a stellar Q3 raise. For quick reference, the Q3 estimates going into the print were for revenue of $173.8M for 3% QoQ growth and 37% YoY growth. Instead, management guided for 27% QoQ growth and 70% YoY growth. The company is also newly GAAP profitable, albeit narrow.  The main point of caution is the intersection of increasing wafer prepayments alongside management not officially updating fiscal year guidance. However, when we connect the dots that hyperscaler capex remains strong per Google’s commentary last night, the probability MaxLinear is securing capacity for demand that doesn’t materialize is fairly low.   Similarly, chances favor analysts raising Q4 and fiscal year consensus in the coming days/weeks given that Q3 was so far above estimates. This kind of disconnect between the initial stock reaction and the underlying report is exactly why we pore over the details.   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 MXL at the time of writing and may own stocks pictured in the charts. Recommended Reading: GE Vernova Q2: Laying a Strong Foundation for Years to Come AI Networking in 2026: What’s in Motion Tends to Stay in Motion Broadcom Offers Strong AI Growth at Scale; Yet Enters Circular Investing Nvidia Fiscal Q1: Perfect Quarter, Imperfect Catalysts

Just like in the Wizard of Oz, the yellow brick road for the AI industry runs right to the emerald city, Nvidia. The sheer pace and scale at which Nvidia is ramping its GPUs is placing immense stress across many parts of the supply chain, from memory to optics, yet the company’s Rubin generation ratchets this up a notch in the one area that relief is hard to come by – 3nm capacity.   Up until this point, Nvidia’s GPU roadmap has been manufactured on TSMC’s 4nm nodes, with Hopper utilizing the N4 process and Blackwell adopting a custom version, N4P, optimized for improvements in power efficiency and performance.   With Rubin, Nvidia is shifting to 3nm in high volume, as  Rubin and Rubin Ultra both are set to adopt TSMC’s N3P process, the third-generation upgrade to 3nm that delivers single-digit gains in power efficiency and performance versus TSMC’s optimized N3E process.  This shift will pit Nvidia’s demand head-to-head with two of TSMC’s larger customers (considering that Nvidia is now its largest), Apple and Broadcom, both of whom have already adopted 3nm in high-volume. Apple is tapping TSMC’s N3E process for its M4 and A18 chips for Macs and iPhones, and it was rumored in 2023 that Apple was purchasing all of the 3nm chips that TSMC could make.   Broadcom transitioned from N5 to N3E for production of Google’s TPU v7 Ironwood in 2025, remaining on N3E for the new TPU8t and broadening to the N3P process for the TPU 8i. Meta’s MTIA 300, which entered production in late 2025, and upcoming MTIA 400 both will adopt 3nm, transitioning from 5nm on the MTIA 200 in 2024.   Additionally, Amazon is transitioning to 3nm for the first time with Trainium3, which is scheduled to ramp in 2H 2026. AMD’s MI350 and upcoming MI400 chips powering its Helios rack will also use N3P. Last but not least, Microsoft is tapping N3P as well for its Maia 200 accelerator now shipping.   The 3nm bottleneck arises with the convergence of major AI accelerator roadmaps moving to and ramping on 3nm alongside Apple’s needs – but the scaling-up of Rubin Ultra towards 576 GPUs adds another layer by helping drive an expected 3X growth in 3nm-based accelerator shipments in 2027, far outpacing capacity growth.   3nm in a Rare State of Overload, Analysts Guess It’s 30-50% Short of Demand  Analysts guessed in Q2 that the “the unconstrained demand for 3-nanometer and below is sort of 30% to 50% above your ability to supply” and could take three to four years to solve, to which TSMC responded that the “gap is very big.” This detail essentially confirms reports from March that 3nm had “entered an extremely rare state of ‘overload’,” and recent reports from June stating 3nm lead times had surpassed one year.  Let’s first go back in time a bit to understand the sheer pressure 3nm faces. At the end of 2025, TSMC’s 3nm capacity at year-end 2025 was implied to be 120-130K wafers/month. TSMC’s initial plan reportedly plotted out 150K/month capacity by year-end 2026, slightly below Nvidia’s request that TSMC reach 160K wafers/month.  Capacity estimates for 3nm have steadily moved higher, and year-end 2026 estimates now sit around 200K/month, a 33% increase from TSMC’s original reported plan. 2027’s initial estimate calls for 250K/month capacity, a 25% increase.   The main takeaway: original capacity estimates pointed to 160K/month by year-end 2026, while now it sits 25% higher at 200K/month and still is significantly short of demand.  TSMC Flexing Pricing Power with 3nm Prices Expected to Rise Through 2027   Given the tightness on 3nm and TSMC’s pole position as the primary volume supplier for the entire AI industry, the chipmaker is expected to flex its muscles and exert its pricing power later this year and through 2027. This is important because 3nm is already TSMC’s second-largest revenue driver, behind 5nm, and increasing prices as multiple high-volume accelerators ramp through 2027 could quickly take the node to its top revenue contributor.   For a quick snapshot, Q2 saw 3nm contribute 30% of TSMC’s revenue, up from 25% in Q1. This corresponds to revenue of ~$13.3 billion, up nearly 84% YoY and 48% QoQ, and marking the first time 3nm revenue surpassed $10 billion. At the midpoint of Q3 guidance for $44.6 to $45.8 billion, and assuming a 3 point step up in mix to 36% as Rubin begins to ramp, 3nm revenue could hit $16.3 billion next quarter, up 115% YoY and 23% QoQ.   TSMC is expected to hike 3nm prices in the second half of 2026 by ~15%, followed by another 5-10% hike in 2027. Given that current 3nm prices are pegged at $21,000 to $22,000, these price hikes could see 3nm wafers cost more than $27,000 by the end of next year.   While it’s clear to see how pricing power and capacity growth could compound for TSMC’s revenue – such as moving from ~$3.4 billion at $21,500 per wafer and 160K/month estimated currently to nearly $6.8 billion at $27,000 and 250K/month possible by the end of 2027 – there are some implications that could be felt further downstream by TSMC’s customers.  For Nvidia, price hikes are likely to pose little concern, as CEO Jensen Huang has publicly stated his full support for 3nm price hikes. Not only does Nvidia have the scale, margins and token performance metrics to offset these hikes, it also has the cash flows to support meaningful supply allocations, potentially at lower prices, as evidenced by its $119 billion in supply commitments.  However, there is a question of whether custom silicon players, notably Broadcom and Marvell, are a bit more at risk from price hikes, considering one of the main value propositions of custom chips comes down to that cost advantage. Rising 3nm prices could squeeze margins if the two are unable to pass on elevated costs to customers, which adversely affects that cost advantage. Margin-related impacts also overhang on 3nm-based DSP suppliers, where both Marvell and Broadcom also play, alongside Credo and other smaller vendors.   Plotting Out the Shortfalls: Accelerator Shipment Forecasts Outpace Capacity Growth  Looking at projections and estimates for accelerator shipments through 2026 and 2027 provides more color on why the node is this overwhelmed.   Supporting a ramp from ~9M accelerator shipments in 2026 to 25M+ in 2027, based on projections and estimates above, suggests that 3nm is likely to remain in a prolonged state of overload. It’s rather simple math – >3X growth in shipment volumes versus 2X growth in 3nm capacity.   The main takeaway is that Nvidia and Google and the two largest drivers of this shipment growth. Analysts are modeling Nvidia shipping up to 30K Rubin racks in 2026, and remaining conservative on 2026 at >1K racks per week like it maintained with Blackwell could see Nvidia put out 60K racks in 2027.   Assuming ~15% penetration next year for the NVL576, Nvidia’s Rubin demand could jump 4X from 2.2 million to 8.9 million; with 20% penetration, Rubin demand moves nearly 5X higher to 10.4 million.  Google is estimated to see a similarly substantial increase in TPU shipments as its 8t and 8i generations ramp into next year. UBS projects Google will ship 4.13 million TPUs in 2026, before more than doubling to 9.87 million units by 2027, helping support multi-GW capacity expansion for Anthropic.   Combining these two giants with AMD ramping Helios, Amazon ramping Trainium, Meta ramping its MTIA roadmap, TSMC must still account for several million more accelerators — as well as millions of CPUs, networking switches and components such as DSPs, and HBM logic dies for HBM4 and HBM4e.  Alleviating 3nm Tightness is No Easy Game  The primary lever to increasing 3nm capacity is via new fabs, which is a lengthier process due to the longer timelines required to build new cleanroom space. TSMC is afforded quicker timelines in Taiwan from fast-tracked permitting processes, with new fab construction estimated at around 1.5 to 2 years, whereas US expansion is said to take around 3 years.   As a result, any new capex being put towards new fabs (not any existing facilities) would not move the needle on 3nm capacity until mid to late 2028 at the earliest, meaning this will have no effect on 3nm tightness through next year. Improvements on the yield and defect side could also play a role in helping solve that 3X shipment growth versus 2X capacity growth shortfall, as it could help TSMC get more viable chips from its current footprint (although this is a smaller lever considering 3nm yields range from 80-90%).  Luckily, TSMC does have fab expansion efforts currently underway, aiming to bring new 3nm capacity online in Taiwan, Arizona and in Japan; however, the main problem here is these fabs are not expected to come online until early 2027 for Taiwan, and late 2027 through 2028 for the international two. These expansion efforts will layer into to the 2027 capacity increase and provide outlets for future growth in 2028 – with Japan offering 15K/month and Arizona offering up to 30K/month — but again, the growth will likely not come soon enough or large enough to alleviate the tightness.   TSMC does have routes to increase near-term capacity, primarily via retrofitting capacity on older nodes, such as on 5nm as detailed in Q1’s earnings and re-emphasized in Q2. JP Morgan believes that TSMC could convert some 4nm capacity previously used for smartphone chips to 3nm, freeing up 20-30K wafers, while 5nm/7nm reallocation could add 5-10K wafers/month. Based on a 160K wafer/month baseline as targeted in Q2, this could create a ~25% uplift in capacity, though considering the elongated lead times and strength of demand, any additional capacity is likely to be immediately soaked up.  Outside of TSMC, Samsung and Intel both offer 3nm manufacturing, though at far smaller scales (simply look at TSMC’s 72.3% market share versus Samsung’s 6.5% in second place). Neither are likely to help aid 3nm constraints via diversification, as Samsung has reportedly faced substantial yield issues on 3nm, per Chosun, while Intel may only be adding pressure as it reportedly is securing 3nm supply at TSMC to help meet Xeon server CPU and GPU demand.   CoWoS Now Rapidly Expanding through 2027 It would not be a TSMC bottleneck discussion without CoWoS, where capacity still remains extremely tight in a rather familiar narrative over the last three years. Management put it quite bluntly in Q2: “packaging capacity is so tight that it is limiting [our] customers’ growth,” even as CoWoS capacity is expected to roughly triple from 2025 to 2027.  Currently, TSMC’s CoWoS supply is estimated to fall ~20% short of demand, though reports suggest this gap could narrow towards 10% by year-end as capacity comes online. However, CoWoS capacity allocations in 2027 show Nvidia’s competitors gaining significant share, hinting that CoWoS may no longer be the number one bottleneck at TSMC beyond this year. This isn’t to say the CoWoS bottleneck will magically be solved, but rather that 3nm could present a tougher bottleneck for TSMC and customers than CoWoS come 2027.  For 2026, TSMC’s CoWoS capacity is estimated to nearly double to 1.275 million wafers, though other supply chain trackers project capacity to reach as much as 120K-130K/month by year-end, or exiting capacity of 1.5 million at midpoint.   For 2027, Goldman Sachs projects an >80% increase in capacity to 2.31 million wafers, with Mizuho also projecting CoWoS to reach 2.34 million. This would essentially triple capacity from 2025 and represent >27X growth from 2021 (or a 74% CAGR).  On the allocation side, Nvidia still is securing a majority of CoWoS capacity, rising from 385K in 2025 to between 650K to 780K in 2026. Estimates from KeyBanc suggest that the 650K allocation could support 5.5-6 million Blackwell GPUs, as well as 1.5 million Rubin and 1 million Hopper GPUs (or a total of up to 8.5 million chips).   2027 allocation estimates for Nvidia see quite a wide range currently – some estimates place its secured supply at 840K wafers for the year, while other estimates, such as those from Morgan Stanley and Mizuho, see Nvidia’s allocation ranging between 1.01 million to 1.22 million. At those upper end forecasts, Nvidia would be securing roughly 44-52% of TSMC’s CoWoS (assuming all goes to TSMC), leaving more than 1 million wafers for the rest of the industry.   It is this broadening of allocation beyond Nvidia that suggests CoWoS will likely not present as tough a headwind as 3nm, as companies with much smaller volumes and much smaller cash flows than Nvidia (ie not able to secure meaningful supply well in advance) are seeing substantial increases in CoWoS allocation in 2027.   To note, these figures above represent global CoWoS capacity, including TSMC, ASE and other outsourced assembly and test providers.   Combined allocations for Broadcom, AMD, AWS, MediaTek and Marvell are expected to jump almost 141% YoY in 2027, per Morgan Stanley, with AMD’s allocation rising more than 300% to 530K wafers, likely supporting a strong Helios ramp. Broadcom is expected to see ~60% growth to 484K wafers, while MediaTek is expected to see its allocation more than quadruple to 180K, with both supporting Google’s newest TPU.   There is still a high chance for CoWoS capacity to remain tight beyond 2026, considering the pace of growth TSMC must maintain to add capacity of more than 1 million wafers YoY. This CoWoS expansion remains limited geographically to Taiwan, as US-based capacity in Arizona is not expected to be ready until 2028 at the earliest. TSMC did announce a $100 billion boost to its Arizona investments to $265 billion, and while some of this will go towards new advanced packaging plants, the majority is likely geared towards <2nm capacity.   TSMC is also beginning to ramp its 3D hybrid bonding packaging tech, SoIC (system on integrated chip), which will vertically stack dies to lower power consumption via reduced interconnect reach. SoIC capacity is expected to reach around 10-15K wafers/month by the end of 2026, ramping in 2027 to support demand from Nvidia, AMD and Broadcom, though the larger tailwind for SoIC capacity is expected to be Nvidia’s future Feynmann generation.  True Relief Comes From New Cleanroom Space = More Capex  As mentioned above, true relief for 3nm capacity will come from accelerating construction of new cleanroom space and new fabs, something TSMC is well aware per Q1 commentary: “we have to speed it up with our build of cleanroom and buying the tools.” Reallocating and retooling older nodes to 3nm is simply a bridge to help meet demand until new fabs can come online.   The result will be a strong upwards trajectory in TSMC’s capex, not just for 2026, but through 2028. Management explained that capex over the last three years was $101 billion, with 2025 seeing a more than $11 billion increase to a new record. 2026 capex was initially guided to be $52 to $56 billion in Q2, as CFO Jen-Chau Huang outlined that TSMC expects capex “in the next 3 years will be significantly higher than the past 3 years.”   TSMC is showing that statement to be true, boosting capex to $60 to $64 billion in Q2, a nearly 15% raise and pointing to almost 51% YoY growth. The primary reason is to expand 3nm capacity to meet a robust multi-year demand pipeline (and a bit of tool inflation). TSMC has also been allocating more towards advanced packaging for CoWoS and SoIC, as seen further below.  At $62 billion in capex, this would represent around 36.2% of its projected revenue of $171.3 billion (as management raised 2026 growth to 40% YoY), a step up from the mid-33% level in both 2024 and 2025. Assuming TSM maintains this level of capex through both 2027 and 2028, to support aggressive 3nm, other advanced node and advanced packaging expansion, capex could reach $80 billion and $100 billion, respectively, based on estimated growth of 28% and 26% YoY in both years. This would put its three-year total for 2026 through 2028 to $242 billion, or nearly 2.5X the $101 billion it spent in 2023 through 2025, aligning with management’s Q2 commentary that the three-year total would be “even more significantly higher.”   This also begs the question – was TSMC caught flat-footed for this upcoming surge in 3nm demand? Put another way, was that $101 billion in cumulative capex from 2023 to 2025 far too low to support the necessary capacity ramps across 3nm and CoWoS to prevent these shortfalls? It’s something to ponder considering how quickly TSMC looks to be raising capex to meet that 30-50% estimated 3nm shortfall and continued CoWoS tightness.   For a view on advanced packaging capex, TSMC has begun stepping this up to ~10-20% of capex, from ~10% as recently as 2024, primarily to support CoWoS. At 15% of capex in 2026, this would roughly project $9.3 billion in advanced packaging capex, or more than 3X the ~$3 billion spent in 2024. Maintaining 15% through 2027 and 2028 (considering that SoIC is expected to cost ~$7 billion per 10K wafers), would see capex of ~$12 billion and $15 billion goes towards advanced packaging in both years.   Who Stands to Benefit  This projected increase in capex on TSMC’s end is rather necessary to prevent it from becoming a bigger choke point on the AI buildout, with rather direct implications downstream for the WFE industry, from equipment vendors such as ASML to testing and metrology providers.   As a result, WFE growth forecasts have seen material raises recently, with SEMI raising its 2026 forecast from $145 billion to $165.9 billion, representing 23.2% YoY growth (versus 7.6% previously). The majority of this is expected to be driven by fab equipment, up 23.1% YoY to $143.9 billion, with test equipment up 31% to $15.3 billion and assembly and packaging rising just 9.6% to $6.7 billion. SEMI also projects strong growth continuing in 2027 and 2028, forecasting 21.3% growth in 2027 to $201.2 billion followed by 14.1% growth to $229.5 billion.  TSMC explicitly called out three key suppliers in Q1 – ASML for EUV tools, and Applied Materials and Lam Research for etch, deposition and process integration tools. However, beneficiaries expand far beyond these three, including some indirect beneficiaries of this 3nm tightness.   ASML’s lithography tools are essential for TSMC, its largest customer per Bloomberg, with EUV systems supporting advanced node manufacturing and DUV systems key for CoWoS packaging. During Q2’s report on Wednesday, ASML boosted its revenue guide for the year to €43-45 billion (from €36-40 billion) and noted that customers continue to “accelerate their capacity expansion plans.” This is driving increasing visibility and stronger order intakes, with ASML aiming to boost EUV and DUV capacity by 30% this year. ASML also hinted at long-term price increases for both systems, though TSMC is said to be resisting these efforts (considering DUV systems are said to cost $90 million and High-NA EUVs up to $400 million). It was also reported in April that TSMC was holding off on buying the High-NA systems due to the cost.  Applied Materials and Lam Research both supply etch and deposition tools, which are becoming increasingly important to help construct the complex transistor and interconnect structures within next-gen GPUs, such as Rubin’s >61% increase in transistor count to 336 billion. Similar to ASML, Applied Materials sees “tremendous” order visibility spanning at least two years. The two also provide specialized process integration tools that aim to increase the speed or efficiency of chip production.  One vital lever in solving the main question raised above – how TSMC can meet potential 3X growth in shipment volumes with 2X growth in capacity – comes down to yield, as even marginal increases in 3nm yield (from the estimated 80-90% range) could help TSMC squeeze more chips out of its existing capacity footprint. KLA provides reticle inspection systems and metrology tools that help identify defects on 3nm and advanced DRAM chips, key in keeping yields high as more complex chips require more inspection steps.   TSMC competitor Intel could also benefit on both the fab and packaging side, as Nvidia has recently indicated will move allocation here in the medium-term and Google is said to have placed an order for 3 million TPUs with Intel in 2028. The two are reportedly exploring having Intel serve as a backup manufacturer to TSMC for its advanced nodes and EMIB-T advanced packaging services. When asked about EMIB-T as a competitive threat, management downplayed that and instead stated that they “welcome that additional flexibility in the market. And so that will help TSMC's front-end wafer business growth” as it could allow for accelerator shipments above TSMC’s CoWoS limits.  Networking Could Face Issues in Getting 3nm Supply  Given the challenges of satisfying demand for Nvidia, Broadcom, Apple and other key customers, one does wonder how TSMC allocates all of this capacity considering not everyone has the pockets of Nvidia to snap up supply.   Credo’s executives made a key concern on this regard from the point of view of the optics supply chain — TSMC cannot overallocate to Nvidia and must reserve capacity for optical DSPs, in order to prevent the optical connectivity bottleneck from worsening.   Here’s what Credo stated:  "And let's talk specifically about 1.6T because everybody sees that's where the market is going. Every solution that we're aware of that does 200 gig per lane, 1.6T, 8 lanes of 200 is done at 3-nanometer. So there's — I don't think there's any 5-nanometer that are going to go to volume production because power is simply too high. …  And by the way, TSMC fully understands that these small complementary connectivity chips are needed to deploy clusters. If you take this small amount of wafers and you short those, you're basically locking in disruption in the entire deployment.”  This is the very tight rope the industry has to walk. 200G and 400G DSPs from Marvell, Broadcom and others, necessary for enabling 1.6T optics, are built on TSMC’s 3nm node, and as you may recall from our Networking thematic, 1.6T is a necessity with Rubin with content per rack expected to double. This will require a similar step-up in DSP demand that has to consume a portion of 3nm capacity – and TSMC is well aware of this.   However, it’s a ‘small fish in a big pond’ for the networking companies, who do not have the budgets or cash flows to go out and pre-pay billions to lock in supply; rather, they sit and wait for TSMC to allocate what is left after the big fish secure their share. This suggests that the largest impacts from the 3nm crunch could fall on the networking stocks and DSP providers, as they also could have to bear the brunt of wafer price hikes on margins.   Quick Note on Geopolitical Risk  One of the other main risks associated with TSMC is geopolitical risk with China, which is nothing new. This extends beyond 3nm to other advanced nodes (5nm/2nm) as well as packaging, as the majority of TSMC’s advanced node and packaging fabs remain in Taiwan.  While TSMC is progressing with its large-scale expansion in Arizona and other expansion efforts in Japan and Germany, meaningful capacity internationally is unlikely prior to 2028. Any escalation in tensions between Taiwan and China could further threaten the pressure than the 3nm node is placing on the AI buildout, as there are no near-term routes to minimizing geopolitical risk or turning to alternative manufacturers at the required scale.  Conclusion  TSMC has been central to one of the AI industry’s longest-standing bottlenecks, CoWoS capacity, since mid-2023 with capacity remaining extremely tight as of Q2 2026. However, the more critical constraint that is emerging is TSMC’s 3nm capacity.   This comes down to a few core factors: the convergence of all major AI accelerator roadmaps ramping in volume through 2027, the transition to 3nm for upcoming critical components such as 200G DSPs, HBM logic dies and networking switches, and smartphone chip production.   To put in perspective the potential 3nm shortfalls, accelerator shipments alone could see 3X growth in 2027 compared to 2X growth in monthly 3nm wafer capacity, with analysts implying in Q2 that unconstrained 3nm demand could be 30-50% higher than TSMC’s ability to supply.  This puts the focus on the WFE industry, from fab equipment suppliers to inspection and metrology players, as the primary relief for 3nm will come from new cleanroom capacity. This is evident within TSMC’s upgraded capex guide to $62 billion this year, up 50% YoY, with potential for 2026-2028 capex to be nearly 2.5X higher than the last three years, setting up a strong runway for WFE growth. Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund do not own shares in TSM 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: The I/O Fund’s Top 20 Stocks for Q3 2026 GE Vernova Q2: Laying a Strong Foundation for Years to Come AI Networking in 2026: What’s in Motion Tends to Stay in Motion Broadcom Offers Strong AI Growth at Scale; Yet Enters Circular Investing