Nvidia Issues First-Ever Full-Year Guide of 70% Growth While Warning Memory Costs Will Squeeze Margins to 71%
Q2 fiscal 2027 earnings call, August 26, 2026 — record $96 billion quarter as AI infrastructure buildout accelerates for a fourth straight quarter
Nvidia delivered a quarter that once again outpaced Wall Street's models, with revenue of $96 billion more than doubling year-over-year, but the headline for investors was not the beat. It was two structural disclosures that changed the shape of the story: the company's first-ever full fiscal year revenue guide, and an unusually direct admission that memory component costs are about to compress gross margins more than previously expected.
A Supply-Constrained 70% Guide, With Demand Running at 100%
For the first time in its history, Nvidia gave investors a full fiscal year outlook, guiding to approximately 70% revenue growth in fiscal 2028. CFO Colette Kress was explicit that this is "a supply-constrained outlook," and CEO Jensen Huang confirmed on the call that underlying customer demand is actually growing near 100%, meaning the company is guiding to roughly two-thirds of what it could sell if capacity were unlimited. Huang told analysts the decision to guide a full year out for the first time was deliberate transparency given the scale of capital now being committed across the ecosystem: "We wanted to make sure that everybody has the same set of information... we've got a huge year coming up next year, and it's going to be pretty extraordinary."
The bottleneck, according to Huang, is not technology or customer appetite but physical infrastructure — land, power, and shell construction that now requires two to three years of lead time, pushing Nvidia further upstream into memory and power supply chains and downstream into land and shell deals. Jim Schneider of Goldman Sachs pressed for a ranked list of constraints — power, DRAM, foundry capacity — and Huang declined to isolate one, saying simply "our entire supply chain is challenged" and that Nvidia currently has visibility into roughly 70% of the capacity it needs to meet demand, with yield and capacity improvements expected incrementally throughout the year rather than in a single step change.
Memory Scarcity Forces a Gross Margin Reset
The most consequential negative data point in the print was gross margin guidance. Nvidia guided Q3 GAAP and non-GAAP margins to 74%, plus or minus 50 basis points, and told investors margins will bottom in Q4 at 71% to 72% before recovering to a 72% to 73% range in fiscal 2028 as executed price increases take hold in the first quarter. Management was unusually blunt that this reset was overdue: "The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year. As a result, we are resetting expectations today," Kress said, adding the company wanted to "be direct about this rather than let it linger as an open question."
Nvidia framed the memory shortage as a byproduct of its own success rather than a pure cost headwind, noting the same AI buildout driving its growth is also draining DRAM and HBM supply industry-wide. The company said it maintains "long-standing deep relationships with all 3 major memory suppliers" and is working to expand capacity against its roadmap, but the near-term margin compression is real and will run through at least the December quarter.
Nvidia Addresses the Circular Financing Question Directly
With roughly $50 billion already invested in frontier AI labs and a new set of financing partnerships announced with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR designed to raise over $500 billion in third-party capital, Nvidia used the call to confront the circularity criticism head-on rather than deflect it. "We recognize the scale of this support, and we know some will call this circular financing. We see it differently," Kress said, arguing the labs are "once in a generation companies" whose growth is capital-constrained rather than demand-constrained, and that Nvidia's compute is fungible enough to be redeployed to other customers if any single relationship were to falter.
Management disclosed that demand from AI labs where Nvidia is leveraging its balance sheet will contribute roughly a quarter of total business next year, and reiterated that all compute shipped will go to investment-grade customers or those backed by one. Vivek Arya of Bank of America pushed further, noting that OpenAI has publicly touted its in-house Jalapeno chip as outperforming Blackwell even as Nvidia continues to fund the company. Huang was unbothered, arguing that custom inference chips built for a single cloud cannot match a platform that spans the full AI life cycle across every cloud: "I have 100% confidence that our technology will continue to be extraordinary for them... I'm delighted to be their friend."
New Deals: AWS Adds 2 Million GPUs, SoftBank Powers OpenAI's Portsmouth Campus
Nvidia disclosed an expanded partnership with Amazon Web Services under which AWS will deploy an additional 2 million GPUs from this quarter through the second quarter of fiscal 2029, alongside Vera CPUs both integrated with Rubin and sold standalone. AWS will also host Nvidia's Nemotron open model family on Bedrock and SageMaker and adopt Nvidia's physical AI stack — Omniverse, Cosmos, Isaac and Jetson — for its warehouse robotics fleet. Separately, Nvidia detailed a new SoftBank Energy partnership securing land, power and shell capacity at a Portsmouth campus expected to support 4.25 gigawatts of AI factory capacity, to be utilized by OpenAI. Management said each generation of systems deployed there could represent approximately 1.5 million Nvidia GPUs, with the site capable of supporting multiple upgrade cycles over 20 years. OpenAI's total committed and planned deployments now stand at roughly 12 gigawatts of Nvidia compute through 2030. For a separate unnamed frontier lab, Nvidia said it will provide "selective credit enhancement" for nearly 2 gigawatts of compute layered on top of capacity that lab has already secured independently.
Groq 3 LPX Reaches Production, Targeting High-ASP Inference
Following up on last year's Groq partnership, Nvidia confirmed that Groq 3 LPX, its first rack-scale LPU system, is now in full production and set a record of nearly 4x the tokens-per-second throughput of the next best alternative on the company's artificial analysis benchmark. Volume shipments begin later this quarter, with Nebius as the first customer. Huang characterized this as a complement rather than competitor to the core NVLink 72 architecture, aimed narrowly at services requiring extreme interactivity: "For services that would like to have super high interactivity, super high-speed token generation... you could bolt on one of our Groq accelerators. But the vast majority of the world's data centers will just be Vera Rubin and NVLink 72."
Revenue Per Gigawatt Keeps Climbing, and Management Says It Isn't Done
Nvidia quantified how its addressable spend per gigawatt of data center capacity has scaled across generations: roughly $18 billion with Hopper, $25 billion with Grace Blackwell, and now $40 billion with Vera Rubin once CPU, NVLink, InfiniBand or Ethernet networking, and Groq LPUs are included. Huang indicated this trajectory continues upward, framing the ideal outcome in almost absurdist terms — "the perfect answer is actually infinity per gigawatt" — while noting return on invested capital for a $50 billion data center is now reportedly under a year, which he said explains why hyperscalers and NeoClouds continue to race toward each new generation rather than sweat the sticker price.
Non-Hyperscaler Business Becomes Half the Data Center Franchise
Data center revenue rose 18% sequentially to $89 billion, split between hyperscale revenue of $49 billion, up 13% sequentially, and the ACIE segment — NeoCloud, industrial and enterprise customers — which grew 25% sequentially and 138% year-over-year to $40 billion. Management said this non-hyperscaler bucket, growing roughly 100% annually according to Huang, is on track to represent roughly half of total data center revenue going forward, a segment he described as largely "invisible" to investors focused only on the large cloud platforms. Sovereign AI revenue, run primarily through regional NeoCloud partners, grew 35% sequentially and more than tripled year-over-year, with new regional buildouts cited across Japan, South Korea, India, Africa, Australia, Malaysia and Armenia.
Nvidia also disclosed a new revenue-sharing financing structure for NeoCloud partners, under which it provides a take-or-pay minimum revenue guarantee on a portion of a facility's capacity to help lenders underwrite the project, in exchange for a share of rental revenue above that floor. Kress described the mechanics plainly: "We get paid twice, once on the hardware sale and again, through the share of rental revenue," calling it a structure with "the potential to drive billions in revenue over the medium to long term."
China Remains Excluded From Guidance
Hopper 200 shipments to China-based customers under U.S. government licenses amounted to less than 1% of total data center revenue in the quarter and are dilutive to corporate gross margin. Management reiterated that its forward outlook contains zero China data center compute revenue given ongoing geopolitical uncertainty, treating any future resumption as pure upside rather than a planning assumption.
Capital Returns and the Quarter's Numbers
Networking revenue also hit a record, up 18% sequentially, with Spectrum-X Ethernet up 2.6 times year-over-year. Vera CPU is now in full production and management's preliminary expectation is for CPU revenue to more than double in fiscal 2028 against roughly $20 billion of total server CPU demand the company says it is tracking. Nvidia returned $26 billion to shareholders in the quarter, split between $20 billion of buybacks and $6 billion of dividends, putting year-to-date capital return at 60% of free cash flow against a stated floor of 50%. Third quarter revenue guidance came in at $108 billion, plus or minus 2%, with Vera Rubin expected to represent about 20% of data center revenue in the period and hyperscale growth reaccelerating into the fourth quarter and fiscal 2028 as Vera Rubin supply builds.