The Framework Doubled to a Trillion and the Buyback Came Back, But the Guide Finally Landed Under the Street

Results vs. Consensus

Q1 FY2027 Scorecard

Year-Over-Year Comparisons

The 1,420bp year-over-year gross margin expansion is not an operating improvement of that magnitude. Q1 FY26 carried a $4.5B charge on H20 excess inventory and purchase obligations, which depressed that quarter's margin to 60.8%. Against the more meaningful sequential comparison, margin was flat.

Quarter-Over-Quarter Comparisons

Quality of Beat

Revenue. The headline growth rate understates the quarter. NVIDIA grew 85% year-over-year while lapping $4.6B of China Hopper revenue that did not repeat, which means the ex-China comparison is materially stronger than the reported one. The sequential addition of $13.5B is the largest in company history in absolute terms, and it arrived on a base that had already grown 20% the prior quarter. Data Center at $75.2B is now 92.2% of the company.

Margins. Flat at 75.0% is the correct outcome and the least interesting line in the release. Management reaffirmed mid-seventies for the full year and did so with the Vera Rubin transition in view, which is the third consecutive quarter it has held that framework. The margin debate that dominated our initiation is now closed for practical purposes, and the risk has migrated to a transition quarter in the second half rather than to an input-cost thesis.

EPS. Free cash flow of $48.6B was 59.5% of revenue, up from 51.2% last quarter, helped by lower cash taxes that management explicitly flagged will reverse: it expects a substantial increase in cash taxes in Q2 for estimated federal and state payments. Days sales outstanding improved to 45 days from 51 on favourable collection timing, and management guided that back to the mid-fifties next quarter. Both are timing effects and neither should be extrapolated.

Segment Performance

Revenue by Market Platform (new framework)

Legacy Data Center Sub-Markets (final disclosure)

Hyperscale: $37.9B, 50.3% of Data Center

Hyperscale grew 115% year-over-year and 12% sequentially, remaining at approximately half of Data Center revenue. The named deployments are unusually concrete this quarter: Microsoft's newest facility, described as the world's most powerful AI data center, is live ahead of schedule on hundreds of thousands of Blackwell GPUs; AWS will add more than a million Blackwell and Rubin GPUs starting this year and is collaborating on Spectrum networking; and Google will offer Blackwell in the cloud including confidential computing capability.

Assessment: A 12% sequential growth rate in the segment everyone worries about is the quietly reassuring number. It is slower than ACIE by design rather than by weakness, and management put hyperscale capital expenditure itself at a trillion dollars this year and expects it to grow from there toward the $3T to $4T end-of-decade framework.

ACIE: $37.4B, Growing 31% Sequentially

The AI Clouds, Industrial and Enterprise cohort grew 74% year-over-year and 31% sequentially, with AI cloud revenue alone more than tripling year-over-year. Sovereign revenue increased more than 80% year-over-year, and NVIDIA infrastructure is now deployed across nearly 40 countries representing $50 trillion of GDP. The number of partner data centers exceeding 10 megawatts has nearly doubled in a year to more than 80 sites.

Asked directly where neoclouds sit and whether ACIE will overtake Hyperscale, management placed them firmly in ACIE and gave a clear directional answer.

"I expect the second category to still grow faster, but both are gonna grow incredibly fast."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: This is the segment disclosure doing exactly the work it was designed to do. ACIE sits $492M below Hyperscale and is growing at more than twice the sequential rate, which puts the crossover in the next quarter on simple arithmetic. The customer-concentration bear case, one of three concerns circulating around the February sell-off, is being retired by the company's own reporting rather than by argument. Management's structural point is also the right one: these buyers want to purchase and operate systems rather than design them, which is a segment semi-custom silicon does not address at all.

Networking Within Data Center: $14.8B, Up 199% Year-Over-Year

In its final disclosure under the legacy split, networking grew 199% year-over-year and 35% sequentially. Two claims stand out. Spectrum-X, the end-to-end Ethernet platform, is now described as larger than all Ethernet networking peers combined. InfiniBand grew more than four times year-over-year on next-generation XDR deployments, which is notable because the consensus view has been that Ethernet displaces InfiniBand rather than both growing.

Assessment: Networking is now folded into the Data Center platform and will be harder to track, which is a real loss of visibility on the line item that has outgrown Data Center compute in each of the three quarters we have covered. We will continue to model it from the legacy trend and from management's qualitative commentary, and we would welcome its reinstatement.

Edge Computing: $6.4B, With a Memory-Price Drag

Edge Computing grew 29% year-over-year and 10% sequentially, driven by robust Blackwell workstation demand and offset by consumer demand that fell modestly on higher memory and system prices. Physical AI, which spans platforms rather than sitting inside this segment, exceeded $9B of revenue over the trailing twelve months, up from the north-of-$6B figure disclosed for fiscal 2026. The Uber partnership will power a robotaxi fleet across nearly 30 cities on four continents by 2028.

Assessment: The consumer softness is a price effect rather than a demand effect, and it is the same memory tightness that shows up as a cost line elsewhere. The physical AI figure growing from above $6B to above $9B in one quarter of disclosure is the number worth tracking here, though the company has not defined the boundary precisely enough to model.

Key Topics & Management Commentary

Overall Management Tone: The most forthcoming call of the three we have covered. Management volunteered a doubled visibility framework, an explicit capital-return payout target, a segment restatement with nine quarters of history, and a candid characterization of its newest acquired technology as a niche product, which is not the kind of thing a promotional management team says. Answers were specific and quantified in every area where the prior two calls were thematic. The one place the tone outran the disclosure was the Q2 guide, which was presented alongside expansive multi-year framing without acknowledging that it landed below the Street.

1. The $500 Billion Framework Becomes $1 Trillion

The cumulative revenue framework has been doubled in size and extended by a year. What was $500B of Blackwell and Rubin revenue from the start of calendar 2025 through the end of calendar 2026 is now $1 trillion from calendar 2025 through calendar 2027, and management attached the word confidence to it.

"Giving us full confidence in $1 trillion in Blackwell and Rubin revenue we foresee from 2025 through calendar 2027."
— Colette Kress, Executive Vice President and Chief Financial Officer

Asked what sits above that trillion, management named three sources in order: continued share gains in frontier AI models, standalone Vera CPU revenue which is explicitly excluded from the framework, and the LPX decode accelerator.

Assessment: A framework that doubles in six months is either a company discovering demand faster than it can describe it, or a company inflating a number that cannot be missed in any single quarter. The evidence favours the first reading: the framework is now backed by $145B of secured supply including prepaids, which is a cash commitment rather than a slide. The exclusion of Vera CPU from the trillion is the analytically useful detail, because it means the framework and the CPU TAM are additive rather than overlapping.

2. Capital Allocation Answered, With a Number

We wrote in February that NVIDIA could end the capital-allocation debate with a single disclosure and had chosen not to. It has now done so comprehensively. The company returned a record amount in the quarter, which it described as approximately $20.0B, comprising $19,312M of repurchases and $243M of dividends, raised the quarterly dividend from $0.01 to $0.25 per share, and added an $80.0B repurchase authorization on top of the $38.5B remaining. Most importantly, it gave a payout framework.

"As we indicated at GTC, we plan to return roughly 50% of free cash flow to shareholders this year."
— Colette Kress, Executive Vice President and Chief Financial Officer

Management framed the priority order explicitly: research and development and strategic investment first, shareholder return as the other key component. Strategic investment did not stop, with $18.6B of non-marketable securities purchased in the quarter against $649M in the year-ago quarter, but it now runs alongside a comparably sized buyback rather than in place of one.

Assessment: This resolves the concern we escalated in February. The February quarter now reads as a timing artifact of the Groq transaction rather than a strategic de-prioritization of shareholder return. A stated payout ratio is the accountability mechanism that was missing, and at roughly 50% of an approaching $200B annual free cash flow base it is a large absolute commitment. The circularity question about the investment programme itself is separate and remains open.

3. LPX: The Groq Technology, Productized and Called a Niche

The $13.0B transaction we flagged in February as sized in a single captioned but unexplained cash flow line has become a product. LPX is the low-latency decode accelerator built on the acquired SRAM-based technology, and management's characterization of it was strikingly unpromotional.

"The LPX is designed for low latency and high token rate. Its throughput is low, Size capacity is low."
— Jensen Huang, Founder and Chief Executive Officer

Management went further, saying the use case is not broad, that the number of customers is not significant, and that LPX and similar SRAM-based decode-focused accelerators will be a niche product for some time. It positioned LPX as an attach sold alongside a Grace Blackwell or Vera Rubin rack to providers who already offer a high-token-rate premium service, rather than as a standalone platform. On whether such premium tokens represent 20% of the market, management said it is a lot less than that today and could reach 20% someday.

Assessment: Two quarters ago we said the generous reading was a Mellanox-style fold-in and the uncomfortable reading was a defensive purchase. Management has now given the answer, and it is closer to the uncomfortable one: NVIDIA paid $13.0B for a technology it publicly describes as niche, with an insignificant customer count and limited breadth. That is rational as insurance and as talent acquisition, and it is not a growth driver. We are marking the custom-silicon risk down rather than up, because the honest characterization is more reassuring than a promotional one would have been. But we would not credit LPX in the model.

4. Vera CPU Opens a $200 Billion TAM That Sits Outside the Framework

The most genuinely new element of the quarter is a standalone CPU business. Vera, built on custom Arm cores and co-designed with Rubin GPUs and NVLink, is claimed to deliver up to 1.5 times the performance per core, twice the performance per watt, and four times the density per rack against x86 alternatives. Management sized the opportunity and the near-term visibility.

"Vera opens a brand new $200 billion TAM for NVIDIA, a market we have never addressed before."
— Jensen Huang, Founder and Chief Executive Officer

Asked whether the $20B of visible CPU revenue this year is standalone or bundled inside Vera Rubin, management was unambiguous.

"The 20 billion is for standalone CPU."
— Jensen Huang, Founder and Chief Executive Officer

The argument for why CPU demand rises rather than falls in an agentic world is the strongest part of it: an agent runs on a harness that handles input and output, orchestration, memory management and tool use, and all of that runs on CPU while the reasoning runs on GPU. Management's framing was that the economics of cloud computing used to be dollars per core, and agents do not rent cores.

Assessment: This is a real and under-modelled line. $20B of standalone CPU revenue visibility, explicitly excluded from the $1 trillion framework, is roughly 5% of our fiscal 2027 revenue estimate arriving from a market NVIDIA did not previously address. The claim that it will be supply-constrained for the entire life of Vera Rubin is the tell that demand is not the binding issue. We are treating the $200B TAM as directional and the $20B as modellable.

5. Rental Pricing Rising on Six-Year-Old Silicon

Management disclosed a datapoint we have not seen it give before, and it is the single best piece of evidence in the release for the durability of the demand cycle: the price of renting an H100 has risen 20% year to date, and A100 cloud pricing is up nearly 15%.

"The price of renting an H100 has risen 20% year to date, while A100 cloud pricing is up nearly 15%."
— Colette Kress, Executive Vice President and Chief Financial Officer

The implication management drew is that customers are generating profitable revenue beyond the depreciable life of their GPUs, which in turn is what makes AI infrastructure financeable by the ecosystem.

Assessment: This is the most useful disclosure on the call for anyone worried about a depreciation-driven unwind. Rising rental prices on prior-generation silicon means the installed base is appreciating in earning power rather than obsolescing, which directly contradicts the bear argument that GPU economics collapse as newer generations arrive. It also quietly supports the residual values that underpin third-party AI infrastructure financing. We would like this metric disclosed every quarter.

6. The New Segmentation and What It Is For

Asked what drove the change, management framed it as a response to the diversity of the business rather than as an investor-relations exercise: AI workloads differ by industry, applications differ, deployment locations differ from hyperscale cloud to on-premises industrial to supercomputing to the edge, and governance differs across public cloud, confidential computing and national security constraints.

The competitive point embedded in the answer is the important one. Management argued that the ACIE cohort contains a whole category of data centers where semi-custom silicon does not apply, because these buyers want to purchase and operate systems rather than design and build them.

Assessment: The framework is a better description of the business and it is also a competitive argument delivered as an accounting change. Splitting out a $37.4B cohort growing 31% sequentially, in which custom accelerators have essentially no presence, is a more effective rebuttal to the ASIC thesis than any performance benchmark. The cost is comparability and the loss of the standalone networking line.

7. Frontier Model Share Is Growing, With Anthropic the Swing

Management stated plainly that its share of frontier AI compute is increasing, and attributed the change primarily to Anthropic moving from essentially no NVIDIA coverage to a deep partnership with capacity being secured across AWS, Azure, CoreWeave and others. The named roster now includes Anthropic, OpenAI, Gemini, xAI, Meta, Microsoft AI, Thinking Machines, Reflection, Perplexity and Cursor.

"Well, we are growing share in inference. And we are growing share in inference very, very quickly."
— Jensen Huang, Founder and Chief Executive Officer

On the next architecture, management made a specific comparative claim: every frontier model company will adopt Vera Rubin from the start, which was not true of Blackwell.

Assessment: This directly addresses the concern we raised in February, when a rival architecture won Anthropic capacity and we downgraded the frontier-roster pillar. The correct reading now is that both things are true: Anthropic buys everything available, and NVIDIA's coverage of it went from near zero to substantial. Share of a growing pie is the right metric and management says it is rising. We are restoring the pillar while noting that the claim is management's own and not independently verifiable.

8. Supply Secured at $145 Billion

Total supply, inclusive of inventory, purchase commitments and prepaids, increased to $145B. The narrower supply-related commitments line reached $119.0B from $95.2B, inventory rose to $25.8B from $21.4B, and multi-year cloud service commitments rose to $30.0B from $27.0B. Management acknowledged it is not immune to supply challenges while expressing confidence in its ability to support growth.

Assessment: The narrow supply-commitments line has grown from $50.3B to $119.0B in two quarters, and total secured supply including prepaids now stands at $145B, while quarterly revenue grew from $57.0B to $81.6B over the same span. Commitments are compounding faster than revenue, which remains the correct posture given the demand signal and the correct thing to monitor if it ever turns. The mitigating evidence continues to improve rather than deteriorate: days sales outstanding fell to 45 days, and management pre-announced its reversion to the mid-fifties rather than letting investors discover it next quarter.

9. Vera Rubin Ramps From Q3, With the Slope Still Undefined

Production shipments of Vera Rubin commence in the second half starting in Q3. The platform integrates seven purpose-built chips across five accelerated racks and is claimed to deliver up to 35 times higher inference throughput and up to ten times greater AI factory revenue against Blackwell. Google will be an early adopter with bare-metal instances supporting up to 960,000 Rubin GPUs across multiple sites.

Asked whether to expect a ramp slope comparable to GB300, which management described as the fastest product ramp in company history, the answer was hedged in a specific and informative way.

"We will start in Q3... we have demand already planned We have got POs. We have got almost all of our major customers ready to go."
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: Purchase orders in hand from almost all major customers is a stronger statement than the ramp-slope hedge is a weak one. The constraint management describes is manufacturing complexity and system integration timing rather than demand. For modelling purposes this means the second half is a supply question, and it means our fiscal 2027 estimate carries more timing risk than level risk.

10. Growing Faster Than Hyperscaler Capital Expenditure, By Design

Asked whether the company should be expected to grow faster than hyperscale capital expenditure, management gave the clearest strategic statement of the call and grounded it in the new segmentation.

"So first of all, we should be growing faster than hyperscale CapEx."
— Jensen Huang, Founder and Chief Executive Officer

The mechanism is that hyperscale capital expenditure, which management put at a trillion dollars this year, drives only about half of the Data Center business, while the ACIE half is driven by a different and far more fragmented buyer base of AI native clouds, sovereigns, and the roughly 250,000 enterprises management referenced.

Assessment: This is the answer to the question that was asked and dodged in February, when management would not say whether NVIDIA could grow into a flat hyperscaler envelope. The segmentation now provides the arithmetic for the answer. It is a genuinely stronger position than the one the company was arguing from two quarters ago.

11. China Remains Zero, and Now Laps a $4.6 Billion Comparative

The United States government has approved licenses for H200 shipments to China-based customers, and NVIDIA has generated no revenue from them, with import permission still unresolved. No Data Center Hopper products shipped to China in the quarter, against $4.6B in the year-ago quarter. The Q2 outlook again assumes zero.

Assessment: The situation is unchanged from February and the significance has shifted. The company has now fully lapped the last material China quarter, which means from Q2 onward the year-over-year comparisons no longer carry a China drag at all. That is a modest arithmetic tailwind to reported growth rates that has nothing to do with the business improving.

Guidance & Outlook

The $91.0B guide implies 11.5% sequential growth and 95% year-over-year growth against the $46,743M printed in Q2 FY26. It is the largest absolute forward number the company has ever guided to and it is the first in this coverage series to sit below consensus. Two smaller changes accompanied it: the full-year tax rate range was lowered to 16.0% to 18.0% from 17.0% to 19.0% on geographic mix, and full-year non-GAAP operating expense growth was raised to the upper forties from the low forties, attributed to higher research and development and to accelerating internal use of AI tools.

Implied Q-over-Q ramp: Sequential growth of 11.5% decelerates from the 19.8% just delivered. That is unsurprising on a base that has grown for fourteen consecutive quarters, and it embeds the Vera Rubin ramp starting only in Q3. If sequential growth holds near 10% through the second half as Rubin contributes, fiscal 2027 revenue lands near $385B.

Street at: Consensus of 41 analysts was $92.07B, so the guide sits $1.07B below it. This is the number that explains the share-price reaction and it deserves to be stated plainly rather than explained away.

Guidance style: The pattern has changed and investors should register it. NVIDIA guided 5.4% above the Street in November, 7.4% above in February, and 1.2% below in May. Nothing in the release suggests a demand problem, and management simultaneously doubled its multi-year framework, which is not the behaviour of a company seeing softness. The more likely explanation is that the Street has finally marked its models up to the point where it is extrapolating faster than management will commit, combined with a second half whose shape depends on a product transition management has repeatedly declined to size. Either way, the era of automatic guidance beats is over, and that is a meaningful change in how the stock will trade.

Analyst Q&A Highlights

What Drove the Change in Segmentation

The opening question asked for the philosophy behind the new reporting framework, whether there are competitive differences between the segments, and how the surprising CPU figure maps across them.

Q: "I guess I would like to ask what drove the change in segmentation? what is the philosophy behind giving us the numbers that way?"
— Joseph Moore, Morgan Stanley

A: "it is the simplest way of understanding our business Each 1 of them have different stacks in a lot of ways. They have different operating systems. They operate in a different way. We go to market very differently in each 1 of them."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: A substantive answer with a competitive argument embedded in it. The distinction management drew is that the hyperscale go-to-market involves five or six counterparties while the second category involves hundreds of thousands of companies, which requires the full acceleration library portfolio to address. The unstated implication, which the answer makes without asserting it, is that a competitor could win a hyperscale socket and still be structurally unable to serve half the Data Center business.

Whether NVIDIA Should Grow Faster Than Hyperscaler Capital Expenditure

A question on growth philosophy noted that Data Center excluding China grew roughly 120% in the quarter while hyperscaler capital expenditure is forecast to grow perhaps 90% to 100% this year, and asked whether outgrowing that envelope is the goal.

Q: "I was just wondering the goal for the company to grow faster than hyperscaler CapEx? Do are you still comfortable in kind of endorsing that view?"
— Ben Reitzes, Melius Research

A: "So first of all, we should be growing faster than hyperscale CapEx."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The most consequential answer of the call, and a direct improvement on February when the equivalent question was answered only for the optimistic case. Management now has a structural argument rather than an assertion: roughly half the Data Center business is not driven by hyperscale capital expenditure at all. This is the answer we said was missing two quarters ago.

Where Neoclouds Sit and Whether ACIE Overtakes Hyperscale

A clarifying question asked which segment the neocloud operators fall into and whether the faster-growing cohort should be expected to overtake the other.

Q: "Where do you put the neo clouds across those 2 segments? Are they in hyperscale, or are they in the AI cloud?"
— Stacy Rasgon, Bernstein Research

A: "I expect the second category to still grow faster, but both are gonna grow incredibly fast."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Answered fully and usefully. The reasoning matters as much as the answer: management argued that AI native clouds need an architecture with broad offtake because their business is renting compute, and characterized NVIDIA's architecture as the most rentable in the world. That is the offtake argument we identified at initiation as the strongest and least-discussed part of the moat, now made explicit by management and attached to a named, sized, fast-growing segment.

Traction With the Acquired Decode Technology

A question probed the traction of the newer workload-specific products, including the decode accelerator built on the technology licensed in the prior quarter, and how they fit the broader platform strategy.

Q: "I wanted to ask about the traction you are getting with some of these custom merchant things you are doing, stuff like, you know, CPX and LPX."
— Timothy Arcuri, UBS

A: "The LPX is designed for low latency and high token rate. Its throughput is low, Size capacity is low."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: A notably candid answer that works against the company's own narrative interest, which is why it is credible. Management described the LPX use case as not broad, the customer count as not significant, and the category as a niche for some time to come. Set against a $13.0B outlay two quarters ago, this is management telling investors the purchase was not about a new revenue line. We read it as insurance and talent, correctly priced or not, and we do not model it.

Whether the CPU Opportunity Is Incremental or Cannibalistic

A two-part question asked whether CPU demand for agentic applications is incremental to GPU demand or cannibalizes it, and whether the $20B CPU figure is standalone or already bundled inside Vera Rubin systems.

Q: "the $20 billion number that you gave, is that for stand alone Vera CPUs, or is that kind of already included in that Vera as part of Vera Rubin?"
— Vivek Arya, Bank of America Securities

A: "The 20 billion is for standalone CPU."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Precisely the clarification the model needed, answered in one sentence. The accompanying argument for incrementality is coherent: agent harnesses handle orchestration, memory management and tool use on CPU while reasoning runs on GPU, so a proliferation of agents raises both. Management expects to be supply-constrained on Vera across the entire life of the Vera Rubin generation, which is a demand statement dressed as a supply one.

Vera Rubin's Effect on Inference Share

A question asked how the next architecture and its co-engineering affect inference market share heading into late calendar 2026 and 2027, given investor focus on that specific battleground.

Q: "how do you see Vera Rubin in your extreme co engineering impacting your share of the inference market, you know, as we look into late 26, 27?"
— Christopher Muse, Cantor Fitzgerald

A: "Every single frontier model company will jump on VeraRubin from the get go. And that was not true before on Blackwell."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: A specific, falsifiable comparative claim rather than a general assertion, which makes it useful. Management grounded the inference share gain in two mechanisms: a larger number of frontier model companies exists than a year ago, and NVIDIA's coverage of Anthropic went from near zero to substantial. It also made the sharper point that in the ACIE segment, essentially all inference runs on NVIDIA, which is where the share arithmetic actually lives.

What Sits Above the Trillion-Dollar Framework

A question noted that the trillion-dollar visibility figure discussed at the company's developer conference excluded several newer product lines, and asked which would be the largest source of upside above it.

Q: "Could you maybe give us a sense about whether the Vera CPUs are going to be the biggest source of upside above and beyond that $1 trillion"
— James Schneider, Goldman Sachs

A: "we did not include any Verus CPU, standalone CPU in that in that number And so I expect that to be the second largest"
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Management ranked the upside sources: frontier model share gains first, standalone Vera CPU second, LPX third. That ordering is more informative than the trillion-dollar headline, because it says the largest incremental opportunity is share within a market NVIDIA already serves rather than a new product line. It also confirms the framework and the CPU TAM are additive, which is the piece the model needed.

What They're NOT Saying

1. Why the Q2 guide sits below consensus: Management guided $91.0B against a $92.07B Street number and did not acknowledge the gap, address it, or characterize the guide as conservative. In a quarter where it doubled a multi-year framework, declining to reconcile the near-term number with the long-term one is a conspicuous omission.
2. The slope of the Vera Rubin ramp: Asked for a third consecutive quarter, deferred for a third consecutive quarter. Purchase orders are in hand from almost all major customers and management still will not size the second-half contribution, which leaves the largest swing factor in the fiscal year unquantified.
3. Any ceiling on the strategic investment programme: $18.6B of non-marketable securities purchased this quarter against $649M a year ago. The capital-return side is now framed at roughly 50% of free cash flow; the investment side still has no stated limit, no return threshold, and no disclosure of what share of revenue comes from invested-in entities.
4. Standalone networking revenue going forward: The legacy split was disclosed one final time at $14.8B and 199% growth. Folding the fastest-growing line in the business into a broader platform removes the ability to track it, and no replacement disclosure was offered.
5. What the $145B of secured supply converts to and when: The figure now includes prepaids, which is a change in composition that was not reconciled against the prior $95.2B commitments number. The two are not directly comparable and management did not say so.
6. Why full-year operating expense growth was raised to the upper forties: The attribution was higher research and development and accelerating internal AI tool usage. A roughly five-point increase in a full-year expense growth guide, given one quarter after the prior guide, deserved more than a clause.
7. Whether the H100 and A100 rental price increases are broad or selective: The most valuable datapoint on the call was given without a source, a sample, or a geography. If prior-generation rental pricing is genuinely rising 15% to 20%, that deserves a standing disclosure rather than a single sentence.

Market Reaction

• Pre-print setup: NVDA closed at $223.47 on May 20, up 19.8% year-to-date, up 10.6% over the trailing thirty days, and up 66.3% over the trailing twelve months. The stock entered the print roughly 5% below its 52-week closing high of $235.74, against a 52-week closing low of $131.29.
• After-hours move: Shares traded modestly lower in the extended session following the release and the call.
• Next-day session: The stock opened at $222.29, down 0.5%, traded between $217.93 and $227.40, and closed at $219.51, down 1.8% or $3.96 from the pre-print close.
• Volume: 203.4 million shares against a 30-day average of 151.9 million, a 1.3x multiple and the lightest of the three post-print sessions we have examined, against 1.8x and 2.1x for the prior two.
• Market context: The S&P 500 rose 0.2% in the same session, so the decline is company-specific but small.

For the third consecutive quarter NVDA beat and the stock fell, but this session is categorically different from the prior two and should not be filed alongside them. In November the stock round-tripped eight points on a beat-and-raise while the index fell 1.6%. In February it fell 5.5% on a beat-and-raise while the index fell 0.5%. Here it fell 1.8% on a beat and a guide that came in 1.2% below the Street, on the lightest relative volume of the three.

This one has an explanation that does not require a sentiment story. A 1.8% decline against a 1.2% guidance shortfall is close to a mechanical response. The stock had also been marked up going in, rising 10.6% over the prior thirty days against just 2.1% into the November print, so expectations were considerably better reflected in the price. There is no evidence in this tape of the buyer exhaustion we identified at initiation, nor of the cluster of concerns that accompanied February's decline.

What we would watch instead. The meaningful change is not the price action but the guidance relationship behind it. Consensus has now caught up with and briefly overtaken what management is willing to commit to, which removes the automatic upside surprise that has supported the stock through this entire coverage period. That does not weaken the investment case at 25 times forward earnings, and it does change the mechanism by which the stock is likely to work: from beating a low bar to compounding into a fair one.

Street Perspective

Debate: Does a Guide Below Consensus Signal Demand Deceleration?

Bull view: Nothing else in the release supports a demand reading. Management doubled its multi-year framework to $1 trillion, secured $145B of supply, committed to sequential growth in every remaining quarter of calendar 2026 after delivering a fourteenth straight, and said it should outgrow hyperscale capital expenditure. A guide 1.2% under a raised bar is arithmetic, not signal.

Bear view: Management guides what it can supply, and the second half depends on a product transition it has declined to size for three consecutive quarters. A guide below the Street from a company that has beaten every quarter is precisely how a deceleration first becomes visible.

Our take: The bull has the better of it on the evidence available, with one caveat that matters. The framework doubling and the supply commitments are hard, cash-backed disclosures that no management team makes into softening demand. But the bear is right that the second-half shape is unknown, and the honest position is that Q2 is a supply-constrained guide whose conservatism cannot be verified until the Rubin ramp is visible.

Debate: Is the ACIE Segment a Genuine Diversification or a Reclassification?

Bull view: ACIE grew 31% sequentially against Hyperscale's 12% and sits within $500M of overtaking it, with AI cloud revenue more than tripling year-over-year, sovereign revenue up more than 80%, and partner data centers above 10 megawatts nearly doubling to more than 80 sites. Those are independent operating metrics, not an accounting artifact.

Bear view: Neoclouds are classified into ACIE while their end demand often originates from the same hyperscalers and frontier labs counted in the other bucket, so the split overstates true customer diversification. The framework changed in the quarter after concentration became the market's stated concern.

Our take: Both are partly right and the bull is more right. The pass-through critique is real: a hyperscaler renting capacity from a neocloud appears in ACIE. But sovereign revenue, on-premises industrial and enterprise deployments are genuinely independent demand, and the 80-plus partner data center count is a physical fact rather than a classification. We treat ACIE as real diversification with a discount applied, and we note that the timing of the disclosure is convenient without being dishonest.

Debate: Was the $13 Billion Decode Acquisition Money Well Spent?

Bull view: NVIDIA acquired the leading low-latency decode team and technology for roughly a quarter of one period's free cash flow, denied it to competitors, and can now attach it to racks for customers running premium high-token-rate services. Mellanox looked expensive at the time too.

Bear view: Management has now described the resulting product as niche, with limited throughput, limited context capacity, an insignificant customer count, and a use case that is not broad. That is $13.0B for an insurance policy, and shareholders learned the price before they learned the rationale.

Our take: The bear wins on the economics and the bull wins on the strategy. Paying $13.0B to remove an argument from the market and to acquire a decode team is defensible for a company generating $48.6B of quarterly free cash flow, and it is not a growth investment. The genuinely positive development is that management said so plainly rather than promoting LPX as a new revenue line, which is the kind of candour that earns credibility for the harder disclosures still outstanding.

Model Update & Valuation Framework

Valuation. At the May 21 close of $219.51, NVDA trades at approximately 24.8x our fiscal 2027 non-GAAP EPS estimate of $8.85, on the stock-based-compensation-inclusive basis the company now reports. That is essentially unchanged from the 23.5x at which we last marked it, because the share price and our estimate rose in similar proportion. Over the three quarters we have covered the stock, the share price has risen 21.5% from $180.64 while our forward EPS estimate has risen 26.4% from $7.00, which means the multiple has compressed slightly even as the stock has appreciated.

Fair value range: $250 to $300, raised from $220 to $270. The range reflects 28x to 34x our fiscal 2027 estimate of $8.85. The midpoint of $275 implies approximately 25% upside from $219.51. We have held the multiple range unchanged and raised the absolute range entirely on the estimate revision, which is the discipline we would want applied to us.

What would change our view. A second consecutive guide below consensus, particularly if unaccompanied by a supply explanation. Evidence that the Vera Rubin ramp is slipping out of Q3. Any softening of the mid-seventies full-year gross margin framework as the transition begins. A reversal of the capital-return commitment below the stated roughly 50% of free cash flow. On the upside, a Q2 guide-and-beat that restores the pattern, first disclosure of Vera Rubin second-half contribution, or standalone Vera CPU revenue tracking meaningfully above the $20B visibility figure.

Thesis Scorecard Post-Earnings

Overall: Thesis strengthened materially. Three of the six bear points were de-escalated this quarter, two of them because management made specific disclosures we had said were missing: a capital-return framework and an honest characterization of the acquired decode technology. A sixth bull pillar was added in standalone Vera CPU. The single remaining unaddressed disclosure gap is a ceiling on strategic investment, which is now the only item carried forward unchanged from February.

Action: Buy. This was the most substantively disclosed quarter of our coverage and it resolved more open questions than it created. At roughly 25x a fiscal 2027 estimate we have raised twice in three quarters, against 78% modelled revenue growth, a reaffirmed 75% gross margin and a stated commitment to return half of free cash flow, the risk-reward remains favourable. We maintain Outperform, raise the fair value range to $250 to $300, and replace capital-allocation disclosure at the top of our watch list with the guide-versus-consensus relationship, which changed this quarter in a way that will alter how the stock trades even if it does not alter what the business earns.