NextFin News - Nvidia gave investors a two-year answer to the question that has haunted the artificial-intelligence trade: the sales surge is not ending soon. Reporting fiscal second-quarter revenue of $96.2 billion, up 106% from a year ago, the chipmaker said it expects fiscal 2028 revenue to grow approximately 70% year over year — and framed the outlook as supply-constrained, with bottlenecks expected to persist at least through the end of fiscal 2028.
The quarter, which beat Wall Street estimates, was not the headline. The horizon was. Chief Financial Officer Colette Kress told analysts on the August 26 earnings call that the company's fiscal 2028 view is "a supply-constrained outlook," adding that although Nvidia will work to close the supply-demand gap, "we expect supply to remain a bottleneck at least through the end of fiscal year 2028." Chief Executive Jensen Huang was more direct about the imbalance: "Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%."
The Quarter: A Beat That Mattered Less Than the Guidance
For the quarter ended July 26, Nvidia posted revenue of $96.2 billion, up 18% from the prior quarter and more than double the $46.7 billion recorded a year earlier. Data Center revenue reached $89.0 billion, up 117% year over year and 18% sequentially. On both a GAAP and non-GAAP basis, gross margin held at 75.0%. GAAP diluted earnings per share were $2.46; non-GAAP diluted EPS were $2.22. Operating income climbed to $63.7 billion, up 124% from a year ago, and the company returned approximately $26.0 billion to shareholders through buybacks and dividends, leaving $99.0 billion under its repurchase authorization.
Wall Street had been braced for roughly $92 billion of revenue and $2.09 of adjusted earnings per share, according to a consensus of analysts tracked ahead of the print. Nvidia cleared both. The company also guided third-quarter revenue to $108.0 billion, plus or minus 2%, with gross margins of 74.0%, plus or minus 50 basis points.
The market read the longer message. Shares, which had traded around $213 earlier on August 26, rose roughly 5% in after-hours trading as investors absorbed a guidance framework that extends the AI buildout narrative well past fiscal 2027, the year that had anchored most analyst models.
Yet the most consequential detail was what Nvidia chose to do for the first time: guide a full year ahead. "It is the case that we have never forecasted or never guided to a year in advance," Huang said on the call. The break from company practice signals management believes visibility into the AI buildout is now long enough to anchor customers, suppliers, and shareholders to a shared multi-year view. It is also a disclosure choice with competitive consequences — a public, supply-capped number tells rivals exactly how much room Nvidia believes it has, and tells customers how much they need to order now to secure allocation.
The Constraint Has Flipped: Scarcity Moved From Chips to the Whole Stack
The first-order story is simple: demand exceeds supply. The second-order story matters more — the bottleneck has migrated. In 2023 and early 2024, the constraint was advanced-packaging capacity for Nvidia's own GPUs. Today it has propagated outward through the entire AI-factory stack: high-bandwidth memory, power infrastructure, rack-level integration, and the financing required to pay for it all.
Kress flagged the memory problem directly. Component costs have risen significantly, she said, and the company is "experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and is headed even higher into next year." The margin consequence is specific: Nvidia expects gross margins to bottom in the fourth quarter in the 71%–72% range before settling at 72%–73% in fiscal 2028.
That is the trade-off embedded in a 70% growth year. Nvidia is willing to absorb margin compression to keep units flowing, because in a supply-constrained market, volume share today is worth more than margin points. Every GPU that ships now locks a customer into Nvidia's software and networking stack — CUDA, NVLink, Spectrum Ethernet — for the life of the deployment. The company is effectively buying lock-in with margin.
Read that way, the 70% figure is a floor set by supply, not a ceiling set by demand. Huang's remark that demand is "much greater than 70%" implies the company could grow faster if memory, packaging, and power caught up. The risk for competitors is not that Nvidia's demand fades; it is that Nvidia's supply chain improves and takes more share before anyone else can scale.
The financing dimension of the constraint is easy to miss. Building an AI factory is no longer just a chip purchase — it is a real-estate, power, and balance-sheet problem. In the quarter, Nvidia announced strategic partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize more than $500 billion of third-party capital for AI infrastructure "over time," subject to definitive agreements. That is not a revenue line today; it is an attempt to relax the demand-side constraint on its customers. If hyperscalers and sovereign buyers can finance more capacity, Nvidia's supply finds more buyers — and the 70% growth year becomes easier to fill.
Agentic AI Turns Compute Into an Operating Expense — and That Changes the Cycle
The deeper mechanism behind the extended outlook is a shift in how AI is consumed. Huang told analysts the industry has crossed an inflection point. "Today, the vast majority of AI is prompted by people," he said. "I believe that this last month it has crossed. Most AI are now agentic."
Those agents are running continuously. They're running in the background.
That distinction matters for the cyclical-versus-structural question. Prompted AI is episodic — a human types, a token is generated, demand rises and falls with human activity. Agentic AI runs continuously in the background. A company with roughly 40,000 employees, Huang suggested, may eventually run hundreds of thousands of agents, each consuming compute around the clock.
When compute shifts from a project budget to an always-on operating expense, the demand profile changes from lumpy capital spending to a utility-like load. That is the structural leg of the thesis, and it is why Nvidia's guidance now stretches to fiscal 2028 rather than stopping at the next quarter.
Huang framed the economics in the line that has become the company's shorthand for the new regime: "Its tokens are productive and profitable. Now, compute is revenue." Frontier labs, he said, "are generating profitable tokens. They are only limited by the amount of compute." If AI workloads are already paying for themselves, the buildout is not dependent on speculative capital — it is funded by operating cash flow, which is stickier and harder to shut off when sentiment sours.
The revenue-per-watt math supports the point. Since Hopper, Nvidia said, its revenue opportunity has grown from roughly $18 billion per gigawatt of data-center capacity to $25 billion with Blackwell and $40 billion with Vera Rubin — a measure of how much more of the data-center stack the company now captures per unit of power. Vera Rubin, it said, delivers 30 times the throughput per megawatt and 35 times lower token cost than the prior generation. Efficiency gains of that size do not kill demand for compute; they expand the set of workloads that can profitably run, which is precisely the condition under which agentic AI scales from pilot to production.
The Counter-Thesis: A Supply Constraint Is Cyclical, Not a Moat
The strongest case against the bullish read is that a supply-constrained outlook is, by definition, self-correcting. Scarcity invites capacity. Memory makers will add output, foundries will expand advanced packaging, and customers will qualify second sources. When supply catches up, the pricing power that has carried Nvidia's 75% gross margins evaporates. The bear case is not that AI demand disappears; it is that the industry overbuilds into 2028 and turns a seller's market into a buyer's market.
History gives the bears ammunition. The semiconductor industry has spent decades oscillating between shortage and glut, and the memory market Kress described is the most cyclically violent segment in technology. A 70% growth year priced on constrained supply is vulnerable the moment capacity arrives ahead of demand. The cyclical investor's rule is simple: the cure for high prices is high prices.
Nvidia's answer is vertical depth. The company is not just selling GPUs; it is selling the rack, the networking fabric, the CPU layer, the inference accelerator, and the software that binds it together. In the quarter it announced Vera Rubin is in full production, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, and Nebius. The Vera CPU — the first processor the company says is built for AI agents — is shipping to Oracle Cloud Infrastructure, SpaceXAI, and AWS. Its Groq 3 LPX inference accelerator is in full production. Beyond Vera Rubin, the company's disclosed roadmap runs through Rubin Ultra in the second half of 2027 and the next architecture, named Feynman, in 2028.
That ecosystem depth is the moat, but it is not impenetrable. The falsifying signal is concrete: if Nvidia's Data Center revenue growth decelerates below 50% year over year while gross margins fall faster than the guided 71%–72% trough — meaning supply has caught up and pricing power has broken simultaneously — the structural thesis is wrong and the cycle has turned. Watch the quarterly Data Center growth rate against the margin floor. A growth slowdown paired with stable margins would point to a supply story; a growth slowdown paired with a margin break would point to a demand story.
What the Roadmap to 2028 Actually Says
A predictable two-year cadence gives customers a reason to keep buying rather than delay purchases for the next generation — and in a market where buyers might otherwise pause, that reduces the risk of an order gap. The financing partnerships matter here too. AI factories cost billions and require power and land that most enterprises cannot secure on their own balance sheets. By helping to mobilize $500 billion of third-party capital, Nvidia is effectively expanding the pool of buyers who can afford to keep building through 2028.
There is a risk inside the roadmap, however. A disclosed cadence also gives customers a reason to wait if they believe the next generation will be meaningfully cheaper per token. Nvidia's defense is that each generation raises the revenue opportunity per gigawatt — $18 billion to $25 billion to $40 billion — so upgrading is not just cheaper, it is denser. Whether customers accelerate or defer purchases across the Vera Rubin, Rubin Ultra, and Feynman transitions will be the second signal to watch alongside the margin floor.
Conclusion: The Binding Constraint Is the Only Forecast That Matters
The base case is that Nvidia delivers the roughly 70% fiscal 2028 growth it has guided, with revenue approaching the high $200 billions as the Vera Rubin and CPU ramps compound. Margins compress toward the guided 72%–73% range, but operating profit keeps climbing on volume. The upside case is that memory supply loosens faster than expected and agentic adoption accelerates — in which case the 70% figure proves conservative, just as Huang hinted. The downside case is that hyperscaler capital spending hits an internal-return wall, or that memory and power constraints bind harder than Nvidia expects, pushing the margin trough below 71% and delaying shipments.
Short term, the stock trades on the gap between the roughly 5% after-hours pop and the memory-margin overhang. Medium term, it trades on whether fiscal 2028's 70% growth is achievable within the supply envelope. Long term, it trades on whether agentic AI converts compute into a durable utility load — the structural question that decides whether this is a cycle or a regime.
The signal to watch is not the next quarter's revenue beat. It is the quarterly Data Center growth rate against the 71%–72% margin floor. Growth above 50% with margins holding at or above the trough confirms the structural read. Growth decelerating below 50% alongside a deeper margin break confirms the cyclical one.
Nvidia's message to the market was not that AI demand will last forever. It was that demand now outruns everything the company can build through 2028 — and in a boom, the binding constraint is the only forecast that matters.
Explore more exclusive insights at nextfin.ai.
