Nvidia Reports $81.6 Billion Quarter — Its Best Ever — and Guides to $91 Billion. Its Stock Slid Anyway.

Nvidia on Wednesday afternoon reported its best quarter in company history, posting $81.6 billion in revenue for the three months ended April 26 — a result that beat Wall Street’s consensus estimate by roughly $2.4 billion, extended the chipmaker’s streak of record-breaking performance to five consecutive quarters, and came with a guidance figure of $91 billion for the current quarter. The stock fell Thursday anyway.

Shares of Nvidia declined about 1.5 percent in Thursday’s session, extending a pattern that has characterized every quarterly report the company has filed since late 2025: earnings beat, stock slides. The broader market also fell — the S&P 500 dropped 0.45 percent and the Nasdaq declined 0.50 percent — so some of the pressure reflects the rising Treasury yield environment that has weighed on technology valuations generally. But Nvidia’s retreat after a blowout beat has its own logic, which will be addressed below.

The Numbers

Total revenue of $81.6 billion rose 85 percent from a year earlier and 20 percent from the prior quarter, the third consecutive quarter of double-digit sequential acceleration. Every major segment beat its internal forecast.

The Data Center division, which sells the GPU systems that train and run artificial intelligence models, generated $75.25 billion — up 92 percent from the same quarter a year ago and accounting for 92 percent of total revenue. That single division is now generating more quarterly revenue than some major technology companies generate in an entire year. CFO Colette Kress, in the accompanying commentary filed with the Securities and Exchange Commission, noted that cloud rental rates for Nvidia’s H100 processor had risen 20 percent year-to-date, with rates for the older A100 architecture up roughly 15 percent — a signal that demand is outpacing supply even as production ramps.

Earnings per diluted share came in at $1.87 on a non-GAAP basis, above the consensus estimate of $1.78, and $2.39 on a GAAP basis, which includes stock compensation expenses. Gross margin held at 74.9 percent GAAP and 75.0 percent non-GAAP, roughly flat with the prior quarter and up from 60.8 percent a year earlier. Operating income reached $53.5 billion, a 147 percent increase from the year-ago quarter.

In a move that underscores the scale of the capital the company is generating, Nvidia announced an $80 billion additional share repurchase authorization and raised its quarterly cash dividend from one cent per share to twenty-five cents per share — a 25-fold increase.

Blackwell and the AI Factory

The driving product is the Blackwell 300 architecture, Nvidia’s current-generation GPU platform, which has been adopted across what Jensen Huang, Nvidia’s co-founder and CEO, describes as “AI factories” — the large-scale computing facilities being built by hyperscaler cloud providers, sovereign government programs, healthcare systems, and enterprise corporations to run AI workloads.

On the earnings call, Huang framed the moment with characteristic directness: “Demand has gone parabolic. The reason is simple: Agentic AI has arrived.”

Agentic AI — systems that can plan, reason, and execute multi-step tasks autonomously rather than just responding to individual prompts — requires substantially more computing infrastructure than earlier AI deployments. Huang said that agentic workloads use GPUs for the high-intensity inferencing work (what he called “the thinking”) and CPUs for the orchestration, memory management, and input-output tasks that surround it. This framing is also a product pitch: Nvidia last month unveiled the Vera CPU, designed specifically for the orchestration role in agentic systems, positioning the company to capture revenue from both the GPU and CPU components of the AI stack.

CFO Kress told analysts that the build-out of AI factories is “accelerating at extraordinary speed” and that the Blackwell platform has been adopted by major hyperscale cloud providers, frontier AI model developers, and sovereign customers — a category that includes national governments building AI infrastructure for defense, intelligence, and civil applications.

Jensen Huang was among the technology executives who traveled with the U.S. delegation to Beijing earlier this month, where AI chip export licenses were among the topics discussed at the margins of the Trump-Xi summit. The current licensing framework limits Nvidia’s ability to sell its most advanced processors to Chinese customers, a constraint that has diverted significant demand to chip architectures designed to meet the restrictions. How those rules evolve will materially affect Nvidia’s addressable market in its second-largest region.

The Stock Reaction Pattern

The decline on Thursday marks the fourth consecutive quarter in which Nvidia’s stock fell after reporting better-than-expected results. In the fiscal fourth quarter of 2025, reported in February 2026, the stock fell roughly 5 percent. The prior two quarters produced post-earnings declines of about 3 percent and 0.8 percent, respectively.

The conventional explanation for selling a company on good earnings is that the results — however strong in absolute terms — failed to exceed what investors had already priced in. Nvidia’s stock trades at a forward price-to-earnings multiple that implies sustained, dramatic growth for years, meaning each quarter’s actual result must clear an escalating expectations bar, not merely the published analyst consensus.

The specific friction in this quarter appears to be the Q2 guidance. The $91 billion midpoint beat the Wall Street consensus of approximately $86.8 billion by a substantial margin, but some analysts had predicted the company could issue guidance as high as $93-95 billion given the Blackwell demand signals and the agentic AI buildout commentary. The $91 billion figure — strong as it is — fell short of those upper-end projections.

The upstream context matters here. TSMC, which fabricates Nvidia’s processors, has noted that AI chip demand has pushed power delivery requirements to the edge of what current data center electrical infrastructure can support. Physical constraints on how quickly power can be delivered to large server clusters impose a ceiling on how fast AI factories can be brought online regardless of how much demand exists, and that ceiling ultimately limits how quickly Nvidia’s revenue can scale in any single quarter.

Why the Buildout Continues

One quarter of slower-than-expected guidance does not change the structural case for Nvidia’s dominant position in AI infrastructure. The company is selling the primary tool for one of the largest capital expenditure cycles in the history of the technology industry.

The scale is worth stating plainly. If Nvidia hits its $91 billion Q2 guidance, the company will have generated roughly $345 billion in annualized revenue — more than Apple’s total revenue for a full year — with the vast majority coming from data center hardware. No semiconductor company in history has grown to this scale this quickly.

The shift to agentic AI applications reinforces the demand trajectory. A simple prompt-response AI interaction requires substantial GPU compute for the inferencing step. An agentic system that autonomously plans a five-step task, calls external tools, retrieves memory, and iterates based on results requires several times that compute per user session. As AI applications migrate from novelty assistants toward operational enterprise tools, per-session compute consumption grows, and the infrastructure required to support it grows with it.

OpenAI’s recent consolidation of its video products after scrapping the Sora-Disney partnership is a reminder that individual AI ventures can stumble. But at the infrastructure level, the demand for training and inference compute has continued expanding regardless of which applications succeed or fail. Nvidia is selling the picks and shovels to every miner in the field.

What Comes Next

Nvidia’s fiscal second quarter ends July 26, with results reportable in mid-August 2026. The $91 billion guidance represents the baseline. Upward revisions are possible as the Blackwell 300 ramp continues and as enterprise AI deployments move from pilot programs to production workloads.

The competitive landscape will get more crowded. Advanced Micro Devices is ramping its MI350 GPU architecture, targeted at the data center market. Intel’s Gaudi 3 accelerator is positioned at cost-sensitive customers. Hyperscalers including Google, Amazon, and Microsoft are all investing in custom silicon designed to reduce their dependence on Nvidia for at least some workloads. None of these alternatives has displaced Nvidia’s share in the current cycle, but the architectural window for competition widens each year.

The more immediate unknown is export policy. The Trump administration’s approach to AI chip sales to China — including the licensing discussions at the Beijing summit — will significantly affect where Nvidia can sell its most capable processors in the second half of 2026. The current market is already pricing a licensing regime that restricts Chinese access to Blackwell-class chips. Any expansion or relaxation of that framework would have a material effect on the demand picture for the back half of the fiscal year.

In the near term, Nvidia’s challenge is less about demand and more about what “enough” looks like when the market expects the extraordinary as the baseline.