Nvidia Reports $81.6 Billion in Revenue and Guides to $91 Billion Next Quarter

One thing doesn't add up about Nvidia's latest earnings report. The company posted the largest single quarter in semiconductor history. Revenue hit $81.6 billion. Data center revenue came in at $75.2 billion. Both figures blew past analyst expectations by billions. Guidance for the next quarter landed at $91 billion, a number that would have seemed impossible two years ago. And the stock fell 5.46% the next day.

Understanding why the stock dropped is actually the best way to understand what's happening in the AI infrastructure market right now, and what Nvidia's results reveal about the pace and composition of the buildout.

What Nvidia Actually Reported

Nvidia reported Q1 fiscal year 2027 results on May 20, 2026. Revenue reached a record $81.6 billion, up 85% from the same period last year and 20% from last quarter. The company had guided to approximately $77.5 billion. Analysts expected around $78 billion. Nvidia beat both by roughly $3.5 billion, extending a streak of consistent estimate outperformance that has defined its last several quarters.

Data center revenue was $75.2 billion, up 92% year over year. That segment now accounts for nearly 92 cents of every dollar Nvidia earns. Within data center, compute revenue came in at $60.4 billion, up 77% year over year. Networking revenue hit $14.8 billion, up 199% from the same period last year. That is not a typo. Networking revenue nearly tripled, driven by NVLink compute fabrics, InfiniBand deployments, and Spectrum-X Ethernet solutions that now ship as part of large Blackwell system builds.

The networking number deserves special attention because most analysis of Nvidia's earnings stays focused on compute revenue or the headline top-line figure. What's happening in networking reflects a structural shift in how AI infrastructure is built. Blackwell-scale deployments require NVLink interconnect fabrics that did not exist at this scale during the Hopper era. A GB200 NVLink system does not only include GPUs. It includes high-bandwidth interconnect hardware that in large configurations approaches the cost of the compute itself. Teams that previously bought individual GPU servers are now buying interconnect-dense cluster configurations that require Nvidia networking products throughout the stack.

GAAP net income was $58.3 billion, up 211% from the prior year. Free cash flow reached $48.6 billion. Operating cash flow was $50.3 billion. These numbers reflect a demand environment where the buyers of AI hardware are not managing a careful budget process. They are spending what they need to spend to stay competitive, and the cash is coming out of Nvidia's bottom line in quantities that few software companies have matched, let alone chip manufacturers.

Nvidia CEO Jensen Huang described the underlying dynamic in the earnings call, as covered in the post-earnings analysis from Sherwood News: "Computing demand is growing exponentially. The agentic AI inflection point has arrived. Grace Blackwell with NVLink is the king of inference today, delivering an order-of-magnitude lower cost per token, and Vera Rubin will extend that leadership even further."

Cost per token is now the primary competitive unit in AI infrastructure. When enterprises and cloud providers choose between GPU configurations, they compare how much it costs to run a billion tokens of inference on different hardware setups. The Blackwell 300 architecture has cut that cost enough to shift procurement decisions away from alternatives in ways that showed up clearly in Q1's customer composition.

Customer mix is one of the most important parts of the Q1 story. Large cloud providers now account for just under 50% of data center revenue, down from a higher concentration in Hopper-era quarters. The other half comes from AI cloud providers, industrial deployments, enterprise organizations, and sovereign AI programs. That shift reduces Nvidia's exposure to concentration risk. When a government makes a long-term national AI infrastructure commitment or a regional AI cloud provider expands inference capacity, those purchases don't get deferred because a US hyperscaler temporarily adjusts its capital plan.

Sovereign AI has moved from a strategic talking point into a measurable revenue line. Nvidia's sovereign AI revenue crossed $30 billion in fiscal year 2026, up more than three times year over year, representing roughly 14% of total company revenue. The buyers include the UK, France, the Netherlands, Canada, Singapore, and India. India's $1 billion sovereign AI project with Nvidia represents one of the larger individual national commitments to domestically owned AI compute. For governments watching state-directed AI programs elsewhere and evaluating their dependency on foreign cloud infrastructure, the calculation on building owned capacity has shifted from optional to strategic.

The Forces Driving the Buildout Forward

The Q1 results confirm what many infrastructure observers suspected but couldn't quantify: the AI buildout is entering a second phase driven by inference and agentic workloads rather than the training buildout that defined the first phase.

Training a large model is a periodic, high-intensity cost that eventually plateaus once a generation of models is built. Running agents continuously at scale is a persistent operational cost that grows with deployment breadth. The more agentic applications an organization deploys, the more inference capacity it needs, and that capacity scales with usage rather than with discrete model releases. Nvidia's guidance to $91 billion for Q2 is a bet that this inference and agentic demand expansion is the durable driver of the next several quarters, not a single training cycle pulling forward spend.

For Q2 fiscal 2027, Nvidia guided to $91 billion in revenue, plus or minus 2%. At the midpoint, that is a 12% sequential increase from an already record Q1. One important note: that guidance assumes zero data center compute revenue from China. Nvidia cannot sell its highest-capability chips to Chinese customers under current US export restrictions. If there were any change in export policy, the $91 billion figure would be conservative. The guidance also signals that Blackwell demand has not peaked, which stands in contrast to the usual late-cycle skepticism about backlog clearance and deceleration.

Nvidia also confirmed details about Vera Rubin, the platform generation that follows Blackwell. It is described as purpose-built for agentic AI workloads, which signals a design philosophy evolution. Where Blackwell was optimized to extend Hopper's training strengths into inference, Vera Rubin appears designed from the start around the persistent, multi-step, tool-using workloads that define agentic systems.

Alongside the hardware roadmap, Nvidia released Dynamo 1.0, an inference software layer that achieves a 7x throughput improvement on existing Blackwell hardware compared to previous configurations. A software multiplier of that scale matters because it means organizations that already committed to Blackwell infrastructure can get dramatically better output without buying new hardware. It also gives Nvidia a software wedge that extends the customer relationship beyond the point of hardware sale, building retention through performance dependencies that are distinct from hardware lock-in alone. This is a familiar pattern in mature enterprise infrastructure markets: the hardware cycle creates the initial revenue, and the software layer creates stickiness and defines the upgrade cycle that follows.

Nvidia raised its quarterly cash dividend from $0.01 to $0.25 per share in a single step, a 25-fold increase, and announced an $80 billion share repurchase authorization with no expiration date. In Q1 alone, the company returned $20 billion to shareholders through buybacks and dividends combined. A company returning $20 billion in a single quarter while investing in the next platform generation is not constrained. It is operating in a rare position where internal growth opportunities cannot fully absorb the cash the business generates. That posture contrasts sharply with the chip industry's traditional cyclical narrative.

How This Quarter Changes AI Infrastructure Planning

The stock fell 5.46% to $184.89 the day after the earnings release. The explanation is mechanical rather than fundamental. Expectations had moved ahead of what even a record quarter could deliver. Some institutional traders had positioned for upside above $85 billion. When Nvidia came in at $81.6 billion, the beat was real but not at the level the most aggressive forecasts had priced in. The $91 billion Q2 guidance was strong, but a subset of market participants had extended models to $95 billion or higher. When the actual number landed below those stretched estimates, the stock repriced. This pattern is a reliable feature of Nvidia's stock behavior over the last two years, not an exception. It says something about the market's expectations for Nvidia. It says nothing negative about the underlying business.

For organizations making infrastructure decisions now, the Q1 results confirm two things. First, the AI buildout cycle is not plateauing. A $91 billion Q2 guide from a company whose entire business is now AI data center hardware means data center capital spending is still rising. The hyperscalers have not hit a ceiling. Enterprise and sovereign buyers are adding to the base rather than substituting for hyperscaler spending. The net effect is a market that continues to expand in aggregate even as individual buyer categories fluctuate.

Second, the diversification of Nvidia's customer base creates a more durable demand profile than the previous hyperscaler-concentrated model. Previous cycles where one or two buyer categories represented the majority of spending were vulnerable to concentration events. The emerging distribution across hyperscalers, AI clouds, enterprises, and sovereign programs represents a structural change in who funds the buildout, not just a temporary widening of the customer list.

For teams planning GPU capacity and understanding where the major infrastructure providers stand heading into the second half of 2026, our AI Infrastructure Companies to Know in 2026 resource covers the competitive landscape as Blackwell deployments set a new performance baseline across the market. And for context on the physical constraints behind these demand numbers, our earlier reporting on how data centers drove a 76% power spike on America's largest grid covers the supply-side reality that makes Nvidia's guidance credible rather than aspirational.

GPU availability and cost will remain constrained through at least the next several quarters. The companies and organizations that secured long-term capacity commitments in 2025 are in a stronger position than those building spot-price strategies. The Q2 print will be the clearer signal for identifying whether the buildout curve is still steepening or approaching an inflection. At $91 billion in forward guidance, Nvidia is clearly not betting on deceleration.