Nvidia Q2 Revenue Hits $96.2 Billion as AI Infrastructure Demand Accelerates

Nvidia AI chip demand accelerated again in the company’s fiscal second quarter, strengthening the argument that the current artificial-intelligence boom increasingly resembles a global infrastructure buildout rather than a conventional semiconductor cycle.

For the three months ended July 26, 2026, Nvidia reported revenue of $96.2 billion. That represented an 18% increase from the previous quarter and a 106% jump from a year earlier.

The company’s Data Center business generated $89.0 billion, up 117% year over year. Data-center products therefore accounted for more than 92% of Nvidia’s quarterly revenue.

GAAP net income reached $59.7 billion, compared with $26.4 billion a year earlier. Gross margin stood at 75%.

Nvidia also forecast approximately $108 billion in revenue for its fiscal third quarter, plus or minus 2%.

The outlook does not assume any Data Center compute revenue from China.

Chief executive Jensen Huang summed up the company’s argument with a simple line: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.”

The numbers suggest customers are still willing to spend heavily to turn that compute into production infrastructure.

Nvidia’s Q2 numbers at a glance

Metric	Q2 FY2027	Q2 FY2026	Change
Revenue	$96.2B	$46.7B	+106%
Data Center revenue	$89.0B	—	+117%
GAAP net income	$59.7B	$26.4B	+126%
GAAP gross margin	75.0%	72.4%	+2.6 pts
Q3 revenue outlook	~$108B	—	—

The quarter also beat market expectations.

Reuters reported that analysts had expected approximately $92.17 billion in total revenue. Nvidia’s $108 billion third-quarter forecast also came in above the roughly $104.19 billion analysts expected.

Those results matter because investors have increasingly questioned whether several years of explosive AI spending can continue.

For now, Nvidia says demand is still growing faster than its ability to supply hardware.

Nvidia makes an unusual 70% growth forecast

One of the biggest surprises did not concern the quarter Nvidia had just completed.

Management gave investors a rare view into the following fiscal year.

Nvidia expects revenue to grow by approximately 70% in fiscal 2028, which ends in January 2028.

Before the announcement, analysts had expected growth of about 44%.

Nvidia normally avoids giving revenue guidance that far ahead.

The company says its confidence comes from a broader group of customers buying AI infrastructure.

Large cloud providers remain major buyers. However, Nvidia is also seeing stronger demand from AI laboratories, specialist AI clouds, enterprises, governments and industrial customers.

The company expects AI labs to contribute roughly one-quarter of its overall business next year.

Finance chief Colette Kress added another striking detail during the earnings call.

Based on customer forecasts, demand could support roughly twice Nvidia’s current growth rate next year. Supply constraints prevent the company from assuming that level of sales.

That gap turns Nvidia’s supply chain into one of the most important parts of the AI infrastructure story.

Nvidia locks in $279 billion of future supply

Nvidia is making unusually large commitments to ensure enough hardware exists to meet future Nvidia AI chip demand.

Its latest regulatory filing shows $279 billion in supply and capacity commitments as of July 26.

That figure was only $119 billion one quarter earlier.

The commitments cover Nvidia’s data-center infrastructure systems. They primarily involve memory and manufacturing capacity needed for current and future product architectures.

The schedule includes approximately:

• $92 billion for the remainder of fiscal 2027
• $87 billion during fiscal 2028
• $88 billion during fiscal 2029
• smaller commitments in later years

High-bandwidth memory is particularly important.

Modern AI accelerators require huge quantities of advanced memory placed extremely close to the GPU. Growing competition for that memory has increased prices and made supply security a strategic priority.

The scale of Nvidia’s commitment signals confidence in future demand. It also creates risk.

If AI infrastructure investment slows sharply, Nvidia could find itself committed to more capacity than the market needs.

Rising memory costs could pressure margins

That supply challenge is already affecting profitability expectations.

Nvidia expects third-quarter gross margin of approximately 74%, compared with 75% in the quarter just reported.

Management says margins could fall further to roughly 71% to 72% in the fourth quarter before improving.

Higher memory and component costs are among the reasons.

This creates an unusual situation for Nvidia.

Demand remains exceptionally strong, but securing enough components to satisfy that demand is becoming more expensive.

The company has therefore started raising some prices to offset higher input costs.

The next few quarters will show how much of those increases customers are willing to absorb.

Amazon adds another 2 million Nvidia GPUs

Amazon Web Services provided one of the clearest signs that large customers are still expanding their AI plans.

AWS and Nvidia announced that Amazon will deploy 2 million additional Nvidia GPUs during 2027 and 2028.

Those units will include Blackwell Ultra, Rubin and Rubin Ultra processors.

The new commitment builds on a plan announced at Nvidia GTC 2026 to add more than 1 million Nvidia GPUs starting in 2026.

Nvidia and AWS say demand has exceeded expectations since that initial announcement.

The expansion means AWS is planning more than 3 million Nvidia GPUs across the two deployment programmes.

That does not mean Amazon is abandoning its own silicon.

AWS continues to develop and deploy custom AI processors. Instead, the arrangement illustrates the current structure of the market: hyperscalers are building proprietary chips while simultaneously purchasing enormous quantities of Nvidia hardware.

The AWS deal goes beyond GPUs

The expanded partnership covers much more than accelerators.

AWS plans to introduce infrastructure based on Nvidia’s Vera CPUs.

The companies are also extending Nvidia’s NVLink Fusion technology and integrating additional Nvidia networking, data-processing and AI-model technologies into AWS services.

Amazon Robotics will use Nvidia’s physical-AI platform for warehouse automation and next-generation robotics.

AWS will also continue supporting Nvidia’s Nemotron open models through Amazon Bedrock and SageMaker.

The companies additionally plan secure AI infrastructure for the U.S. government.

That programme includes 100,000 Nvidia GPUs intended for federal and national-security workloads on AWS infrastructure.

The breadth of the partnership shows how Nvidia is trying to position itself beyond the accelerator itself.

Its product increasingly includes processors, networking, software, models and entire data-center architectures.

Vera Rubin begins contributing to the next growth cycle

Nvidia’s next-generation Vera Rubin platform is now moving into production.

The company says complete systems are running with partners including Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, CoreWeave and Nebius.

Nvidia expects Vera Rubin to contribute roughly one-fifth of Data Center revenue during its current quarter.

That is an important transition.

Semiconductor companies often face a period when customers delay purchases while waiting for a new architecture.

Nvidia is instead arguing that demand is high enough for existing products and the new generation to scale together.

Rubin also expands Nvidia’s strategy beyond individual GPUs.

The platform combines the Rubin GPU, Vera CPU, networking, storage infrastructure and other processors into a complete AI-computing system.

That supports Nvidia’s broader effort to sell the architecture of an “AI factory,” not simply chips installed inside one.

Hyperscalers could spend more than $1.5 trillion in two years

Nvidia’s results sit inside an even larger capital-spending boom.

S&P Global Market Intelligence estimates hyperscaler capital expenditure could reach approximately $682 billion in 2026.

The forecast rises to about $878 billion in 2027.

Combined, that represents roughly $1.56 trillion over two years.

Much of that money will go toward data centres, GPUs, networking equipment, power systems and related AI infrastructure.

This spending explains why Nvidia can generate nearly $90 billion of Data Center revenue in a single quarter without exhausting demand.

Amazon alone raised its 2026 capital-spending outlook to approximately $220 billion following strong growth at AWS.

The infrastructure race also extends beyond the traditional hyperscalers.

Specialist AI-cloud companies, sovereign computing programmes, frontier-model developers and large enterprises are increasingly building dedicated AI capacity.

Nvidia is helping finance the infrastructure it supplies

Nvidia is not simply waiting for customers to finance new data centres.

The company is becoming more involved in helping capital reach the projects that buy its hardware.

In August, Nvidia partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing structures designed to mobilise more than $500 billion in third-party capital for AI infrastructure.

Nvidia could provide guarantees covering part of certain deals.

The basic logic is straightforward.

AI data centres require enormous upfront investment. By helping customers access financing, Nvidia can expand the number of organisations capable of purchasing Nvidia-powered infrastructure.

That strategy could increase access to compute.

It also creates questions about how much Nvidia should financially support companies that subsequently buy its products.

Nvidia backs massive OpenAI data-center project

The OpenAI relationship provides one example.

Nvidia agreed to provide a guarantee of up to $105 billion connected to OpenAI’s long-term use of an enormous SB Energy data-centre campus in Pike County, Ohio.

Nvidia is also investing $1.5 billion in SB Energy.

The planned campus could eventually reach 8 gigawatts of capacity, although development will occur in stages. An initial 800 megawatts is targeted for operation by 2028.

Nvidia is expected to supply the project’s computing hardware.

These arrangements show how the company is moving deeper into financing the physical infrastructure surrounding its chips.

They also increase scrutiny of whether some AI infrastructure demand is becoming financially interconnected.

Nvidia pauses one financing experiment

That scrutiny is already having an effect.

A separate Nvidia programme involving smaller AI-cloud companies has reportedly been paused.

The initiative offered financial support to AI clouds while allowing Nvidia to participate in revenue generated from their customers.

The structure raised concerns among some investors about circular financing: Nvidia helping finance a customer that then buys Nvidia hardware.

It also generated internal concerns about how much influence Nvidia could exercise over participating cloud providers.

Pausing that specific arrangement does not mean Nvidia is abandoning infrastructure finance.

The wider $500 billion capital-mobilisation effort remains part of the company’s strategy.

However, the episode demonstrates that investors increasingly care not only about the size of AI demand but also about how that demand is financed.

Custom AI chips are becoming serious competitors

Nvidia also faces a growing challenge from custom silicon.

Amazon, Google and Microsoft are all building their own AI processors.

OpenAI joined that group with Jalapeño, an inference accelerator developed with Broadcom.

OpenAI published its first benchmark results on August 25.

The company said Jalapeño delivered stronger throughput per kilowatt and lower token latency than the commercial systems included in its comparison for several large language models.

However, the benchmark has important limits.

Jalapeño is an inference chip. It does not replace Nvidia GPUs for model training.

OpenAI also did not compare the chip with Nvidia’s newer Vera Rubin platform.

The result therefore shows that specialised accelerators can challenge Nvidia in specific workloads. It does not establish that Nvidia has lost its broader infrastructure advantage.

Nvidia’s software ecosystem remains a major defence

Hardware performance is only one part of Nvidia’s position.

CUDA has become a widely used programming platform for accelerated computing.

Nvidia also provides networking, inference software, libraries, deployment systems and increasingly complete rack-scale architectures.

Customers switching to another accelerator therefore need to consider more than the price or performance of one chip.

They must evaluate software compatibility, engineering effort, networking and deployment tools.

That ecosystem remains one of Nvidia’s strongest competitive advantages.

Still, hyperscalers have strong incentives to reduce dependence on a single supplier.

Even modest movement toward custom silicon could become significant when annual AI infrastructure spending reaches hundreds of billions of dollars.

What could slow Nvidia’s AI growth?

Nvidia’s results support the bullish case for AI infrastructure, but several risks remain.

Power availability is one of the largest.

A data centre cannot begin operating simply because its GPUs arrive. New campuses require grid connections, transmission infrastructure, cooling and large amounts of electricity.

Memory and manufacturing capacity present another bottleneck.

Nvidia’s $279 billion supply commitment demonstrates how seriously the company takes that constraint.

Customer concentration also creates strategic tension.

Amazon, Microsoft and Google are some of Nvidia’s largest customers, yet all three are building competing AI chips.

China remains unpredictable.

Nvidia excluded China Data Center compute revenue entirely from its current-quarter forecast because export restrictions continue to complicate sales.

Finally, there is the question of return on investment.

The infrastructure boom can continue only if companies eventually generate enough revenue or productivity gains from AI to justify enormous spending on data centres and computing equipment.

Why Nvidia’s $96.2 billion quarter matters

This quarter matters because Nvidia is increasingly functioning as more than a semiconductor supplier.

The company provides accelerators, CPUs, networking, software and complete data-centre systems.

It is securing hundreds of billions of dollars in future manufacturing capacity.

It is helping organise financing for customers that need new AI factories.

And its largest cloud partners continue to expand GPU deployments even as they develop competing processors.

That combination makes Nvidia one of the clearest indicators of the wider AI infrastructure cycle.

For now, the indicator still points upward.

Data Center revenue reached $89 billion. AWS added another 2 million GPUs to its deployment plans. Nvidia forecast roughly $108 billion of revenue for the current quarter and about 70% growth next fiscal year.

The bigger test comes next.

If companies continue turning AI compute into valuable products and services, today’s data-centre construction could become the foundation of a much larger computing economy.

If returns disappoint, the warning signs will likely appear in delayed data centres, lower equipment orders, tighter financing and more workloads shifting toward custom silicon.

For the moment, Nvidia’s shipment data suggests the buildout remains firmly in expansion mode.