Q2 2026

Nvidia Posts Another Blockbuster Quarter, But Debt is Rising in AI Frenzy

co-written by newsrooms.ai27. August 2026, 12:29

Nvidia has just reported results for the second quarter of fiscal 2027, and once again they came in bigger than expected. Revenue climbed to $96.2 billion, up 106 percent from the year-earlier quarter and 18 percent from the previous one. The data center business that carries the company accounted for $89.0 billion of that on its own, up 117 percent. Gross margin came in at 75.0 percent. Operating profit grew even faster than revenue: $63.7 billion marks a gain of 124 percent year over year and 19 percent quarter over quarter. Earnings per share landed at $2.46 on a GAAP basis and $2.22 adjusted.

What moved the market, though, was the outlook. For the current quarter Nvidia guided to $108 billion in revenue, give or take two percent. Analysts had penciled in roughly $104.2 billion and adjusted earnings of $2.09 per share. The company cleared both marks.

How the Stock Reacted

Heading into the report, Nvidia shares had slipped 1.6 percent in regular trading to close at $209.66. After the bell the stock turned positive and added around five percent to roughly $219.55. In premarket trading the next morning the gain ran to more than seven percent, and once regular trading opened it settled at about six percent up. The move rested mainly on the guidance rather than on the quarter itself.

Analysts followed suit. Goldman Sachs and Citigroup were among those raising their price targets on the stock. Worth recalling: after the previous quarterly report, investors had stayed skeptical despite record revenue.

Nvidia also returned about $26.0 billion to shareholders during the quarter through buybacks and dividends. Roughly $99.0 billion remains under the current repurchase authorization, and the next quarterly dividend of $0.25 per share is due in early October.

The Wins

The most important product signal: the Vera Rubin platform is in full production and already running in racks at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. Nvidia has managed the generational handoff from Blackwell without a visible dip in revenue, which in the past has been the weakest operational moment of any such transition.

In inference, the fastest-growing slice of the AI market, the $20 billion acquisition of Groq is starting to pay off: the Groq 3 LPX accelerator is now in full production. Alongside it comes Vera, Nvidia’s first in-house CPU built specifically for AI agents, with SpaceXAI among the announced adopters.

The single largest deal around the results, however, came from Amazon. AWS and Nvidia simultaneously announced a sweeping expansion of their partnership. Two million additional GPUs from the Blackwell Ultra, Rubin and Rubin Ultra lines are set to go into AWS global infrastructure over the next two years. At the most recent GTC, AWS had talked about more than one million chips; demand has since run well ahead of that plan, according to both companies.

The agreement goes well beyond unit counts. AWS is adding Vera CPUs to its lineup, connecting its own Trainium chips to Nvidia’s new custom memory technology NVHBM over NVLink Fusion, and building AI factories for the U.S. government, including 100,000 GPUs on secure infrastructure for federal and national security workloads at Impact Level 6 and above. Nemotron open models on Bedrock and SageMaker, GPU-accelerated data processing on EMR and OpenSearch, and the use of Nvidia’s physical AI stack at Amazon Robotics round out the package. The notable part is that the hyperscaler with the most advanced in-house AI chip is deepening its reliance on Nvidia and folding both architectures into a shared rack-scale design.

On the financing side, Nvidia has set up platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR intended to mobilize more than $500 billion of third-party capital for AI infrastructure. That secures customer demand while keeping the sums off Nvidia’s own balance sheet.

The geographic footprint keeps widening as well: national AI infrastructure in Japan, gigawatt-scale projects in Korea with SK Telecom, NAVER and Brookfield, a multiyear memory partnership with SK hynix, and 35 new AI supercomputers in development across Europe. In the edge business ($7.2 billion, up 27 percent), the Windows push with Microsoft around RTX Spark and the DRIVE Hyperion robotaxi stack with Foxconn, VinFast, Uber and HUMAIN add to the momentum.

Where the Problems Could Come From

China remains the largest blank space. The $108 billion forecast contains not a single dollar of data center revenue from China. That leaves room on the upside should export conditions ease, and it also shows how completely Nvidia has written off one of the world’s biggest data center markets.

Second, the margin. Nvidia guided to a gross margin of 74.0 percent for the current quarter, down from 75.0 percent. Operating expenses are meanwhile growing at 55 percent year over year, with pricier memory and more complex rack systems working their way through.

Third, the composition of profit. Adjusted earnings usually come in above the GAAP figure, because items such as stock-based compensation get stripped out. This time it runs the other way: $59.7 billion on a GAAP basis against $54.0 billion adjusted. The gap of roughly $5.7 billion comes from effects outside the operating business that Nvidia itself backs out, typically valuation gains on equity stakes. That portion of reported profit therefore stems from rising company valuations, while chip sales stay untouched by it. The sequential comparison makes the same point: GAAP profit grew two percent, adjusted profit 18 percent. Because Nvidia holds stakes in many of these companies and supplies them at the same time, this is exactly where the debate about circular financing between OpenAI, Nvidia and Oracle begins. In parallel, the company’s debt has risen from $8.5 billion to $33.4 billion, on top of multiyear infrastructure commitments worth roughly $366 billion.

Fourth, customer concentration. The bulk of data center revenue rests with a manageable number of hyperscalers, neoclouds and frontier labs. Commitments like the AWS order for two million GPUs show the flip side of that growth: a single customer can swing a substantial share of future utilization. Several of those customers are also building their own hardware, OpenAI’s Jalapeño chip being the most prominent example, and Amazon continues to develop Trainium, which connects through NVLink Fusion yet competes for the same workloads over the long run.

Finally, the physical bottleneck. The partnership with SB Energy to secure land, power and shell capacity at a technology campus in Ohio suggests that the limit on growth now sits with energy and real estate more than with the chips themselves.