Nvidia's second-quarter earnings report, published on Aug 28, 2026, confirmed that demand for its AI chips remains intense. The company posted revenue of $96.2 billion, up 106% year-over-year, a figure that handily beat market expectations. But beneath the headline numbers, the report exposed a series of fragilities: shrinking margins, heavy customer concentration, and a balance sheet increasingly tied to financing its own customers' AI infrastructure.
The results were closely watched as a barometer for the broader AI boom. Investors wanted reassurance that the spending spree on data centers would continue. They got it, at least for the near term. Yet the fine print suggests that Nvidia is now carrying a significant portion of the financial risk for that boom itself.
Data-Center Demand Still Roaring
The core engine of Nvidia's business, its data-center segment, delivered $89 billion in revenue for the quarter. That represents the bulk of total sales and shows no sign of slowing. Hyperscale customers, the large cloud providers that support frontier AI labs like Anthropic and OpenAI, purchased $48.7 billion worth of chips, up 13% from the previous quarter.
Growth is broadening beyond those giants. Revenue from AI clouds, industrial customers, and enterprises, a category Nvidia calls ACIE, rose 25% quarter-over-quarter and 138% year-over-year, totaling $40.3 billion. That mix suggests demand is no longer the sole province of a handful of tech titans.
CEO Jensen Huang, in an interview with CNBC's Jim Cramer, said demand is "super strong" and accelerating. He also provided forward guidance for data-center revenue growth of 70% for fiscal 2028, which begins in late January 2027. It is a striking number, especially given that Huang previously refrained from commenting on future demand at all.
CFO Colette Kress characterized the 70% growth forecast as "a supply-constrained outlook." In other words, Nvidia believes it could sell even more if it could make more. The company's latest platform, Vera Rubin, which succeeded the Blackwell generation, reached full production in August 2026. Kress called the ramp "the fastest product ramp in Nvidia's history." Vera Rubin is expected to generate 20% of data-center revenue in the current quarter. Clients with operational systems already using the platform include CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, and Nebius.
Margins Under Pressure
The good news on demand is tempered by a clear deterioration in profitability. Gross margins are expected to fall from 75% to 74% in the third quarter, then settle between 71% and 72% in the fourth quarter. The erosion is driven primarily by memory shortages, which are pushing up component costs.
That is a meaningful shift for a company that has enjoyed extraordinary pricing power. The margin decline is not catastrophic, but it signals that Nvidia is no longer immune to supply-chain friction. The company is paying more for the memory chips that go into its systems, and it is passing less of that cost onto customers.
The product transition to Vera Rubin appears to be going smoothly, with no technical issues reported. Solid initial sales have lifted the transition. But the margin compression suggests that the era of ever-expanding profitability may be leveling off.
Concentration Remains a Weak Spot
A more troubling pattern emerges when looking at who is buying. One customer accounted for 16% of Nvidia's revenue last quarter. That is a single point of failure that analysts have flagged repeatedly. The concentration is even starker on the receivables side. According to the company's 10-Q filing, five customers made up 70% of accounts receivable.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Days sales outstanding rose to 60 days, up from prior levels. That means Nvidia is waiting longer to collect cash. To ease the strain, the company extended payment terms for select investment-grade clients to up to one year. More generous credit policy does not necessarily indicate poor revenue quality or financial distress, but it does move risk onto Nvidia's balance sheet.
Vendor credit helps finance clients' purchases, but it also means Nvidia is effectively acting as a lender. That is a role the company has not historically played, and it changes the risk profile of the business. The size and concentration of these arrangements are the issues that worry analysts most.
Nvidia's Growing Financial Exposure
The credit extension is just one piece of a much larger financial web. Nvidia has $36 billion in "AI-cloud commitments" to help clients secure data center capacity, land, and power. That is money the company is putting at risk to ensure its customers can actually build the facilities needed to run its chips.
Total "future commitments" stand at $366 billion, most of which, $279 billion, is tied to supply and capacity. These are obligations Nvidia has taken on to lock in manufacturing and infrastructure. They are enormous numbers, and they represent a direct exposure to the AI infrastructure buildout that has raised concerns across the analyst community.
Beyond commitments, Nvidia extends customer guarantees with a maximum exposure of $108.5 billion. The bulk of that, $105 billion, is associated with a single project: a 4.25-gigawatt OpenAI data center in Ohio. That is an extraordinary concentration of risk in one customer and one location.
To diversify that risk, Nvidia signed agreements to raise more than $500 billion in third-party capital for its customers. The company is partnering with major financial institutions, including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The idea is to bring outside money into the AI buildout rather than relying solely on Nvidia's own balance sheet.
Addressing the Circular Financing Question
The arrangement has drawn scrutiny. Critics have raised concerns about "circular financing," where Nvidia lends money to customers who then use it to buy Nvidia chips, creating the appearance of demand that is partly self-funded. Kress addressed those concerns directly, stressing that third-party capital providers conduct independent underwriting. She also argued that Nvidia's compute is durable and fungible, meaning the chips retain value and can be redeployed if a customer defaults.
That defense may reassure some investors, but it does not eliminate the risk. The guarantees, the extended payment terms, and the commitments all point to a company that is deeply intertwined with the financial health of its customers. If the AI boom falters, Nvidia would not just lose sales. It would be on the hook for billions in obligations.
The scorecard for this earnings report is mixed. Data-center demand is strong and growing beyond hyperscale customers. The product transition is lifted by solid initial sales and no technical issues. But the weak links are persistent customer and receivables concentration, and the high direct exposure as Nvidia finances the AI infrastructure boom.
For now, the market is rewarding the growth. The demand for semiconductor chips far exceeded supply, and the AI boom is not short of demand, at least in the next several months. But signs of fragility still exist. The margin decline, the concentration, and the financing exposure are all real. Nvidia is no longer just a chipmaker. It is a financier of the AI economy, and that comes with risks that no amount of revenue growth can fully mask.