Nvidia reported $81.6 billion of revenue for the quarter ended April 26, 2026, up 85% from a year earlier; Data Center contributed $75.2 billion [1]. The strength of that demand is clear. For an investor financing capacity, the harder issue sits on the customer’s balance sheet: whether contracted receipts will cover installation, power, depreciation and debt service before the equipment loses its earning power.
What Nvidia's release shows
Under its previous sub-market reporting, Nvidia put Data Center compute revenue at $60.4 billion, up 77% from a year ago, and networking revenue at $14.8 billion, up 199% [1]. GAAP gross margin was 74.9% [1]. Gross margin is struck before research and development and overhead, so it describes the margin on products sold and says little about what remains after those costs. The figures are also recognised revenue. The release excerpt includes no cash-flow statement, so it cannot tell us when customers paid.
GAAP and non-GAAP earnings per diluted share were $2.39 and $1.87 [1]. The excerpt does not explain the gap between them, so this note does not lean on earnings.
Nvidia returned approximately $20.0 billion to shareholders in the quarter through repurchases and cash dividends, and $38.5 billion remained under its repurchase authorisation at quarter end [1]. On May 18 the board approved a further $80.0 billion without expiration, and the quarterly dividend rises from $0.01 to $0.25 per share, payable June 26 [1]. Of these capital-return figures, the $20.0 billion describes completed distributions. The $80.0 billion is permission to repurchase, with no stated pace. A company usually returns cash it expects to have, so the distributions say something about Nvidia's own position. They say little about its customers' returns.
Nvidia is also changing how it reports. It will show Hyperscale, which includes public clouds and the largest consumer internet companies, and ACIE, covering AI clouds, industrial and enterprise buyers [1]. The excerpt gives no revenue split between the two. That matters here because a public cloud operator and an enterprise building its own AI facility can face different financing and utilisation risk, and the current figures do not separate them.
What Microsoft's figures cover
Microsoft reported revenue of $82.9 billion for the quarter ended March 31, 2026, up 18%, operating income of $38.4 billion, up 20%, and net income of $31.8 billion, up 23% on a GAAP basis [2]. Those are company-wide results, and none can be traced to AI alone. Its chief executive said the AI business passed an annual revenue run rate of $37 billion, up 123% [2]. Microsoft Cloud revenue was $54.5 billion, up 29%, and commercial remaining performance obligation rose 99% to $627 billion [2].
These numbers should not be set against Nvidia's. A run rate extrapolates a recent measure, while Nvidia's $75.2 billion is revenue already recognised for one quarter [1][2]. AI revenue and Microsoft Cloud revenue are different categories. Remaining performance obligation is contracted revenue not yet recognised, it covers the whole commercial business, and the excerpt does not isolate AI [2]. The two companies' customer and supplier populations overlap without being identical. Neither release supplies customer capital spending or sector cash-flow totals, so any estimate of how much of Nvidia's sales a given customer funded, or how quickly cash came back, would be invented.
Diligence questions for a financed AI site
What can be done is to list what an investor or lender would want to know about a particular data centre. The relevant terms belong to the individual project’s contracts. Aggregate supplier results cannot replace that diligence.
Power and site. Is the grid connection secured and dated? If energisation slips after equipment has been bought, someone carries the idle cost. The contracts should say whether that is the developer, the operator or the customer.
Utilisation. Is capacity contracted before the build, and on take-or-pay or usage terms? Under usage terms the owner carries any unused capacity. Under take-or-pay terms the customer's credit becomes the question.
Useful life. Compare the depreciation life of the accelerators with the customer contract term and the loan maturity. Debt that outlasts the economic life of the equipment leaves a refinancing gap.
Obsolescence. If a newer chip generation lowers rental rates for older systems, the loss in residual value falls on the owner, the lessee or the lender, depending on the documents.
Owner and operator. Where the building, the chips and the operations sit in different entities, each contract assigns shortfalls differently, and the party with the thinnest protection absorbs them first.
An attractive site would match firm delivery of power to the start of customer payments and cover debt service under a realistic utilisation case. If customer contracts expire before the loan matures, the residual exposure should be valued as refinancing risk rather than assumed away. Higher throughput from the next chip generation could weaken rental rates before accounting depreciation catches up. Strong supplier sales are compatible with poor returns on a particular facility; a sound investment case needs project-level receipts, costs and contractual protection.
Sources
[1] NVIDIA Corporation - NVIDIA Announces Financial Results for First Quarter Fiscal 2027
[2] FY26 Q3 - Press Releases - Investor Relations - Microsoft