On August 26, 2026, Nvidia reported $96.221 billion of revenue for the quarter ended July 26, 2026, up 106 percent from a year earlier. Data Center revenue was $89.023 billion, up 117 percent. The 10-Q filed the same day says where that money came from, and the list is short.
Hyperscale accounted for $48.710 billion of the quarter. That is 50.6 percent of all revenue, from one buyer category. One direct customer, on its own, represented 16 percent of total revenue. Nvidia guided the current quarter to $108.0 billion, plus or minus 2 percent.
The most recent nationally representative measurement of everybody else was taken before that quarter even began. Economists at the US Census Bureau found 18 percent of US firms using AI in a business function. Among those that had adopted, 57 percent were running it in three functions or fewer.
What Nvidia's filing actually discloses
Note 7 of the quarterly report is the part worth reading. Five direct customers accounted for 22, 14, 13, 11 and 10 percent of accounts receivable as of July 26, 2026. Those five held 70 percent of the balance owed to the company.
Nvidia states the position in its own words: "Our revenue is concentrated among a limited number of direct and indirect customers and this trend may continue." For the first half of fiscal 2027, three direct customers represented 16, 15 and 13 percent of total revenue.
The filing also discloses that Nvidia estimates one AI research and deployment company contributed a meaningful amount of revenue by purchasing cloud services from Nvidia's own customers. Part of the demand travels through the customer base rather than past it.
One line is absent. Nvidia does not report enterprise as a separate market platform. Data Center splits into Hyperscale at $48.710 billion and a single bucket called AI Clouds, Industrial and Enterprise at $40.313 billion. The enterprise share of this build is not a number the disclosure produces.
What the Census Bureau measured
In April 2026, six economists including John Haltiwanger published the results of the second AI supplement to the Business Trends and Outlook Survey (Bonney et al., CES 26-25). The survey is nationally representative and weighted to the universe of US firms. The reference period is November 2025 to January 2026.
Eighteen percent of firms used AI in a business function, rising to 32 percent on an employment-weighted basis, with 22 percent expected within six months. Very large firms in Information, Professional Services and Finance reach 50 to 60 percent, and 60 to 70 percent employment-weighted. Depth is where the picture changes. Among adopters, 57 percent use AI in three or fewer business functions, most often Sales and Marketing (52 percent), Strategy and Business Development (45 percent) and IT (41 percent). At task level, 65 percent of firms limit use to three or fewer tasks, and 66 percent of users rely on AI solely to augment work.
The number that moves with commercial performance is not whether a firm adopted AI. It is how many functions the adoption reached. Hikari Blue · operator note
The finding that belongs in the board pack
The paper runs regressions on firm outcomes. It reports a positive correlation between commercial performance and the breadth of AI integration, measured across functional deployment, task-level use and operational investment. The authors present this as correlation. So do we.
Set that beside the 57 percent. The variable associated with performance is precisely the one most adopters have not moved. Breadth, not adoption, is the reported quantity that carries information.
The labor result is separable and sharper. Functional breadth and operational investment are positively associated with employment decreases. Worker-task integration shows no significant link to headcount reduction once those two are accounted for. Employment effects trace to the formal deployment channel, not to staff using a model on their own tasks. Across the sample, AI-related employment decreases occurred in 2 percent of firms.
The objection, stated plainly
These two measurements do not share a period, a universe or a unit. Nvidia's quarter ended July 26, 2026. The survey's reference period ended in January 2026. Nvidia sells worldwide, and 38 percent of its second-quarter revenue came from customers headquartered outside the United States.
We are not claiming the two figures should reconcile. The claim is narrower and survives the mismatch. No nationally representative US measurement shows the breadth of firm-level AI use expanding at anything close to the growth rate of supplier revenue. Nvidia's own concentration note says where that revenue lands instead. The two documents agree on direction: this build is being paid for by depth in a small number of buyers, not by breadth across the economy.
The lag is itself the operator's problem. Supplier revenue updates every quarter. Firm-level adoption data updates on a research cadence, two to three quarters behind. Capital decisions get made against the fast series and justified with the slow one.
What changes on Monday
Retire the adoption number. "We use AI" describes 18 percent of US firms and tells a board nothing about which of them is compounding. Replace it with a breadth line: how many business functions run AI in production, who owns each one, and what operational investment sits behind it.
Expect the constraint to be structural, not technical. Moving from three functions to eight is not blocked by model quality. It is blocked by data access, permission topology and the evidence each new function must produce before it runs unattended. In financial services, the sector the Census data puts at the top of the adoption range, that evidence requirement decides whether function four ever ships. This is the unglamorous work of an AI operating layer, and it is the actual gate on breadth.
How many business functions run AI in production today, who owns each one, and against which revenue or cost line is that spend booked?
If the answer arrives as a count of licenses rather than a list of functions, the firm is measuring the wrong variable. Carry three countable numbers into the next quarter: functions in production, cost per function, and the line each one touches. That is what the Census data associates with performance, and what an AI strategy can be held to.
- NVIDIA Corporation (August 26, 2026). Form 10-Q for the quarterly period ended July 26, 2026, filed with the US Securities and Exchange Commission. Primary source for total revenue of $96,221 million against $46,743 million a year earlier, Data Center revenue of $89,023 million against $41,096 million, Hyperscale revenue of $48,710 million, AI Clouds, Industrial and Enterprise revenue of $40,313 million, gross margin of 75.0 percent, United States headquartered customer revenue of $60,074 million, the statement that 38 percent of second-quarter revenue came from customers headquartered outside the United States, the disclosure that one direct customer represented 16 percent of total revenue for the second quarter and three direct customers represented 16, 15 and 13 percent for the first half, the Note 7 disclosure that five direct customers accounted for 22, 14, 13, 11 and 10 percent of accounts receivable as of July 26, 2026, the sentence "Our revenue is concentrated among a limited number of direct and indirect customers and this trend may continue", and the estimate that one AI research and deployment company contributed a meaningful amount of revenue by purchasing cloud services from Nvidia's customers. The 50.6 percent and 70 percent figures used above are arithmetic on these disclosed numbers and can be recalculated from the filing. sec.gov, NVIDIA Form 10-Q, quarter ended July 26, 2026
- Bonney, K., Breaux, C., Dinlersoz, E., Foster, L., Haltiwanger, J. and Pande, A. (April 2026). The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks. US Census Bureau, Center for Economic Studies, Working Paper CES 26-25. Primary source for the November 2025 to January 2026 reference period of the second AI supplement to the Business Trends and Outlook Survey, the 18 percent firm-level AI use rate and 32 percent employment-weighted rate, the 22 percent six-month expectation, the 50 to 60 percent range for very large firms in Information, Professional Services and Finance, the finding that 57 percent of users integrate AI in three or fewer business functions, the function shares for Sales and Marketing (52 percent), Strategy and Business Development (45 percent) and IT (41 percent), the 65 percent of firms limiting use to three or fewer tasks, the 66 percent relying on AI solely to augment tasks, the 2 percent of firms reporting AI-related employment decreases, the positive correlation between commercial performance and breadth of AI integration, and the divergence in which functional breadth and operational investment associate with employment decreases while worker-task integration does not. CES working papers have not undergone the review accorded Census Bureau publications, as the series states. census.gov, CES Working Paper 26-25
- NVIDIA Corporation (August 26, 2026). NVIDIA Announces Financial Results for Second Quarter Fiscal 2027. Company announcement, attributed as such. Sole source for the third-quarter fiscal 2027 revenue outlook of $108.0 billion plus or minus 2 percent. That figure is guidance issued by the company, not a result, and is the only forward-looking number used above. nvidianews.nvidia.com, second quarter fiscal 2027 results
- US Census Bureau (May 26, 2026). Large Firms With at Least 20 Employees Biggest AI Users. Cited for the separate Business Trends and Outlook Survey series running to a May 3, 2026 reference date, showing 37 percent AI use among firms with at least 250 employees, and for the November 2025 wording change from AI use "in producing goods or services" to use "in any business function". The two series are not interchangeable and are kept distinct above. census.gov, AI use by business size
The Hikari Blue team · Austin, September 2026