The $5 Trillion Machine

Every figure in this article is drawn from Nvidia’s SEC filings and earnings releases, or from reporting by CNBC, the Financial Times, Bloomberg, CNN and Fortune. Sections marked as analysis are identified as such.

On April 26, 2026, Nvidia closed the books on a quarter in which it took in $81.6 billion — up 85 per cent on the same quarter a year earlier — at a gross margin of 74.9 per cent. For the quarter now under way, it told investors to expect about $91 billion. GAAP net income for the three months was $58.3 billion.

The company that designs the chips the artificial-intelligence boom runs on had already become, on October 29, 2025, the first company in history to end a trading day valued above $5 trillion. Its fiscal 2026 revenue came to $215.9 billion, up 65 per cent in a year.

Those are the numbers everyone quotes. The more revealing ones sit further down in the filings: who the money actually comes from, where Nvidia’s own money is going, and what happened when one of its largest markets disappeared.

Ninety-two cents of every dollar

Nvidia built its name selling graphics cards to gamers. That company no longer really exists. In the April quarter, the data-centre business — the chips, systems and networking sold to the builders of AI infrastructure — took in $75.2 billion of the $81.6 billion total, up 92 per cent in a year. Everything else Nvidia does — gaming, professional visualisation, automotive — now fits inside the remaining eight cents of each dollar.

The product cycle behind that number keeps shortening. The Blackwell generation is shipping at scale; its successor platform, Vera Rubin, was announced alongside the quarter, with the first systems due in the second half of 2026. Chief executive Jensen Huang’s description of the moment was characteristically unrestrained:

“The buildout of AI factories — the largest infrastructure expansion in human history — is accelerating at extraordinary speed.”

The unnamed customers

Growth of this size usually implies a broadening base of buyers. Nvidia’s own filings describe the opposite.

In its fiscal 2026 annual report, the company disclosed that a single direct customer accounted for 22 per cent of total revenue — roughly $47 billion — and a second direct customer for another 14 per cent. Two buyers, thirty-six cents of every dollar. A year earlier, the largest customer stood at 12 per cent.

Nvidia does not name them. “Direct customers” are frequently intermediaries — the system builders and distributors that assemble Nvidia silicon into servers for the giant cloud companies — which means even sophisticated investors cannot say with certainty whose demand this is. When the pattern first emerged in mid-2025, CNBC reported that two mystery customers made up 39 per cent of a single quarter’s revenue, and that analysts were left guessing at their identities.

The company’s own risk disclosures acknowledge the point, stating that revenue is concentrated among a limited number of direct and indirect customers, and that the trend may continue.

Analysis: the circle

The following section is analysis, drawn from the reported facts.

In September 2025, Nvidia and OpenAI announced a letter of intent: OpenAI would deploy at least 10 gigawatts of Nvidia systems, and Nvidia intended to invest up to $100 billion in OpenAI — progressively, as each gigawatt came online, with the first gigawatt due in the second half of 2026 on Vera Rubin hardware.

Read the structure twice. Nvidia invests in OpenAI. OpenAI spends the money on Nvidia systems. The deployment triggers the next tranche of investment. The customer’s purchases are funded, in part, by the supplier — and each purchase releases more funding.

Nor is OpenAI a one-off. Bloomberg has mapped the pattern across the industry — under the plain-spoken heading “How Microsoft, OpenAI and Nvidia Keep Paying Each Other” — identifying Nvidia stakes in or commitments to customers including CoreWeave and Nebius alongside OpenAI. The concern this raises has a name, circular financing: when a supplier funds its own customers’ purchases, reported demand can look stronger than the end-user appetite underneath it.

Nvidia rejects the framing; Huang has called the criticism “ridiculous.” But the company’s own handling of the OpenAI figure is instructive. The day it was announced, $100 billion was a headline number. By February 2026, Huang was telling reporters the figure “was never a commitment” — the letter of intent was non-binding, and Nvidia’s actual contribution to OpenAI’s funding round would not approach it. Both statements can be true. A number can move markets on the way in and be walked back on the way out, and it is reasonable to notice that it did both.

A $17 billion market, written to zero

What makes Nvidia’s growth stranger is that it happened while an entire national market vanished from its books.

In fiscal 2025, China brought Nvidia roughly $17 billion, about 13 per cent of revenue. In April 2025, the U.S. government told the company its China-market H20 chip would require an export licence indefinitely; Nvidia took a $4.5 billion charge for stranded H20 inventory in a single quarter.

Then came an arrangement with no precedent in American trade policy. In August 2025, as first reported by the Financial Times, Nvidia and AMD agreed to pay the U.S. government 15 per cent of their revenues from certain China chip sales in exchange for export licences. President Trump said publicly that he had originally asked for 20 per cent, and credited Huang with negotiating him down.

The licences were granted. The sales never materialised. Chinese authorities urged domestic companies not to buy the H20 on security grounds, and Nvidia eventually stopped manufacturing it. The company’s current guidance assumes zero data-centre compute revenue from China.

Follow the arc: a market worth one dollar in every eight was restricted, ransomed, and finally warned out of existence in about eighteen months — and Nvidia grew 85 per cent anyway. That is either the strongest possible evidence of how deep AI demand runs, or a measure of how much a single wave of infrastructure spending is carrying.

Analysis: what has to stay true

The following section is analysis.

None of this suggests the results are anything but real. The revenue is audited, the margins are genuine, and the cash is immense. Companies with accounting problems do not report $58 billion of quarterly net income and 75-cent gross margins for years on end.

The questions are structural, and there are three. Concentration: when two unnamed buyers supply more than a third of revenue, a purchasing decision at either one moves the company — and shareholders cannot see who they are. Circularity: some share of demand is financed by the supplier itself, and no disclosure separates that share out. Continuation: the valuation assumes the buildout Huang calls the largest in human history keeps running, which depends on Nvidia’s customers eventually earning returns on it — something Nvidia’s filings cannot show and its customers have not yet proven.

A $5 trillion valuation is not a number a company earns once. It is a number the future has to keep agreeing to pay.

Sources

• NVIDIA earnings release, first quarter fiscal 2027 (quarter ended April 26, 2026) — revenue $81.6B, data-centre revenue $75.2B, gross margin 74.9%, GAAP net income $58.3B, ~$91B guidance, zero-China assumption, Huang “AI factories” quote, Vera Rubin platform.
• NVIDIA Form 10-K, fiscal 2026, filed with the SEC — direct-customer concentration of 22% and 14% of total revenue; customer-concentration risk disclosures; fiscal 2026 revenue $215.9B.
• NVIDIA fiscal 2026 first-quarter filings — April 2025 H20 export-licence requirement; $4.5 billion H20 charge.
• CNBC, August 28, 2025 — two unnamed customers at 39% of quarterly revenue; CNBC/Financial Times, August 10, 2025 — the 15% China revenue-sharing arrangement.
• CNN and Al Jazeera, August 11, 2025 — export-licence arrangement; Chinese state-media warnings against the H20.
• CNBC, NBC News and Forbes, October 29, 2025 — first $5 trillion close.
• OpenAI–NVIDIA joint announcement, September 22, 2025 — 10-gigawatt letter of intent; up to $100 billion progressive investment.
• Fortune, February 2026 — Huang: the $100 billion “was never a commitment”; Bloomberg, 2026 — “How Microsoft, OpenAI and Nvidia Keep Paying Each Other.”
• Brookings Institution, 2026 — the U.S. exit from China’s AI chip market.

Figures are in U.S. dollars as reported. Fiscal years are Nvidia’s, ending late January. The $47 billion figure for the largest customer is TNN’s calculation (22% of $215.9B) and is identified as an estimate.