Nvidia's profits are real, audited and fully disclosed — and the interesting question was never whether they exist. It is what they are made of. In fiscal 2026, Nvidia reported revenue of $21
Nvidia's profits are real, audited and fully disclosed — and the interesting question was never whether they exist. It is what they are made of. In fiscal 2026, Nvidia reported revenue of $215.9 billion and GAAP net income of $120.1 billion. But $8.918 billion of that profit did not come from selling a single chip. It came from marking up the value of equity stakes Nvidia holds in other companies — many of them the same AI firms that buy its GPUs. That single line is larger than Nvidia's entire $6.386 billion stock-based compensation expense, and it produced an oddity that almost never occurs at a company this size: GAAP earnings per share ($4.90) came in higher than non-GAAP earnings per share ($4.77). The adjusted number was the conservative one. That inversion is the whole story.
Here is the synthesis nobody is putting together. Strip Nvidia's declared profit into its components and roughly one-seventh of it is not chip margin. The $8.918 billion of equity gains represents 7.4% of GAAP net income. Separately, Nvidia paid $21.383 billion of tax on $141.45 billion of pre-tax income — an effective rate of 15.1%, well below the 21% US federal statutory rate. Had it paid 21%, the bill would have been about $29.7 billion, roughly $8.3 billion higher. Add the equity gains to the tax advantage and you get about $17.2 billion, or 14.4% of declared net income, arriving from sources that are not the sale of a GPU. None of this is improper. All of it is in the filings. But it means the answer to "how much money does Nvidia actually make?" depends entirely on what you count.
Key Facts
• FY2026 revenue: $215.938 billion, up 65% year-on-year — NVIDIA, Q4 FY2026 results • GAAP net income $120.067 billion; non-GAAP net income $116.997 billion — NVIDIA • Gains from non-marketable and publicly-held equity securities: $8.918 billion — NVIDIA • Stock-based compensation: $6.386 billion — smaller than the equity gains, which is why GAAP EPS ($4.90) beat non-GAAP ($4.77) • Income tax expense $21.383 billion on $141.45 billion pre-tax — an effective rate of 15.1% • Free cash flow $96.575 billion — equal to 80.4% of GAAP net income, against Nvidia's own long-run average of ~98% since fiscal 2018 • Nvidia committed more than $40 billion to AI equity investments in the first four months of calendar 2026, including $30 billion into OpenAI — CNBC
Start with the operating business, because it is genuinely extraordinary and the rest of this article should not obscure that.
On $215.938 billion of revenue, Nvidia posted a GAAP gross margin of 71.1%. That implies cost of goods sold of roughly $62.4 billion — the wafers, the packaging, the memory, everything TSMC and its suppliers charge. Operating income was $130.387 billion, which means total operating expenses — all R&D, all sales, all general and administrative costs across the entire company — came to about $23.1 billion.
Sit with that ratio for a second. Nvidia spends around $23 billion running itself and generates $130 billion of operating income from it. The operating margin is 60.4%. There is no comparable business at this scale in the history of semiconductors. Data Center revenue alone was $193.7 billion, up 68%, which is 89.7% of the company — Nvidia is now, functionally, a data-centre company that still sells some gaming cards.
"Computing demand is growing exponentially — the agentic AI inflection point has arrived," said Jensen Huang, Founder and Chief Executive Officer of NVIDIA (NVIDIA).
He is not exaggerating, and the cost base proves it. This is a company that has to spend almost nothing incrementally to serve demand that is growing 65% a year. The gross margin is not an accounting artefact; it is what happens when you hold a near-monopoly on the input everyone needs.
Which is exactly why the $8.918 billion of equity gains is worth interrogating. A business earning 60% operating margins does not need to dress up its earnings. So why does a line item that has nothing to do with chips sit inside the profit number at all?
Because Nvidia has become one of the largest venture investors on earth, and it invests in its own customers.
In the first four months of calendar 2026, Nvidia committed more than $40 billion to AI equity investments. The largest single line was $30 billion into OpenAI in late February. It put $2 billion into CoreWeave in January and $2 billion into Nebius in March. It has participated in rounds for Anthropic and xAI. It holds an option to invest $2.1 billion in IREN at $70 per share.
The mechanism is straightforward, and it is legal. Nvidia takes an equity position in an AI company. That company signs a long-term GPU purchase commitment with Nvidia. Nvidia books the GPU sales as revenue at a 71% gross margin — and, separately, marks up the value of its equity stake as that company's valuation rises, booking the gain through the income statement. The same capital cycle touches the profit line twice.
We have covered both ends of this loop directly. CoreWeave carries $35 billion of debt and a $99.4 billion backlog built to serve compute it buys from Nvidia, which owns a slice of it. IREN's $3.4 billion Nvidia contract came bundled with that $70-a-share equity option. Nebius took $2 billion and signed on for AI infrastructure deployment. Each of these is a real business with real demand. Each is also, in part, a customer whose purchasing power Nvidia helped fund.
Nvidia's own framing of the numbers is far more modest than the headlines. The company has rejected the widely circulated "$610 billion circular financing" claim as unfounded, stating that its strategic investments totalled around $3.7 billion in the third quarter and roughly $4.7 billion year-to-date. That is an order of magnitude below the $40 billion-plus of commitments reported by CNBC — and the gap is definitional, not deceptive. Commitments are not the same as deployed balance-sheet investments, and Nvidia's fiscal year does not line up with the calendar year. But that definitional fog is precisely why this argument cannot be settled from the outside.
The cleanest test of earnings quality is not the income statement. It is whether the profit shows up as cash.
Nvidia has made this argument itself, and it is a strong one: the company points to long-term free cash flow running at roughly 98% of GAAP net income since fiscal 2018 as evidence that its revenue is real and its accounting is conservative. For most of its history, that has been true and it is a genuinely powerful rebuttal.
In fiscal 2026, the ratio was 80.4%.
MetricFY2026What it tells youRevenue$215.938bn+65% year-on-yearCost of goods sold (derived)~$62.4bnImplied by the 71.1% gross marginTotal operating expenses (derived)~$23.1bnAll R&D, sales and admin, company-wideOperating income$130.387bn60.4% operating marginNon-operating income (derived)~$11.1bnIncludes the $8.918bn of equity gainsIncome tax$21.383bn15.1% effective rate vs 21% US statutoryGAAP net income$120.067bnThe declared numberFree cash flow$96.575bn80.4% of net income — vs ~98% long-runSources: NVIDIA Q4 FY2026 results (revenue, margins, income, tax, FCF). Derived figures calculated from NVIDIA's reported gross margin and operating income. Long-run FCF-to-net-income ratio per NVIDIA's own analyst memo.
So where did the $23.5 billion go? The gap between $120.067 billion of declared net income and $96.575 billion of free cash flow is $23.5 billion. Part of it is ordinary: a business growing 65% a year is building inventory and receivables, and that consumes cash. But part of it is structural. The $8.918 billion of equity gains is a non-cash markup — Nvidia did not sell those stakes, it revalued them. That profit exists on the income statement and not in the bank. This does not make the earnings fake. It makes them less cash-backed than Nvidia's own historical benchmark, in the specific year when its investment portfolio ballooned. The company deployed $40.086 billion on buybacks and $974 million on dividends in the same period, so the cash it does generate is very real. But the widening of that gap is the number a sceptic should be watching, not the headline EPS.
This is where the argument turned hostile, and where precision matters most.
Michael Burry — the investor best known for shorting subprime mortgages — alleged that AI infrastructure companies have extended the useful-life assumptions on their hardware in a way that "artificially boosts earnings — one of the more common frauds of the modern era." He estimated that depreciation across the industry could be understated by $176 billion between 2026 and 2028, and singled out Meta and Oracle as potentially overstating profits by 20.8% and 26.9% respectively by 2028.
The mechanism he is describing is real and simple. If a GPU is depreciated over six years instead of three, the annual expense halves and reported profit rises — without a single extra dollar coming in. Whether six years is honest depends entirely on how quickly Nvidia's own product cycle makes the previous generation obsolete, which is an awkward question for the company whose roadmap sets that pace.
Nvidia pushed back hard and in detail, circulating a memo to analysts that named Burry directly and rebutted the fraud framing with cash-conversion data. The sell-side largely sided with the company. "The narrative of systemic fraud is inconsistent with NVIDIA's fundamentals and the structure of this investment cycle," said Simon Leopold, analyst at Raymond James (Investing.com).
Leopold is right on the narrow point, and it needs saying plainly: there is no evidence Nvidia has committed fraud. Its margins are verified by cash. Its revenue is contracted. Its disclosures exceed what the rules require. The depreciation allegation is aimed principally at Nvidia's customers, not at Nvidia — and if those customers are overstating profits, the risk to Nvidia is second-order: it lands on their ability to keep buying, not on Nvidia's books.
But "not fraud" and "not circular" are different claims, and only the first has been established.
The non-GAAP number is about to get worse, on purpose. From the first quarter of fiscal 2027, Nvidia will stop excluding stock-based compensation from its non-GAAP measures, calling it "a foundational component" of its compensation programme. That is a genuinely conservative move — it removes a $6.4 billion add-back that most of the sector still takes — and it is the single strongest counter-argument to anyone claiming Nvidia flatters its numbers. Watch whether peers follow. Most will not.
Expect the equity gains line to become the story. With more than $40 billion committed to AI equity in four months, the mark-to-market swings on that portfolio will grow. In a rising market they inflate GAAP profit; in a falling one they will reverse, and Nvidia will report a quarter where chip demand was fine and earnings missed anyway. That quarter will be badly misread, and it is coming.
The tax rate is the quiet risk. A 15.1% effective rate on $141 billion of pre-tax income is a $8.3 billion annual advantage over the statutory rate. It is legal, it is disclosed, and it depends on a global tax architecture that is under active political pressure. A shift there hits net income directly, and no amount of GPU demand offsets it.
How much money does Nvidia actually make? In fiscal 2026, Nvidia reported revenue of $215.938 billion and GAAP net income of $120.067 billion. Free cash flow was $96.575 billion — 80.4% of declared net income. The gap is largely non-cash items, including $8.918 billion of gains on equity stakes that were revalued, not sold.
Why was Nvidia's GAAP EPS higher than its non-GAAP EPS? Because its $8.918 billion of equity investment gains exceeded its $6.386 billion of stock-based compensation. Non-GAAP strips out both. With the gains larger than the add-back, the adjusted figure ($4.77) came in below GAAP ($4.90) — an inversion that is very rare at this scale.
What is Nvidia's effective tax rate? Roughly 15.1% in fiscal 2026 — $21.383 billion of tax on $141.45 billion of pre-tax income. That is below the 21% US federal statutory rate, worth about $8.3 billion a year. Nvidia guides to a 17–19% GAAP and non-GAAP tax rate for fiscal 2027.
Is Nvidia's "circular financing" real? Nvidia invests in AI companies that then buy its GPUs, and it books both the chip revenue and the equity markup. That loop is real and disclosed. Nvidia rejects the "$610 billion" characterisation, stating strategic investments of about $3.7 billion in Q3 and $4.7 billion year-to-date — far below the $40 billion of commitments reported externally. The gap is definitional.
Is Nvidia committing accounting fraud? There is no evidence of it. Michael Burry's depreciation allegation targets AI infrastructure buyers — Meta and Oracle were named — rather than Nvidia itself, and Nvidia's cash conversion has historically run at about 98% of GAAP net income. The legitimate criticism is about the quality and durability of earnings, not their existence.
This article is informational analysis only and is not investment advice. All figures are drawn from NVIDIA's published fiscal 2026 results and other cited sources; derived figures are calculated from reported data and labelled as such. Nothing here alleges wrongdoing by any company or individual. Equities are volatile and can lose value rapidly. Do your own research and consult a regulated financial adviser before making any investment decision.
