NVIDIA Q2 FY2027: revenue up 18% QoQ, operating cash flow down 52%

NVIDIA beat on revenue. It beat on earnings. It beat on data-center revenue. It guided the next quarter to $108 billion — above what Wall Street was modelling — and it did so while assuming it will sell nothing to China. Every headline number was a beat, and several were records.

That is a different puzzle from the one AMD posed three weeks earlier, where a genuine double beat sat next to one flat guidance line. The interesting question here is what the headline leaves out: after 106% revenue growth and a $108 billion guide, what is not visible on the first page of the release? The answer is not on the income statement. It is one statement over.

	Q2 FY27	Consensus
Revenue	$96.2B	~$92.2B	beat
Non-GAAP EPS	$2.22	~$2.10	beat
Data Center revenue	$89.0B	~$86.3B	beat
GAAP gross margin	75.0%	~74–75%	in line
Q3 revenue guide	$108.0B ±2%	below guide	raise

Revenue grew 106% year over year and 18% sequentially to a company record of $96.2 billion. Data Center revenue grew 117%, to $89.0 billion, and is now 92% of the entire company. GAAP gross margin was 75.0%, GAAP operating income was $63.7 billion — a 66% operating margin — and GAAP diluted EPS of $2.46 was up 128% on a year ago. On the non-GAAP basis the company now prefers, EPS was $2.22, up 120%.

Every one of those numbers is extraordinary. And two of them are not quite what they look like.

Did NVIDIA beat its own guidance in Q2 FY2027?

Yes. Revenue of $96.2 billion came in above the top of the $89.2–92.8 billion range NVIDIA guided to in its May 20, 2026 8-K — an outlook of $91.0 billion plus or minus 2%, with the range our arithmetic. GAAP gross margin of 75.0% was within the guided 74.9% plus or minus 50 basis points, and GAAP operating expenses of $8.41 billion came in just under the approximately $8.5 billion guided. Against the year-ago quarter, revenue rose 106% and GAAP diluted EPS 128%, to $2.46.

The number that isn’t chips: $7.8 billion

Start with the gap between the two EPS figures. GAAP diluted EPS was $2.46. Non-GAAP was $2.22. For almost every company you will ever read, non-GAAP EPS is the higher of the two — non-GAAP adds back stock compensation and one-off charges, so it flatters. Here it is lower. That inversion is a signal, and it is worth understanding.

The reconciliation in NVIDIA’s own release explains it in one line. GAAP “other income, net” was $7,773 million this quarter, of which $7,771 million was gains from equity securities. NVIDIA’s non-GAAP measure strips those gains out — so non-GAAP other income was just $300 million. Roughly $7.8 billion of pre-tax profit, and about $5.7 billion of after-tax net income, came not from selling GPUs but from the rising value of NVIDIA’s investment portfolio.

That portfolio is now enormous and growing fast. Marketable equity securities went from $12.9 billion at the start of the fiscal year to $42.8 billion. Non-marketable securities — private stakes — went from $22.3 billion to $51.2 billion. NVIDIA is holding roughly $94 billion of equity investments on a $320 billion balance sheet, and the mark-to-market on the public slice of that is now a material driver of GAAP earnings.

This is not an accounting trick, and it is not hidden — it is disclosed plainly and the non-GAAP number is specifically designed to remove it. But it changes how you read the headline. “NVIDIA earned $59.7 billion” is a GAAP fact that includes $7.8 billion of investment gains. “NVIDIA earned $53.9 billion from operations” is the cleaner statement of the chip business. Both are true. If those equity holdings fall next quarter, GAAP EPS takes the hit and the operating business will not have changed at all. Know which number you are holding.

One thing that makes NVIDIA’s non-GAAP unusually honest: beginning this fiscal year, it stopped excluding stock-based compensation. SBC of $2.0 billion is now inside the $2.22. Most large-cap technology companies still add it back; NVIDIA no longer does. So the non-GAAP figure here is more conservative than the label usually implies — the only large adjustments left are the equity-securities gains and about $0.2 billion of acquisition-related costs.

The number that grew slower than profit: cash

Now the statement the headline skips entirely.

	Q1 FY27	Q2 FY27	Change
Revenue	$81.6B	$96.2B	+18%
GAAP operating income	$53.5B	$63.7B	+19%
Operating cash flow	$50.3B	$24.1B	−52%
Free cash flow	$48.6B	$21.3B	−56%

Read that table twice. Revenue rose 18% sequentially. Operating income rose 19%. And operating cash flow fell by more than half, from $50.3 billion to $24.1 billion. Free cash flow fell from $48.6 billion to $21.3 billion. A company does not usually grow its profit a fifth and halve its cash generation in the same quarter — unless something on the balance sheet is absorbing the difference.

It is. Accounts receivable consumed $22.3 billion of cash in the quarter, and the receivable balance climbed to $63.1 billion from $38.5 billion at the start of the year. Inventories took another $5.8 billion. Put simply: NVIDIA booked the revenue and the profit, but a large and growing share of it is still owed to the company rather than collected. Add the working-capital build to the picture and the $24.1 billion of operating cash is what actually arrived.

There is an innocent reading and a cautious one, and honesty requires stating both.

• Innocent: demand is so far ahead of supply that shipments are back-end loaded within the quarter and large customers are on normal terms; receivables that big are simply what a $96-billion quarter growing 18% looks like, and they convert to cash next quarter.
• Cautious: a receivable balance rising faster than revenue is the classic early signature of demand being pulled forward, of customer concentration (a handful of buyers on extended terms), or of revenue recognised ahead of collection. It is not evidence of any of those — but it is the line where they would first appear.

You cannot resolve which it is from one quarter. You can mark it as the single most important thing to check in the Q3 report: does operating cash flow re-converge with operating income, or does the receivable balance keep outrunning revenue?

NVIDIA is now helping finance its own demand

The cash and the investment portfolio connect to the story that has quietly become the real debate about NVIDIA, and this quarter put hard numbers on it.

In the same three months, NVIDIA:

• returned about $26.0 billion to shareholders — $19.7 billion of buybacks and $6.0 billion of dividends — with $99.0 billion still authorised;
• raised about $24.9 billion of new debt, taking long-term debt from $7.5 billion at the start of the year to $32.4 billion;
• poured tens of billions into equity stakes (the $42.8 billion marketable and $51.2 billion non-marketable balances above); and
• announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise, in its words, “over $500 billion of third-party capital” to build AI infrastructure over time.

Line those up and a pattern appears that is worth naming without alarm: NVIDIA is increasingly part of the financing of the customers who buy its chips. It invests in AI builders; those builders raise capital; that capital buys NVIDIA hardware; NVIDIA recognises revenue and, when the stakes appreciate, books an investment gain on top. Each step is legitimate and each is disclosed. But it means a growing slice of NVIDIA’s demand and its GAAP profit is downstream of NVIDIA’s own balance sheet, rather than independent of it. That is a different risk profile from “the world buys our chips with its own money,” and it is the fair version of the circularity concern — not that it is improper, but that it reduces the independence between NVIDIA’s revenue and NVIDIA’s capital.

Gross margin is guided down, on purpose

One forward number moved the wrong way and management flagged it: third-quarter gross margin is guided to 74.0%, down from 75.0%. On a company guiding revenue up 12% sequentially, a half-to-full point of margin give-back is the mechanical cost of the Vera Rubin platform — NVIDIA’s new rack-scale generation — ramping into full production. Early yields on a new architecture and a richer mix of bought-in system content dilute the percentage even as the dollars grow. This is the same physics that flattened AMD’s margin, arriving at NVIDIA from a far higher starting point. At 74%, it is a rounding concern, not a thesis — but it is a reminder that even NVIDIA does not get rack-scale systems at chip-scale margins.

The China line: guiding to $108 billion with a zero in it

The most quietly significant sentence in the release is in the outlook: “NVIDIA is not assuming any Data Center compute revenue from China in its outlook.”

Read that carefully. The $108 billion Q3 guide — up about 12% sequentially and roughly 89% year over year at the midpoint — is built with zero contribution from data-center compute in China. A market that was once a meaningful part of the franchise is now modelled at nothing, a casualty of the export-control regime that has run through NVIDIA’s numbers for over a year. There is an optimistic corollary: if any China data-center compute is permitted and materialises, it is upside the guide does not contain. But the base case is now a China-free base case, and the company is telling you the $108 billion does not need China to be real.

For the year-over-year comparison, this matters in the other direction too. A year ago, H20-related charges distorted the accounts; this quarter the H20 impact was insignificant, and the year-ago quarter carried only a small release. So unlike some AI-era comparisons, the 106% and 117% growth rates here are not flattered by lapping a prior-year write-off. They are clean.

Segments: one engine, and everything else in one box

NVIDIA now reports its business in two buckets rather than five:

	Q2 FY27 revenue	YoY	QoQ
Data Center	$89.0B	+117%	+18%
Edge Computing	$7.2B	+27%	+13%

The consumer and edge businesses — what used to be reported as gaming, professional visualization, automotive and robotics — are now consolidated into a single Edge Computing line of $7.2 billion. That is a 27%-growth, $29-billion-a-year business that at any other company would be the whole story, and here it is 7.5% of revenue and a rounding item in the narrative. The concentration is the point: Data Center is 92% of NVIDIA, and 92% of the investment case rests on the durability of AI infrastructure spending by a small number of very large buyers. There is no diversification in this revenue mix to cushion a pause. That is not a criticism of the quarter — it is the structural fact that any position in the stock is taking on.

The bear case, stated fairly

Not “the numbers are bad” — they are historic. The narrower, defensible version:

1. A widening slice of GAAP profit is investment gains, not chips. $7.8 billion this quarter, $23.7 billion across the first half of the year. If those marks reverse, GAAP earnings fall while the operating business is unchanged.
2. Cash conversion broke stride. Operating cash flow fell 52% sequentially as receivables grew $22 billion. The profit is real; the collection has lagged, and that is the line where a demand pull-forward or concentration risk would first show.
3. NVIDIA is financing its own demand. New debt, a near-$100-billion investment book, and $500 billion of customer-financing partnerships reduce the independence between its revenue and its balance sheet.
4. Growth is decelerating off a vast base. 106% becomes ~89% guided; the law of large numbers is arriving.
5. Margin is guided down, and 92% concentration in Data Center leaves nothing to absorb a hyperscaler pause.

None of that is refuted by this quarter. Together they are the reasons to read past the headline: the revenue and guidance lines are strong, and the cash and the investment gains are where the caveats sit.

What to check at the Q3 report

Specific, falsifiable, in rough order of importance:

• Operating cash flow versus operating income. The single most important line. Does cash re-converge with profit, or does the receivable balance keep outrunning revenue? This is the tell for whether Q2’s gap was timing or something structural.
• Accounts receivable and inventory on the balance sheet. The leading indicators, ahead of both margin and the income statement.
• Gross margin against the 74% guide. Below it means the Vera Rubin mix is diluting faster than modelled; at or above means the ramp is on plan.
• Revenue against $108 billion, and whether the China assumption changed — any dollar of China data-center compute is upside the guide excludes today.
• Equity-securities gains in GAAP net income. Whether the gap between GAAP and non-GAAP EPS widens further as the investment book grows.
• Any softening of the AI-infrastructure demand language, especially from the named hyperscalers and the new financing partners.

The read

NVIDIA delivered the largest quarter in the history of the semiconductor industry: $96.2 billion of revenue, 117% data-center growth, a 75% gross margin, and a $108 billion guide that does not even count China. The operating business is not the question.

The useful way to hold this report is to separate the headline from the mechanics. The headline — record revenue, record profit, a guide above the Street — is the part everyone sees. The mechanics are where the new information is: $7.8 billion of the profit came from investments rather than chips, operating cash flow fell by half as receivables ballooned, and the company is increasingly financing the demand it reports. None of those undoes an exceptional quarter. All of them are things you can only see by reading past the first line of the release.

---

Sources and method. Company figures in this piece — revenue and its 106%/18% growth, Data Center and Edge Computing revenue, GAAP and non-GAAP gross margin, operating expenses, operating income, other income and the $7,771 million of equity-securities gains within it, net income, GAAP and non-GAAP diluted EPS, the effective tax rate, operating cash flow, capital expenditure and free cash flow, the accounts-receivable and inventory changes, the marketable and non-marketable securities balances, debt issued and outstanding, the buybacks, dividend and remaining authorisation, the third-quarter revenue, gross-margin and operating-expense guidance, the China assumption, and the H20 footnote — are NVIDIA’s own disclosures from its Q2 FY2027 results release (Form 8-K, Exhibit 99.1) and Form 10-Q, both filed 2026-08-26 for the quarter ended 2026-07-26, reported after the close. Prior-year comparatives are from NVIDIA’s Q2 FY2026 results release. Every figure above is traceable to the primary documents linked in the sources; the segment sums, the GAAP-to-non-GAAP reconciliation, and free cash flow (operating cash flow less capital expenditure and principal payments) were checked against the release and reconcile within rounding.

Not company figures, and attributed accordingly: the consensus estimates (revenue ~$92.2 billion, non-GAAP EPS ~$2.10, Data Center ~$86.3 billion) are from CNBC’s live coverage and the pre-earnings previews cited above, and the description of the $108 billion Q3 guide as above the Street’s expectation follows CNBC’s report that the guidance “sailed past analyst estimates.” The interpretation — that the receivable build explains the cash-flow gap, that the equity-securities gains flatter GAAP earnings, that NVIDIA is increasingly financing its own demand, and that these are the points worth scrutinising — is our analysis, not a company statement, and is not a forecast. Separately from those estimates, the guidance verdict compares the results only with NVIDIA’s own previously issued outlook (Form 8-K, Exhibit 99.1, filed 2026-05-20) and with the year-ago quarter. Our arithmetic, labelled as such: the $89.2–92.8 billion revenue range implied by the $91.0 billion plus-or-minus-2% outlook. This piece makes no claim about the share-price reaction.

Related: AMD Q2 2026, the same rack-scale margin physics one rung down the market, and Microsoft Q4 FY2026, one of the hyperscalers whose capital budget is the other side of NVIDIA’s $89 billion.

Tradune shows what you actually hold across every account, and whether a position has beaten a plain index fund since you bought it. Educational analysis, not investment advice.