Nvidia reported $96.221 billion of revenue for the July quarter, up 105.9% from a year ago and 17.9% from April, against its own guide of $91.0 billion plus or minus 2%. Non-GAAP earnings per share of $2.22 beat the $2.10 consensus by 5.7%. Data Center revenue was $89.0 billion, up 117%. On any ordinary reading this was the fourth straight quarter in which growth accelerated, at a scale where acceleration should be arithmetically hard.
Almost none of that was the news. Two things said on the call were, and they point in opposite directions.
The first: Nvidia guided a full fiscal year it has not started, which it has never done. Revenue in fiscal 2028 is expected to grow approximately 70%. The second: gross margin, which has been climbing for two years and printed 75.0% this quarter, will fall to 74% next quarter, bottom at 71–72% in the January quarter, and settle at 72–73% across fiscal 2028. Memory prices are the reason, and the company said so without being asked.
The headline numbers
| Metric | Q2 FY27 | Q1 FY27 | Q2 FY26 | YoY |
|---|---|---|---|---|
| Revenue | $96.221B | $81.615B | $46.743B | +105.9% |
| Data Center | $89.023B | $75.246B | $41.096B | +116.6% |
| — Hyperscale | $48.710B | $43.050B | $24.168B | +101.5% |
| — AI clouds, industrial & enterprise | $40.313B | $32.196B | $16.928B | +138.1% |
| Edge Computing | $7.198B | $6.369B | $5.647B | +27.5% |
| Gross margin | 75.0% | 74.9% | 72.4% | +2.6 pts |
| Operating income | $63.734B | $53.536B | $28.440B | +124.1% |
| Operating margin | 66.2% | 65.6% | 60.8% | +5.4 pts |
| Diluted EPS — GAAP | $2.46 | $2.39 | $1.08 | +127.8% |
| Diluted EPS — non-GAAP | $2.22 | $1.87 | $1.01 | +119.8% |
| Operating cash flow | $24.077B | $50.344B | $15.365B | +56.7% |
| Free cash flow | $21.400B | $48.587B | $13.471B | +58.9% |
One line in that table needs reading twice: GAAP earnings per share are higher than non-GAAP earnings per share. That is not a typo and it is not usual. Nvidia's GAAP other income carried $7.8 billion of gains on equity securities this quarter, which the non-GAAP line strips out. The company's investment book — non-marketable securities went from $22.3 billion in January to $51.2 billion in July — has become large enough to move the headline number on its own. Anyone comparing this quarter's $2.46 against a consensus struck on the $2.22 basis will read a beat roughly three times the real one.
A second basis change matters as much: from the April quarter, Nvidia's non-GAAP figures no longer exclude stock-based compensation, and the history has been restated to match. Non-GAAP $2.22 is not comparable with pre-2027 non-GAAP prints unless the older number is adjusted the same way.
The growth is coming from the half nobody models
Nvidia now splits Data Center into hyperscale and everything else — AI clouds, industrial and enterprise, which it calls ACIE. The half that is easy to model, hyperscale, grew 13% sequentially. The half that is hard to model grew 25% sequentially and 138% year over year, and the company expects it to settle at roughly half of Data Center revenue.
That is where the growth rate is, and it is the part with no public capex disclosures behind it. Sovereign business alone grew 35% sequentially and more than tripled year over year. Nvidia's NeoCloud partners are expected to exit this year with 8 gigawatts installed, against about 3 gigawatts at the end of 2025. Jensen Huang's framing on the call: "That part of the world is invisible to everybody. The reason for that is because they do not buy custom chips. They do not buy chips one at a time. They really need an entire factory platform built for them."
The strategic argument underneath it is a number Nvidia has now put on the record three ways. Revenue opportunity per gigawatt of data centre: $18 billion with Hopper, $25 billion with Grace Blackwell, $40 billion with Vera Rubin. Every generation, the same gigawatt of land and power buys more Nvidia. That is the company's answer to the question of what happens when power, not silicon, becomes the constraint.
The 70% is a supply number, not a demand number
Nvidia has never guided a year ahead, and the reason it gave for starting is logistical rather than promotional: everybody in the chain — customers, suppliers, shareholders — is committing capital on a multi-year horizon and needs the same picture. CFO Colette Kress put the gap plainly:
Customers' forecasts point to our growth doubling next year. However, as I mentioned earlier, we expect to grow approximately 70% as we are supply-constrained.
Asked what separates the two, Huang did not soften it: "Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%. We are going to continue to work with our supply chain to increase on that." And later, more bluntly: "at this moment, we have supply for 70%. We have more supply than 70%, but about 70%. Our demand is much higher than that, and we've got to go work hard, or we're going to be disappointing customers."
Supply is expected to remain the binding constraint at least through the end of fiscal 2028.
It is worth sizing what 70% means. The first half of fiscal 2027 produced $177.8 billion, and the October quarter is guided to $108.0 billion. Hold the January quarter flat with October — conservative, since the company says hyperscale re-accelerates into it — and fiscal 2027 lands near $394 billion. Roughly 70% on that is about $670 billion of revenue in fiscal 2028, from a company that did $46.7 billion in a quarter fourteen months ago.
The margin reset is the part that costs money
Nvidia volunteered the bad news, which is itself unusual:
We are experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year. As a result, we are resetting expectations today.
The path: 74% ±50bp in October, 71–72% at the January trough, 72–73% for fiscal 2028 once "executed price increases take effect" in the April quarter. Against 75.0% today, the trough is roughly a 350 basis point give-back — about $3.8 billion of gross profit a quarter at this revenue scale, before the price increases claw any of it back.
The balance sheet had already said this. Supply and capacity commitments went from $119 billion last quarter to $279 billion, which the company attributes primarily to procuring memory. Total future commitments now stand at $366 billion. That is not a company expecting memory to get cheaper; it is a company buying three years of it at whatever it costs.
Kress's defence of the cost line is genuinely interesting rather than defensive: memory scarcity, she argued, "is being driven in large part by the AI build-out itself," so unlike an ordinary input-cost shock it is a symptom of the same demand that is driving the revenue line. That is a coherent argument. It is also unfalsifiable in the short run, and it does not change the gross profit.
Cash conversion broke, on purpose
Operating cash flow fell to $24.1 billion from $50.3 billion in April, while net income rose. The gap is working capital: a $22.3 billion build in receivables and a $5.8 billion build in inventory. Days sales outstanding went from 45 to 60, which Nvidia attributes to extended payment terms on large multi-quarter agreements with investment-grade customers. Inventory is at $31.6 billion, ahead of the Vera Rubin ramp.
Both are defensible and both are real. A quarter where free cash flow is $21.4 billion on $59.7 billion of net income is a quarter where a third of the profit is sitting in someone else's payment terms. The company still returned a record $26.0 billion — $19.7 billion of buybacks and $6.0 billion of dividends — and funded the gap by issuing $25 billion of senior unsecured notes. Long-term debt went from $7.5 billion in January to $32.4 billion.
The financing question now has Nvidia's own numbers attached
Kress addressed the "circular financing" characterisation directly, and in doing so supplied the figures to size it: nearly $50 billion invested in frontier AI labs; partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise over $500 billion of third-party capital; OpenAI commitments representing approximately 12 gigawatts of Nvidia compute through 2030; selective credit support for nearly 2 gigawatts for another lab. And then the sentence that matters most:
We recognize the scale of this support, and we know some will call this circular financing. We see it differently. […] we expect demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year.
Separately, guarantees. Nvidia's maximum gross exposure is now $108.5 billion — $3.5 billion of land, power and shell guarantees for AI cloud partners, plus $105 billion entered in August with SB Energy covering roughly 4.25 gigawatts at the PORTS-Pike campus in Ohio, which will host Nvidia infrastructure under 20-year leases to OpenAI. Nvidia's own estimate is that each generation deployed there is about 1.5 million GPUs, or $150–200 billion of Nvidia revenue.
None of that is hidden and none of it is obviously wrong. It is, however, a materially different risk profile from the one Nvidia had a year ago, and it is now quantified enough that it can be argued about honestly. We are writing that argument up separately rather than resolving it in a results piece.
What it does to our model
Two things in this print land directly on our Nvidia model, which was last struck on 17 August.
Net cash is stale by design and now stale by size. The model carries $41.8 billion. After the $25 billion note issuance, cash and marketable debt securities of $56.6 billion against $33.4 billion of total debt leave roughly $23.2 billion — an $18.6 billion swing in one quarter, most of it a deliberate financing decision rather than an operating one.
The margin assumption is now contradicted by the company. Our data-centre compute vertical runs a 76% EBITDA margin. Nvidia has just guided gross margin to a 71–72% trough, and EBITDA cannot exceed gross profit. That assumption has to come down, and the revision deserves its own piece with the arithmetic shown rather than a line in an earnings recap.
What to watch
- Whether the January trough is the trough. 71–72% is a forecast about memory contracts that have not all been signed. Kress said prices are "headed even higher into next year."
- Whether the April price increases stick. The entire 72–73% fiscal-2028 recovery depends on customers absorbing them, at a moment when those same customers are being asked to commit to multi-year capacity.
- Receivables. 60 days from 45 is explainable once. Two more quarters of that and the revenue line and the cash line are telling different stories.
- Vera Rubin's ramp. Guided at about 20% of Data Center revenue in October, and described as the fastest product ramp in the company's history, with purchase orders from every major hyperscaler, AI cloud and system OEM.
- China, which is now zero. Hopper shipments were under 1% of Data Center revenue this quarter and there is no China data-centre compute revenue in the outlook at all — so any change there is upside the guide does not carry.
What we learned
- Growth accelerated again, for a fourth consecutive quarter. Revenue of $96.221B, +105.9% year over year and +17.9% sequentially, against a $91.0B guide. Data Center was $89.0B, up 117%. On a trailing-twelve-month basis Nvidia scores 125 on the Rule of 40 — 83.4% revenue growth plus a 41.9% free-cash-flow margin.
- The company guided a full year ahead for the first time, and capped it at supply. Fiscal 2028 revenue is expected to grow approximately 70%, while the CFO said customer forecasts point to a doubling. Supply is expected to bind at least through the end of fiscal 2028.
- Gross margin is going down, and Nvidia said so unprompted. 75.0% today → 74% in October → a 71–72% trough in January → 72–73% for fiscal 2028. Memory is the named cause, and supply commitments went from $119B to $279B in one quarter, primarily to procure it.
- A third of the quarter's profit did not convert to cash. Operating cash flow fell to $24.1B from $50.3B on a $22.3B receivables build and a $5.8B inventory build; DSO went 45 → 60 days. Nvidia still returned a record $26.0B and issued $25B of notes to do it.
- GAAP EPS ($2.46) exceeded non-GAAP EPS ($2.22) because $7.8B of equity-securities gains sit in GAAP other income — and the CFO put a number on the related exposure: labs whose purchases lean on Nvidia's balance sheet are roughly a quarter of next year's business, against $108.5B of maximum guarantee exposure.
Nvidia (NASDAQ: NVDA) reported its fiscal second quarter — the three months ended 26 July 2026 — after the US close on 26 August 2026. Revenue was $96.221B (+105.9% year over year), GAAP diluted EPS $2.46 and non-GAAP diluted EPS $2.22. Quotations are from the earnings call held the same afternoon, 26 August 2026. Operating margin, effective tax rate, free cash flow (operating cash flow less capital expenditure) and the fiscal-2028 revenue estimate implied by the ~70% growth guide are ours, derived from the reported figures; every other number is the company's. Shares closed at $209.66 on 26 August, down 1.59%, and traded near $218.90 after hours, up about 4.4%. The full figure set for the quarter is on the Nvidia Q2 FY2027 earnings page.
