The AI Capex Scorecard Inside a 96 Billion Dollar Nvidia Quarter

Nvidia's 96.2 billion dollar quarter is the clearest AI-capex scorecard yet published — a map of who owns the compute grid and who merely rents time on it.

The Print: 96.2 Billion Dollars in a Single Quarter

On 26 August 2026 Nvidia reported revenue of 96.2 billion dollars for the quarter ended 26 July 2026 — up 106% year on year and 18% on the prior quarter, according to the NVIDIA newsroom release. Data Center alone brought in 89.0 billion dollars, up 117% year on year and roughly 92% of the total, per the NVIDIA investor relations statement.

The split inside Data Center matters more than the headline. Hyperscale customers accounted for about 48.7 billion dollars; a second bucket Nvidia labels AI Clouds, Industrial and Enterprise added about 40.3 billion. GAAP gross margin held at 75.0%, and GAAP net income reached roughly 59.7 billion dollars — up 126% from the 26.4 billion dollars Nvidia posted a year earlier, per StockTitan's breakdown. Nvidia returned 26.0 billion dollars to shareholders in the quarter through buybacks and dividends.

One oddity earns a mention. GAAP diluted earnings came in at 2.46 dollars a share, above the non-GAAP figure of 2.22 dollars — an inversion driven by a favourable one-time tax item, not by weaker operations. CNBC's earnings coverage shows the 96.2 billion dollar print clearing the roughly 92 billion dollar analyst estimate with room to spare.

AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue. And demand is accelerating.

— — Jensen Huang, NVIDIA CEO

Compute Is Revenue — and Nvidia Wants to Be Paid Twice

Jensen Huang's line on the call was blunt: compute is now revenue. The evidence sits in the platform maths. Revenue captured per gigawatt of deployed compute rises each hardware generation — about 18 billion dollars on Hopper, about 25 billion on Grace Blackwell, and about 40 billion on Vera Rubin, according to Nvidia's own figures.

Nvidia has gone further than selling silicon. Alongside the results, Nvidia announced financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise over 500 billion dollars of third-party capital for AI-compute build-outs. CFO Colette Kress noted that take-or-pay minimum-revenue guarantees mean Nvidia effectively gets paid twice — once on the hardware sale, again on the rental upside. Analysis of Blackwell and Vera Rubin visibility points to cumulative demand near one trillion dollars through 2027.

The Guidance Hides a Warning

Nvidia guided Q3 FY2027 revenue to 108.0 billion dollars, plus or minus 2%, with gross margin at 74.0%. The striking part sits in the assumption. The outlook assumes zero Data Center compute revenue from China — in Nvidia's own words, the company is not assuming any Data Center compute revenue from China. A geopolitical fact becomes a line item.

Then came the number that moved the stock. Kress guided roughly 70% revenue growth for fiscal 2028, far above the roughly 44% Wall Street consensus, according to CFO Dive — a forecast that, CNBC reports, would put Nvidia on track to become technology's second-largest company by revenue. Huang framed 70% as a supply ceiling, not a demand ceiling, constrained by memory supply. Vera Rubin is expected to reach about 20% of Data Center revenue in Q3 FY2027, with partner racks already running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.

Our demand is much higher than that.

— — Jensen Huang, on the 70% growth forecast

The Scorecard Nobody Reads: Who Is On the Grid

Here the scorecard turns into a map. 89 billion dollars a quarter of AI compute flows overwhelmingly to United States hyperscalers and AI-cloud renters. The compute-is-revenue model — hardware sold, then rented back under take-or-pay contracts — turns compute into a rent-extracting asset class. China, zeroed out of Nvidia's own guidance, shows how fast access can be switched off. Read honestly, the AI capex scorecard records not just earnings but access.

Here is the case for Emergent Intelligence (EI) — the dignity-first frame I use for what the world more commonly calls AI. The Global South does not own the factories. Zambia and South Africa, at best, rent capacity priced in 40-billion-dollar-per-gigawatt economics, and more often sit off the map entirely. An AI epoch financed by over 500 billion dollars of take-or-pay contracts entrenches compute-access asymmetry as structural, not incidental. Infrastructure access is a precondition for agency — not a luxury bought after the fact.

The scorecard is not just an earnings report. The scorecard is a map of who holds the grid — and dignity begins with a seat on it.

Frequently Asked Questions

These are the questions people are asking about Nvidia's Q2 FY2027 result. Short answers follow, drawn from the NVIDIA press release, the earnings call, and independent coverage.

What is Nvidia's Q2 FY2027 AI-capex scorecard?

In short, the scorecard is Nvidia's Q2 FY2027 result read as a measure of the AI build-out: 96.2 billion dollars in revenue, 89.0 billion from Data Center. Research and Nvidia's own data show Data Center at roughly 92% of the total, up 117% year on year.

How does Nvidia's compute-is-revenue model work?

Simply put, Nvidia sells the hardware and then captures rental upside as customers deploy the same compute under take-or-pay contracts. Nvidia's figures show revenue per gigawatt rising from about 18 billion dollars on Hopper to about 40 billion on Vera Rubin, according to the earnings call.

Why is the China assumption significant?

The key is exposure. Nvidia's Q3 FY2027 guidance of 108.0 billion dollars assumes zero Data Center compute revenue from China. Analysis shows how a single geopolitical decision can be written straight into a corporate forecast.

Who is Jensen Huang and what did Huang forecast?

In other words, Jensen Huang is Nvidia's chief executive, and Huang backed a forecast of roughly 70% revenue growth for fiscal 2028 — well above the 44% consensus. Evidence from the call, reported by CFO Dive, frames the ceiling as memory supply, not demand.

What are the risks of the AI-capex build-out?

The answer is concentration. Data reveals 89 billion dollars a quarter flowing to a handful of United States hyperscalers, with the Global South largely renting or excluded. Research on compute sovereignty shows infrastructure access shaping who gets agency in the AI epoch.

Sources:

NVIDIA newsroom — Q2 FY2027 results · NVIDIA investor relations release · CNBC — Q2 FY2027 earnings coverage · CNBC — 70% growth, track to tech's No. 2 · CFO Dive — 70% FY2028 revenue forecast · StockTitan — balance-sheet detail · NVIDIA — 500 billion dollar financing platforms · Related on this site: The Rubin AI Infrastructure Regime · Africa's AI Data-Centre Sovereignty · 600 Billion, Zero Consensus