NVIDIA's $96.2 Billion Quarter: When Compute Turns Into Revenue

$96.2 billion of revenue and $63.7 billion of operating income, in a single quarter. That is NVIDIA’s second quarter of fiscal 2027. The AI capex-peak argument that weighed on the market a year ago is hard to hold against one segment — data center — booking $89 billion, up 117% from a year earlier.

The interesting part of this print is not the size of the surprise but a shift in structure. AI compute has moved from something that looks like R&D spending to something that looks like recurring cash flow. That was the thrust of Jensen Huang’s line on the call: compute is revenue.

The 75% margin behind $89 billion of data center revenue

Total second-quarter revenue of $96,221 million was more than double the $46,743 million reported a year earlier, and up 18% sequentially. Of that, $89 billion came from data center. Not gaming, not professional visualization — a single block of AI accelerators and networking gear made up roughly 92% of company revenue.

GAAP and non-GAAP gross margin both landed at 75.0%. Holding that level while high-bandwidth memory prices are climbing says pricing power still sits with the supplier. Operating margin rose to 66.2%.

The net income line needs unpacking. GAAP net income of $59,688 million includes about $7.8 billion of gains on equity securities the company holds. Non-GAAP net income, closer to the operating business, was $53,954 million — still up 118% year over year. Diluted EPS was $2.46 GAAP and $2.22 non-GAAP.

Capital returns set a quarterly record: $19,732 million of buybacks plus $6,047 million of dividends, or $26 billion returned, with $99 billion left on the repurchase authorization.

What “compute is revenue” actually means

Third-quarter revenue guidance is $108 billion. That number assumes zero China data center compute revenue — and China was in fact under 1% of data center revenue this quarter. The plan, in other words, is to clear $100 billion in a quarter on the rest of the world alone, leaving the US export-control hole exactly where it is.

The company also offered a preliminary view that fiscal 2028 revenue could grow about 70%. The reasoning is that demand sits well above what NVIDIA can supply. Management added that cloud customers have already placed more than $2 trillion of orders in backlog.

One way to frame it: training investment is project-based, so it moves in cycles. Inference — the compute that runs every time users around the world make a live request — has to keep adding capacity as traffic grows. The more revenue weight shifts from training toward inference, the more NVIDIA’s revenue behaves like a recurring utility bill than a one-time equipment sale.

Vera Rubin’s ramp and a $500 billion infrastructure financing group

Vera Rubin, the next-generation platform, began production shipments in early August and is expected to account for roughly 20% of third-quarter data center revenue. Inventory rose to $31.6 billion from $25.8 billion the prior quarter to prepare for the ramp. The generational handoff from Blackwell to Rubin is happening without a revenue gap.

The more notable structural move is how the buildout gets financed. NVIDIA announced a group — with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to bring more than $500 billion of outside capital into AI infrastructure construction. It is still a concept, contingent on definitive agreements.

The logic is straightforward. Data center construction costs move off big-tech balance sheets and onto private-equity and alternative-asset managers, while NVIDIA supplies the reference design and the chips. Customers keep building without carrying the leverage themselves. And it thickens the moat, since more capital ends up locked into the NVIDIA ecosystem.

Memory costs and China: the variables to check now

A strong quarter does not clear the risks.

First, margins come down. Third-quarter gross margin is guided to 74.0%, and the fourth quarter is expected to fall to 71-72% on higher memory costs. NVIDIA has said it will not absorb that cost but pass it to customers. How far that pass-through holds is the thing to watch.

Second, power. A gigawatt-class AI factory draws as much electricity as a small city. If transmission buildout and generation permitting lag, the equipment gets bought and cannot be switched on. NVIDIA securing power capacity directly at its Ohio campus is a response to exactly this bottleneck.

Third, end demand. If the companies pouring money into infrastructure cannot show matching software profit, the next investment cycle can slow. The $2 trillion backlog still has to pass that test before it turns into revenue.

What to file away from this quarter

Three lines for the file:

Q2 revenue $96.2 billion, data center $89 billion, gross margin 75.0%. AI compute is hardening into recurring revenue.
Q3 guidance $108 billion (China modeled at zero), a preliminary view of ~70% revenue growth next fiscal year, and a cloud order backlog above $2 trillion.
The risk is not the earnings, it is the margin path: 71-72% in Q4, memory-cost pass-through, and power bottlenecks.

Checkpoints for next quarter:

Whether Q4 margin holds the 71-72% band or breaks lower
Whether Vera Rubin actually fills its 20% share of Q3 data center revenue
Whether the $500 billion financing group moves from concept to definitive agreements

Sources checked

Disclaimer — This article is for general informational purposes only and is not a recommendation to invest in any specific security or product. Investment decisions and their consequences are solely the reader's responsibility.

Add TrueMyPick as a preferred source on Google.