Nvidia Reports Q2 Tonight: $92B Revenue Would Double Last Year, $85B From Data Center Alone
Nvidia reports fiscal Q2 2027 earnings after the bell on August 26. Wall Street consensus has settled at $91.8-$92.2 billion in quarterly revenue, representing approximately 96-97% year-over-year growth. A year ago Nvidia’s entire business generated about $47 billion in the same period. The data center segment — driven by Blackwell GPU shipments to hyperscalers and cloud providers — is expected to account for roughly $85 billion of that total, itself a figure that exceeds last year’s full-quarter consolidated revenue.
Consensus EPS is approximately $2.09.
What the Numbers Mean
A beat on data center revenue extends what is already the most concentrated earnings run in semiconductor history. A miss, or a cautious guidance print, tests whether the infrastructure buildout is as durable as the capital commitments suggest.
The complicating variable is China. Nvidia’s official guidance excludes revenue from Chinese data center operations, given ongoing export restrictions on H100 and H200 class hardware. If Nvidia has found compliant pathways to ship meaningful volumes into Chinese hyperscalers under the current rules, that revenue could surface as upside not embedded in consensus. The guidance exclusion makes the China line an opaque wildcard in either direction.
Infrastructure Commitments Extend Well Past This Quarter
Beyond the quarter itself, the forward signals are unusually concrete. Nvidia announced in August 2026 a partnership with BlackRock and Goldman Sachs aimed at mobilising more than $500 billion for AI infrastructure projects — an arrangement that effectively turns Wall Street debt and equity capacity into Nvidia’s downstream demand funnel.
The company’s own projections for the Blackwell architecture and the upcoming Rubin platform put cumulative revenue from both at $1 trillion between 2025 and the end of 2027. That number requires sustained data center demand at current or increasing rates through the next six quarters.
The Thesis Being Tested
The AI infrastructure supercycle argument holds that hyperscaler and sovereign AI capital expenditure is in an early, durable phase — that the buildout is not speculative excess but foundational spend required to make frontier model inference economically viable at scale. Nvidia’s sequential quarters have been the primary evidence cited for that thesis.
Fiscal Q1 2027 set the prior benchmark. Q2 will show whether Blackwell ramp has continued at the pace needed to sustain the $1 trillion forward projection, or whether hyperscaler order patterns have shifted.
For model inference economics, a sustained Nvidia revenue trajectory matters in a specific way: it is the primary input into GPU supply projections, which in turn determine when inference costs fall enough to enable the next tier of agentic deployments. $85 billion in data center revenue per quarter means approximately $340 billion annually in new AI compute being deployed. At current Blackwell pricing, that represents roughly two to three million new H100-equivalent GPU-months of capacity per quarter entering the market.
Key Figures at a Glance
| Metric | Consensus Estimate | YoY Change |
|---|---|---|
| Total Revenue | $91.8-$92.2B | ~+97% |
| Data Center Revenue | ~$85B | >+100% |
| EPS | ~$2.09 | — |
| BlackRock/Goldman Partnership | $500B targeted | Announced Aug 2026 |
| Blackwell + Rubin cumulative projection | $1T by end-2027 | — |
Results are expected after 4:00 PM Eastern on August 26.