Nvidia Doubles Down: Q2 FY27 Revenue Hits $96.2B, Guides $108B as AI Buildout Goes 'Full Steam'

Nvidia just delivered the most closely watched earnings report of the AI era, and it was a blowout by almost every measure that matters. On Wednesday, August 26, the company reported revenue of $96.2 billion for its fiscal 2027 second quarter — up 106% from a year ago and up 18% sequentially — while guiding next quarter’s revenue to $108 billion, a figure that would have seemed absurd for a full year not long ago.

The numbers land amid genuine anxiety about whether the AI infrastructure boom is sustainable. Nvidia’s answer, delivered in headline form, is that demand is not just holding but accelerating. “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue,” said Jensen Huang, Nvidia’s founder and CEO. “And demand is accelerating. This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem and physical AI coming online.”

The numbers

The Q2 FY27 scorecard, for the quarter ended July 26, 2026:

• Revenue: $96.2 billion, up 106% year over year and 18% quarter over quarter, beating consensus estimates of roughly $92 billion
• Data Center revenue: $89.0 billion, up 117% year over year, ahead of the ~$85.7 billion analysts expected
• GAAP EPS: $2.46 (up 128% YoY); non-GAAP EPS of $2.22 more than doubled from $1.01 a year ago and beat the ~$2.08 consensus
• Gross margin: 75.0% (GAAP and non-GAAP), up 2.6 points from a year ago
• Operating income: $63.7 billion GAAP, up 124% YoY
• Q3 FY27 guidance: $108 billion ± 2%, implying ~89% annual growth, with 74.0% gross margins expected

Notably, the outlook assumes zero data center compute revenue from China — a conservative stance given ongoing export controls.

The quarter’s growth was driven by the ramp of Blackwell Ultra infrastructure. Within the data center segment, hyperscaler revenue reached $48.7 billion (up 101.5% YoY), while AI Clouds, Industrial and Enterprise revenue hit $40.3 billion (up 138.1% YoY), reflecting demand from AI-native companies, enterprises, sovereign customers, and hyperscalers using AI clouds. Edge Computing added $7.2 billion, up 27% YoY, powered by Blackwell workstations even as consumer PC sales softened on higher memory and system prices.

Nvidia also returned roughly $26 billion to shareholders during the quarter through buybacks and dividends, with about $99 billion remaining under its repurchase authorization.

What the call revealed

The earnings call itself produced the most forward-looking datapoints of the night. CFO Colette Kress told analysts that Nvidia’s backlog now exceeds $2 trillion and that growth has now accelerated for four consecutive quarters. She offered a revenue growth forecast for fiscal 2028 of roughly 70% — and pointedly noted that the figure is “supply-constrained,” meaning demand is running ahead of what Nvidia can physically ship.

Kress also forecast hyperscaler capital expenditure of more than $800 billion this year and $1.3 trillion in 2027, and announced an expanded partnership with Amazon’s AWS. The market responded in real time: Nvidia shares turned from slightly lower to up nearly 5% in after-hours trading as the call progressed.

Huang, asked about the state of AI adoption, was characteristically expansive: “About half of our business is growing about 100% a year, and that’s beyond the cloud.” He described AI infrastructure as a jobs creator across the supply chain and teased the road ahead: “We’ve got a huge year coming up next year. It’s going to be extraordinary.”

Vera Rubin enters full production

The product story beneath the financials is the Vera Rubin platform, which Nvidia says is now ramping into full production, with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. Spectrum-6 switch systems supporting pluggable and co-packaged optics are arriving at what the company calls “gigascale AI factories,” and Vera — which Nvidia bills as the first CPU built for AI agents — is seeing broad adoption planned across leading technology providers. Huang tied the platform directly to the moment: “Vera Rubin, now in full production, was built to power exactly this moment.”

The quarter also brought a wave of adjacent announcements: the Groq 3 LPX interactive inference accelerator reaching full production, the DSX platform for designing and operating AI factories at scale, the Isaac GR00T reference humanoid robot, Cosmos 3 (billed as the first fully open frontier omnimodel for physical AI), and a multiyear memory partnership with SK hynix.

The $500 billion financing question

Perhaps the most debated announcement is Nvidia’s plan to establish independent compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — partnerships intended to mobilize more than $500 billion of third-party capital for AI infrastructure buildout over time, subject to definitive agreements.

Critics have framed such arrangements as “circular financing”: chipmakers investing in customers who buy their chips, with financial institutions intermediating. Kress confronted the criticism head-on during the call, emphasizing that the financing structures are independent and that the underlying demand — “skyrocketing” usage, in her words — is what justifies the capital flow. The $2 trillion backlog figure is her strongest evidence: that is contracted demand, not speculative financing.

Sovereign AI is the other quiet giant in the numbers. Nvidia highlighted a national AI infrastructure launch with the Japanese government, gigawatt-scale sovereign AI infrastructure in Korea with SK Telecom, NAVER and Brookfield, and newly secured land, power and shell capacity at the PORTS-Pike Technology Campus in Ohio with SB Energy. Geographic diversification matters because it reduces dependence on any single hyperscaler cohort — and because U.S. private AI investment alone reached $285.9 billion in 2025 per the Stanford AI Index.

What it means

Three takeaways for anyone tracking the AI economy:

1. Compute is now a measured, monetized commodity. Huang’s “compute is revenue” line is more than rhetoric — it reflects a shift where token generation has become a metered business with visible unit economics. When tokens are “productive and profitable,” AI inference infrastructure behaves less like speculative capex and more like a utility buildout.

2. The constraint has flipped from demand to supply. A $2 trillion backlog against $96 billion of quarterly revenue means roughly five years of demand visibility at current run rates. Kress labeling FY28 guidance as supply-constrained tells you where the bottleneck sits: memory, advanced packaging, power, and racks — not customers.

3. Watch gross margins and the financing stack. Q3 guidance of 74.0% gross margin is a slight step down from Q2’s 75.0%, and the sheer scale of third-party financing now propping up the buildout is the legitimate bear case. If token demand plateaus while financed capacity keeps arriving, the industry would face a painful digestion period. For now, the numbers say the opposite: usage, backlog, and capex forecasts are all still pointed up.

Nvidia reports again in late November. Until then, the $108 billion Q3 guide — and whether Vera Rubin shipments can satisfy a $2 trillion queue — is the number the entire AI supply chain will be watching.