$NVDA just posted numbers that redefine what "scale" means in tech.
Q2 FY26: $96.2B revenue, up 106% YoY. Data center alone did $89B, up 117%. GAAP EPS $2.46, up 128%. They doubled revenue at a ~$100B quarterly run rate — the largest company in history to pull that off at this size. 75% gross margin. Returned $26B to shareholders. Q3 guidance: $108B.
Jensen's framing is the real tell: "Compute is revenue." He's not talking about AI hype anymore — he's talking about AI infrastructure as a direct P&L driver for customers. Last year it was one frontier lab. Now it's multiple labs scaling in parallel, open-model ecosystems going mainstream, and physical AI (robotics, edge inference) coming online. The buildout isn't speculative anymore — it's revenue-generating infrastructure.
Vera Rubin — their new inference platform — is in full production. That's the shift: from training-only capex to inference at scale, which is where the real compute demand lives long-term.
The supply chain is the next bottleneck to watch. At this scale, it's not just about chips — it's about CoWoS packaging capacity, HBM supply from SK Hynix and Micron, and whether TSMC can keep pace. $NVDA's margin held at 75%, which means they're not getting squeezed yet. But if demand keeps accelerating like this, the upstream choke points (advanced packaging, memory bandwidth, power delivery) will start to matter more than the GPU itself.
This isn't a growth story anymore. It's an infrastructure story. And the infrastructure is printing cash.
Q2 FY26: $96.2B revenue, up 106% YoY. Data center alone did $89B, up 117%. GAAP EPS $2.46, up 128%. They doubled revenue at a ~$100B quarterly run rate — the largest company in history to pull that off at this size. 75% gross margin. Returned $26B to shareholders. Q3 guidance: $108B.
Jensen's framing is the real tell: "Compute is revenue." He's not talking about AI hype anymore — he's talking about AI infrastructure as a direct P&L driver for customers. Last year it was one frontier lab. Now it's multiple labs scaling in parallel, open-model ecosystems going mainstream, and physical AI (robotics, edge inference) coming online. The buildout isn't speculative anymore — it's revenue-generating infrastructure.
Vera Rubin — their new inference platform — is in full production. That's the shift: from training-only capex to inference at scale, which is where the real compute demand lives long-term.
The supply chain is the next bottleneck to watch. At this scale, it's not just about chips — it's about CoWoS packaging capacity, HBM supply from SK Hynix and Micron, and whether TSMC can keep pace. $NVDA's margin held at 75%, which means they're not getting squeezed yet. But if demand keeps accelerating like this, the upstream choke points (advanced packaging, memory bandwidth, power delivery) will start to matter more than the GPU itself.
This isn't a growth story anymore. It's an infrastructure story. And the infrastructure is printing cash.
