1/
Jensen confirmed it. Vera CPUs are now at Anthropic, OpenAI, SpaceX, and Oracle. 🤖
This is Nvidia's second data center CPU. First was Grace (2021). Vera is purpose-built for one thing: agentic AI.
2/
Grace = general HPC. Vera = orchestration layer for AI agents.
When Claude or GPT-4o runs a multi-step autonomous task, the GPU handles inference. The CPU handles the decision logic, tool calls, memory retrieval, loop control. That's Vera's job. 💻
3/
Why those four customers first?
Anthropic + OpenAI: production agentic systems already running at scale.
SpaceX: Starlink's autonomous ground network needs exactly this kind of CPU.
Oracle: cloud infrastructure partner, early deployment makes sense.
4/
Jensen also said China will "gradually open." No timeline.
But Vera going to the top four AI labs while China sits out the H200 trade tells you where Nvidia's near-term growth is coming from. It's not China. It's every other AI lab on the planet. 🇺🇸
Watch Nvidia's Q3 earnings. Vera attach rate in data center revenue is the new number to track.