Overview
- Nvidia began pitching the Vera data‑center CPU to Chinese clients on Friday and is accepting orders with possible availability as soon as August.
- Vera is an 88‑core Arm‑based processor built for agentic AI workloads and is sold alongside Nvidia’s Rubin platform as a rack and server product.
- One major Chinese cloud buyer plans an initial test order of more than 300 dual‑CPU servers, but wider adoption will depend on benchmarks, software compatibility and integration work.
- Analysts estimate a single Vera chip will cost well over $20,000 and a fully configured 256‑chip rack about $10 million, while Nvidia projects roughly $20 billion in Vera revenue this fiscal year.
- The push gives Nvidia a route back into China where its GPU business stalled under U.S. export controls, but competition from Intel, AMD and custom chips, supply constraints and potential changes in export policy remain key risks.