Three people familiar with the matter say Nvidia has started pitching its new Vera CPU — designed specifically for agentic AI workloads — to Chinese customers, with shipments potentially starting as soon as August. The company has already begun taking orders. The accelerated push is an attempt to regain a foothold in China after tightened US export controls blocked shipments of its H200 chip, effectively driving Nvidia's market share in the country down to near zero.
Some Chinese customers have expressed interest in the standalone CPU, according to the sources. One major Chinese cloud provider is planning to order more than 300 servers, each equipped with two Vera processors. But the customer intends to first test the chips in overseas data centers before deciding on follow-up orders. Whether Vera sees widespread adoption remains uncertain, given software ecosystem compatibility issues and the broader push to shift workloads to domestic AI chips.
According to estimates from semiconductor research firm Semi, a single Vera processor will cost well over $20,000 before volume discounts, and a full rack with 256 chips runs roughly $10 million. The global AI market's shift from training to inference is exacerbating CPU shortages. Nvidia has projected $20 billion in Vera sales for the fiscal year ending January 2026. The push also intensifies competition with Intel and AMD, both of which are ramping up server CPU production for AI data centers.