Nvidia is actively pitching its first standalone data center CPU, Vera , to Chinese customers. Not subject to US export restrictions, the chip – built on an 88‑core custom Arm architecture – will be available as early as August 2026 , with orders already open.

Vera targets agentic AI and reinforcement learning backend tasks and is already in mass production. Nvidia claims it runs 1.8x faster than competitors , and Jensen Huang expects it to become a multi‑billion‑dollar business. China’s Alibaba and ByteDance are among the firms in early discussions, initially planning overseas data center tests.

Vera intensifies competition with Intel and AMD in server CPUs, as AI shifts from training to inference. Intel recently warned of 6‑month delivery lead times for server CPUs in China; AMD also flagged tight global supply. SemiAnalysis estimates Vera’s undiscounted per‑chip price exceeds $20,000** , with a full 256‑chip rack costing around **$10 million . Nvidia expects Vera to contribute ~$20 billion in revenue in the current fiscal year (ending Jan 2027).

Key adoption hurdles in China: software ecosystem compatibility and migration costs from domestic AI chips.

ICgoodFind: Nvidia’s Vera CPU enters China’s AI inference market, bypassing export curbs – a three‑way server CPU battle with Intel and AMD looms.