Nvidia pitches Vera chip to China as it tries to regain ground

Nvidia has begun selling Chinese customers on a new way back into a market that U.S. export controls have all but frozen. The company’s Vera central processor, built for agentic AI systems, is being pitched as available as soon as August, with orders already open, in a sign that Nvidia is trying to rebuild momentum rather than abandon China.

Vera is Nvidia’s first standalone CPU for workloads that include code execution, tool use, sandboxing, analytics, data pipelines and orchestration. Nvidia launched the chip on March 16 and said it was the world’s first processor purpose-built for agentic AI and reinforcement learning. By May 31, the company said Vera was in full production, that systems using it would arrive in the second half of 2026, and that the chip delivered 1.8 times faster task completion than x86 CPUs.

The Chinese pitch lands after months of stalled H200 shipments to China, a dead end that has left Nvidia unable to turn approved sales into deliveries. Jensen Huang has said Nvidia’s share in China fell from about 95% to zero, underscoring how sharply U.S. restrictions have reversed the company’s position in one of its most important markets. Vera looks like an attempt to bypass that blockage by selling a different layer of the AI stack, the CPUs and rack systems that sit behind the better-known accelerators.

The economics are steep. A single Vera processor is said to cost well north of $20,000 before bulk discounts, while a fully configured rack could approach $10 million. Nvidia’s own platform materials say a Vera rack can hold 256 Vera CPUs in dense, liquid-cooled infrastructure. That price tag makes the product a serious commitment, not a casual replacement, especially for Chinese buyers already balancing domestic alternatives and tighter scrutiny around foreign technology.

Interest has emerged, but not yet conviction. One Chinese cloud company is said to be considering more than 300 servers, with two Vera CPUs in each, though early deployments could begin in overseas data centers for testing before any larger rollout in mainland China. Software compatibility remains a major obstacle, along with the challenge of moving workloads built around domestic AI chips. That matters at a moment when Beijing is pressing companies to favor local suppliers such as Huawei.

Nvidia’s gamble is clear: export controls may have blocked one route, but they have not ended the China market. Vera is the company’s test of whether Chinese customers see a U.S.-approved workaround as a meaningful option or a stopgap while homegrown rivals keep gaining ground.