Nvidia has told Chinese customers that its new Vera central processing unit, an 88-core Arm-based chip designed for AI data centres, could be available as early as August and that orders can begin now.
The outreach represents a deliberate pivot. US export controls have blocked Nvidia's Blackwell GPUs from China entirely, restricted Hopper shipments and reduced the company's Chinese AI revenue to near zero. Jensen Huang said earlier this year that Nvidia's China GPU share had dropped to nothing.
Vera is not a GPU. It is a CPU. And CPUs are not currently subject to the same export restrictions.
The product
Vera is an Arm-based processor built for agentic AI workloads, the inference and orchestration tasks that run AI agents in production rather than the training workloads that require GPUs. It was unveiled in March 2026 and is now in full production.
The chip is part of the broader Vera Rubin platform, which pairs the CPU with Nvidia's Rubin GPUs. The China pitch focuses on the CPU in isolation, precisely because the GPU component would attract regulatory scrutiny.
Pricing is significant. A single Vera processor costs well above $20,000 before bulk discounts. A fully configured rack of 256 chips runs approximately $10m. Nvidia has projected Vera as a nearly $20bn revenue line by fiscal 2027.
Alibaba and ByteDance are among the hyperscalers being approached. One major Chinese cloud provider is reportedly planning to order more than 300 servers, each containing two Vera CPUs, for initial testing before committing to larger deployments.
Regulatory question
The Bureau of Industry and Security has not classified CPUs under the same export restrictions that apply to advanced GPUs. That classification could change. The precedent is not encouraging for Nvidia: Washington cleared H200 GPU sales to 10 Chinese firms earlier this year, but shipments stalled for months as regulatory reviews dragged on.
If Vera gains commercial traction in Chinese data centres, the BIS faces a decision. A CPU optimised for AI inference, sold at scale to Chinese hyperscalers, sits in a grey zone that the current rules were not designed to address. Leaving it unrestricted allows Nvidia to rebuild Chinese revenue through an alternative product category. Restricting it closes another door and pushes Chinese customers further towards domestic alternatives.
Competitive context
Nvidia is not entering an empty market. Intel and AMD both sell server CPUs, and both have warned that global CPU supply is tight, with 2026 capacity largely sold out and price increases of 10% to 15% planned. Alibaba, Tencent and Baidu have all launched self-developed AI chips, reducing their incentive for large-scale procurement of foreign processors.
Chinese interest in Vera is real but cautious. Initial deployments are reportedly planned for overseas data centres rather than domestic facilities, a detail that may reflect both regulatory caution and the practical reality that Chinese cloud providers are accelerating domestic substitution.
Nvidia lost its GPU business in China. Vera is the attempt to replace it with something the export controls do not yet cover. Whether Washington lets that stand depends on how successful the pitch turns out to be.
