Nvidia's Vera CPU Becomes Its Side Door Into China

For most of this year, Nvidia has watched its China business shrink to almost nothing. American export controls have kept its advanced AI accelerators out of the country, and chief executive Jensen Huang has said the company's market share there has effectively fallen to zero. On 12 June, Nvidia showed it has found another way through the door.

The company has told Chinese clients they can begin placing orders for its Vera server processor now, with shipments expected as soon as August, according to Tom's Hardware. The pitch matters because Vera is not a graphics chip. It is a central processor, and that distinction is the whole point.

Why a CPU and not a GPU

Washington's restrictions were written for the chips that train and run large AI models, the high-bandwidth GPUs that Nvidia is famous for. Regulators have not placed advanced CPUs in the same category. Nvidia is reading that gap closely. Vera is an Arm-based, 88-core processor that the company unveiled in March and has since moved into full production. It was built for what Nvidia calls agentic AI workloads, the orchestration and data-handling side of modern AI systems rather than the raw model training that GPUs handle.

The numbers are not small. The Next Web reports that a single Vera processor will cost well above $20,000 before bulk discounts, and that at least one major Chinese cloud provider is lining up an initial order for more than 300 dual-Vera servers. That is a meaningful first sale in a market Nvidia had all but written off.

A workaround with a short shelf life

The risk is hard to miss, and Nvidia surely knows it. Selling into China through a regulatory gap is the kind of move that invites Washington to close that gap. Reuters reporting, carried by Yahoo Finance, frames the strategy as a calculated bet that CPUs will stay outside the export-control net long enough to count. If American officials decide that powerful server processors feeding Chinese AI data centres deserve the same scrutiny as GPUs, the side door shuts.

For now, the timing fits a larger picture. China is pouring state and private money into domestic AI infrastructure, the same build-out behind stories like Moonshot AI's pursuit of a $30 billion valuation, and those data centres still need processors. If Nvidia cannot sell the GPUs, selling the chips that sit beside them is the next best thing. It keeps the company present in a market it spent years building, and it keeps Chinese buyers at least partly tied to Nvidia's roadmap.

The real question is whether this is a lasting channel or a brief opening. Export policy has moved fast before. What looks like a clever workaround in June can become the next thing regulators target by autumn. Nvidia is betting the August deliveries land before that happens.