Key Points
- Nvidia opened Vera CPU orders to Chinese customers around June 12.
- Shipments could begin as early as August.
- US export controls target high-end GPUs, not general-purpose CPUs.
- Jensen Huang said Nvidia's China GPU share fell to near zero.
Nvidia has found a side door back into China. The company has told Chinese customers they can place orders now for its Arm-based Vera CPUs, with deliveries possible as early as August, according to Reuters reporting. It is the clearest sign yet of how Nvidia plans to rebuild a market that US export controls all but erased.
The workaround
Washington's restrictions are aimed at high-end GPUs — the chips that train frontier AI models — not general-purpose CPUs. Vera, the Arm-based server processor Nvidia unveiled in March, falls on the permitted side of that line. Selling a CPU into China is less fraught than selling an H200, and Nvidia is moving to exploit the gap: one major Chinese cloud provider is reportedly preparing to order more than 300 servers, each carrying two Vera chips.
How far the GPU business fell
The context is a near-total collapse. Jensen Huang said last year that Nvidia's share of China's GPU market had fallen to roughly zero under export controls. Even the partial thaw has stalled: the US licensed about ten Chinese firms to buy the H200, but the same export-control machinery now reaching into AI models cuts both ways — Beijing has withheld approval to favor domestic chipmakers, and not one of those H200s has shipped. The CPU is the one product Nvidia can actually move.
The read
Nvidia projects roughly $20 billion in Vera revenue by the end of January, with TSMC building the chips on its 3nm node. China is too large to abandon, but this is a side door, not the front gate: a general-purpose CPU is a worse fit for cutting-edge AI training than the GPUs that are blocked, and Beijing's own push toward domestic silicon means even the open lane could narrow. For now it lets Nvidia keep a foot in the market — and keep Alibaba and ByteDance, early Vera collaborators, inside its ecosystem rather than a rival's.
Source: The Next Web, Tom's Hardware
Comments