Nvidia has told Chinese clients its Arm-based Vera server CPUs could ship as early as August, opening a path into a market where its AI GPUs remain stuck. Server CPUs face lighter export controls than accelerators, and the timing places Chinese buyers near the front of an allocation queue for a product Nvidia expects to clear $20 billion in revenue by late January.
Nvidia has informed Chinese clients that its Arm-based Vera server CPUs could be available as soon as August, with orders open now, according to a Reuters report citing three sources familiar with the matter. The move would route Nvidia silicon into a market where its data center GPUs remain frozen, months after CEO Jensen Huang said the company's share in China had effectively dropped to zero.
The distinction between the two product lines explains why one can move and the other cannot. Server CPUs sit under far lighter U.S. export restrictions than the accelerators that anchor Nvidia's data center business. The H200, the GPU at the center of the standoff, illustrates how tight that gate has become. Washington licensed roughly 10 Chinese firms to buy the part, yet not a single unit has been delivered. Chinese officials, working to build up domestic chipmakers, have withheld approval on their own side. Beijing, not just Washington, now functions as a gatekeeper.
That second obstacle shapes how Vera will be deployed. The Reuters sources say Chinese cloud companies are already testing more than 300 Vera servers, and at least one major cloud provider intends to place an order. Initial deployments, however, will be confined to those companies' overseas data centers. Putting U.S. silicon into domestic Chinese facilities would invite scrutiny from Beijing, so the hardware lands where it draws the least political attention.
What Vera actually is
Vera started as the CPU half of the Vera Rubin superchip, first shown at last year's GTC. Nvidia split it out as a standalone product at GTC San Jose this March and paired it with a rack design that holds 256 liquid-cooled Vera CPUs and sustains more than 22,500 concurrent CPU environments. By Computex, the company said the chip had reached full production and claimed 1.8 times faster task completion than x86 processors on agentic workloads.
The Arm-based design follows the path set by its predecessor, Grace, which has shipped close to 2.5 million units to date. That installed base matters. It gives Nvidia a credible software and tooling story for customers weighing a switch away from x86, and it gives the company manufacturing experience with the architecture before Vera scales. The 1.8x figure is Nvidia's own benchmark and should be read as a vendor claim on a specific class of workloads rather than a general performance verdict, but the targeting is deliberate. Agentic AI is exactly where host processor demand is climbing fastest.
The agentic squeeze on host processors
The broader pull behind Vera comes from a shift in how AI runs. Workloads are moving from training toward inference and agentic execution, and agentic systems lean heavily on the host CPU for tool calls, code execution, and data handling. Each step an agent takes, calling an API, running a snippet, parsing a result, routes back through the processor rather than staying on the GPU. As agent counts climb, CPU demand has outrun supply.
The supply picture across the market backs this up. Intel has quoted Chinese customers lead times of up to six months. AMD has described the global CPU market as tight, with demand outpacing its forecasts and supply constraints expected to persist. Telling Chinese buyers they can hold Vera silicon in August, in the middle of that shortage, signals that those customers sit near the front of the allocation queue.
The revenue stakes make the prioritization understandable. Nvidia expects the Vera line to generate $20 billion by the end of its fiscal year in late January. The August timeline also lines up with what the company said at GTC Taipei during Computex, where it indicated Vera systems would reach customers through system builders and cloud partners starting this fall.
For Nvidia, the calculation is straightforward. GPUs remain its highest-margin and most contested product in China, blocked from both directions. CPUs offer a way to keep a commercial relationship alive, hold mindshare with Chinese cloud operators, and book revenue against a product line in genuine shortage. Whether the overseas-only restriction holds, and whether Beijing tolerates even that arrangement, will determine how much of the China opportunity Nvidia can actually capture while its accelerators stay on the wrong side of two governments.
More detail on the report is available via Reuters, and Nvidia's product disclosures for the Vera Rubin platform are posted on the Nvidia newsroom.
Comments
Please log in or register to join the discussion