Last October, Nvidia’s CEO Jensen Huang admitted that its market share in China had fallen to effectively zero. However, it seems Nvidia is not giving up on the Chinese market, as it is now reportedly taking orders from Chinese customers for its new Vera processor.
People familiar with the discussions told Reuters that the chip could ship as early as August.
Despite its efforts, Nvidia has failed to maintain a foothold in China due to U.S. export controls, which have blocked shipments of its most powerful AI accelerators to China.
Beijing, in response, has steered domestic buyers toward homegrown alternatives such as Huawei and others.
Per reports, no Chinese customer has received any H200 GPU despite 10 firms receiving U.S. licenses, and this is because Chinese authorities have withheld their own approval.
Why is Nvidia pushing the Vera chips to China?
With Vera, Nvidia is entering the server CPU market, a segment that currently faces fewer export restrictions than high-end AI accelerators.
The chip is Nvidia’s first standalone central processor, designed for the computing tasks that AI agents depend on, such as database queries and code compilation.
According to Nvidia, Vera chips, which are built on Arm architecture, are 1.8 times faster than comparable x86 processors on workloads tied to autonomous AI systems. Intel’s Xeon and AMD’s Epyc lines have dominated server processors for decades using x86 designs.
“AI agents will be the largest users of computing,” Huang said at an event, adding that “Vera is the first CPU designed for that future.”
Are Chinese buyers showing interest in Nvidia’s chip?
At least one major Chinese cloud provider plans to order more than 300 servers, each with two Vera processors, for initial testing, according to one of the people close to the matter.
However, it is currently unclear if those trials will convert to large-scale purchases. Software compatibility issues and the difficulty of migrating workloads already built around domestic chips could slow adoption, a second source said.
There’s also the pricing hurdle, as a single Vera processor costs north of $20,000 before volume discounts, and a full 256-chip rack runs to approximately $10 million depending on memory configuration, according to estimates from SemiAnalysis.
One of the sources also added that Chinese buyers are planning on deploying the chips in data centers outside of China, as domestic usage is still considered sensitive politically.
How important are the Vera chips in the AI industry?
The global AI industry is shifting from training large models to running them at scale, a phase called inference.
That transition favors CPUs and custom silicon alongside GPUs, and it has resulted in the shortage of CPUs. Intel flagged this development in February when it told Chinese customers to expect six-month lead times on server processors.
AMD said last month that the global CPU market is “tight,” with demand exceeding current supply and forecasts.
Nvidia is expecting $20 billion in Vera revenue by the end of its current fiscal year, which closes in January 2027. Cloud providers Alibaba and ByteDance are reportedly among the companies collaborating on Vera deployments, although it has not been confirmed if they placed orders.
Nvidia’s Q1 fiscal 2027 results, reported in May, showed $81.6 billion in total revenue, an 85% year-over-year increase, with the data center segment accounting for $75.2 billion of that total.
What is the geopolitical risk hovering over Vera?
If the U.S. determines the CPU processors contribute to advanced AI development in China, new restrictions could follow.
Beijing’s push for semiconductor self-sufficiency adds another layer to the risk. Chinese customers may treat Nvidia hardware as a temporary fix until domestic alternatives catch up, which is bad news for Nvidia.
Nvidia has also moved to shore up its political position in Washington. The company recently hired Bruce Andrews, a former Commerce Department official under Obama and Intel’s former head of government relations, to lead its government affairs operation.