Nvidia's Vera CPU is its alternative route back into China.

      Nvidia has begun promoting its Vera CPU to clients in China. Orders can be placed immediately, with deliveries potentially starting as early as August, according to a Reuters report on Friday, which referenced three individuals familiar with the discussions.

      This move appears to be a workaround disguised as a product launch. Nvidia's business in China has been significantly impacted by US export regulations. In October, Jensen Huang stated that the company’s market share in that region had declined to nearly zero. The introduction of the Vera CPU could serve as a pathway for re-entry.

      The reason the Nvidia Vera CPU might evade restrictions stems from its chip classification. The US export controls predominantly focus on high-end GPUs rather than general-purpose CPUs, making the sale of a processor less complicated.

      However, it is not entirely straightforward. It remains uncertain whether Nvidia will require an export license for the advanced CPU like Vera. The differences with GPUs are notable: the US has permitted around ten Chinese companies to acquire the H200, but none have been delivered due to Beijing's reluctance to approve imports in favor of domestic chips.

      Huang has suggested some possibilities, responding last month in Taipei to whether Nvidia’s CPU forecasts included China by saying, “I would think so.”

      Vera represents Nvidia’s inaugural standalone data-center CPU, which is an Arm-based chip designed for AI applications and is currently in full production. Nvidia claims it operates up to 1.8 times faster than competing processors, with Anthropic and OpenAI being among its initial users.

      The price point is significant, as a single Vera chip costs "well north" of $20,000 before discounts, and a rack featuring 256 chips is priced at around $10 million, as per SemiAnalysis. Nvidia anticipates generating $20 billion in revenue from Vera by the end of January, considering the Chinese market too large to disregard.

      The timing is also favorable. The focus of the AI competition is shifting from training models to executing them, and this inference work relies more heavily on CPUs. Demand is so high that Intel has informed Chinese buyers of six-month waiting periods.

      Vera also poses a challenge to Intel and AMD, as it directly competes in the CPU market traditionally dominated by x86 chips. Selling in China, where there is a shortage of server CPUs, expands the competition. The leader in GPUs is now targeting the processor market as well.

      Although there is interest, actual adoption is uncertain. A significant Chinese cloud provider is looking to order over 300 servers, each equipped with two Vera chips, for initial testing. Currently, Chinese companies intend to operate Vera solely in their overseas data centers.

      Large-scale implementations are not guaranteed and will depend on software compatibility, as well as the costs associated with transitioning from domestic chips. The Chinese government is also strongly advocating for self-reliance. Interestingly, Alibaba and ByteDance are mentioned as collaborators on Vera, even as China develops its own alternatives.

      Thus, this approach serves as a side entry rather than a direct access point. Nvidia has established a legal pathway to resume selling in China, while its top GPUs remain restricted, and a gray market has emerged, with B300 servers priced around $1 million within China. The actual commitment from Chinese buyers, who might prefer to manufacture their own solutions, will ultimately determine the success of this initiative before the August deadline.