June 12 (Reuters) – Nvidia has told Chinese clients that its new “Vera” central processors for AI data centres could be available as soon as August and that they can begin placing orders, three sources familiar with the matter said.
The outreach underscores how the world’s most valuable company is quickly pivoting to the new product to revive its rapidly declining fortunes in China, as shipments of its second-most powerful AI chip, the H200, to the country have stalled for months.
Nvidia’s market share in China has effectively fallen to zero, its CEO Jensen Huang said in October, hurt by U.S. export controls on advanced chips and Beijing’s push for self-reliance in key technologies.
The move also raises the stakes in its competition with major CPU firms Intel and AMD, which are racing to increase supplies of server CPUs for AI data centres.
The sources said some Chinese clients have shown interest in the Vera chip, Nvidia’s first standalone central processing unit (CPU) built for agentic AI — systems that perform tasks autonomously. The sources declined to be identified because the discussions are private.
Now in full production, the Vera chip is built for the behind-the-scenes computing that AI agents rely on, and Nvidia says it runs up to 1.8 times faster than comparable processors from its rivals.
Unveiling Vera in March, Huang expected it to become the company’s next multibillion-dollar business. Nvidia also said at the time that leading cloud firms including Alibaba and ByteDance were collaborating with it to deploy Vera, but it did not say whether the actual ordering process for the chip had begun.
Nvidia declined to comment. Alibaba and ByteDance did not respond to requests for comments.
POSITIVE FEEDBACK
One major Chinese cloud company plans to place an order for more than 300 servers, each containing two Vera CPUs, one of the sources said. The company plans to deploy the systems for testing first and decide whether to place official orders based on the results, the person added.
Whether that initial interest translates into large-scale adoption remains to be seen, partly due to issues involving software ecosystems and compatibility as well as the constraints of migrating workloads built around domestic AI chips, another source said.
A single Vera processor will cost “well north” of $20,000 before bulk discounts, and a fully configured rack of 256 chips would run to around $10 million, depending on memory chip configuration, according to SemiAnalysis.
Most of the chips are initially going into the large, ready-to-install racks favoured by hyperscalers, with the simpler two-processor servers expected to ramp up later, the research firm said.
Nvidia expects $20 billion in revenue from Vera chip sales by the end of this fiscal year to end-January.
The Chinese interest in Vera comes as the global AI race pivots from model training to inference computing, the process of answering queries, where graphics processors face greater competition from CPUs and custom chips.
The shift has helped create a CPU shortage. Intel has notified Chinese customers of server CPU delivery lead times of up to six months, Reuters reported in February. Its rival AMD last month flagged that the global CPU market is “tight,” with demand outpacing forecasts and supply constraints expected to persist.
Based on Arm technology, Vera puts Nvidia in direct competition with Intel and AMD, which have long dominated the processor market with the x86 architecture.
Selling CPUs in China could prove less fraught than selling graphics processing units, which face tighter U.S. export controls. Washington has licensed about 10 Chinese firms to buy the H200, a GPU, but not a single delivery has been made. Chinese officials, keen to nurture domestic suppliers, have withheld approval.
Chinese clients plan to initially deploy Vera chips only in their overseas data centres for testing, one of the sources said.
(Reporting by Liam Mo, Ryan Woo in Beijing and Wen-Yee Lee in Taipei; Editing by Miyoung Kim and Sonali Paul)
Comments