Tech giant Nvidia has approached Chinese customers about purchasing its latest “Vera” central processing units designed for artificial intelligence data centers, with potential delivery beginning in August, according to three sources with knowledge of the discussions.
This marketing effort highlights how the technology company is rapidly shifting focus to this new product line as it attempts to recover its severely damaged business in China, where deliveries of its H200 AI chip have been stalled for several months.
The company’s CEO Jensen Huang revealed in October that Nvidia’s market presence in China has essentially dropped to nothing, damaged by American export restrictions on advanced semiconductors and China’s drive toward technological independence.
This development also intensifies the competitive battle with major processor manufacturers Intel and AMD, as all companies work to boost their server CPU production for AI data center applications.
According to the sources, several Chinese companies have expressed interest in the Vera processor, which marks Nvidia’s inaugural standalone central processing unit designed specifically for autonomous AI systems. The sources requested anonymity due to the confidential nature of these business discussions.
Currently in full manufacturing mode, the Vera processor handles the background computational work that AI systems require, with Nvidia claiming performance speeds up to 1.8 times faster than competing processors.
During the March introduction of Vera, Huang projected the product would generate billions in revenue for the company. Nvidia announced at that time that major cloud service providers including Alibaba and ByteDance were working with the company on Vera deployment, though it remained unclear when actual ordering would commence.
Nvidia chose not to provide comment for this report. Both Alibaba and ByteDance failed to respond to requests for statements.
One significant Chinese cloud computing firm is considering purchasing more than 300 servers, with each containing two Vera processors, according to one source. The company intends to use these systems for initial testing before making final purchasing decisions based on performance results, the source explained.
However, whether this early interest will lead to widespread adoption remains uncertain, partly due to software compatibility challenges and the complexity of switching from existing domestic AI chip infrastructure, another source noted.
Research firm SemiAnalysis estimates that individual Vera processors will cost “well north” of $20,000 before volume discounts, while a complete rack containing 256 chips could reach approximately $10 million, depending on memory specifications.
Initially, most chips will be installed in large, ready-to-deploy racks preferred by major cloud providers, with simpler two-processor server configurations expected to increase production later, the research company stated.
Nvidia anticipates generating $20 billion in Vera chip revenue by the conclusion of its fiscal year ending in January.
Chinese interest in Vera emerges as the worldwide AI industry transitions from model development to inference computing – the process of responding to user queries – where graphics processors encounter increased competition from CPUs and specialized chips.
This industry shift has contributed to a CPU supply shortage. Intel informed Chinese customers in February about server CPU delivery delays extending up to six months, according to previous reporting. Competitor AMD indicated last month that global CPU markets remain “tight,” with demand exceeding projections and supply limitations expected to continue.
Built on Arm technology, Vera positions Nvidia as a direct competitor to Intel and AMD, companies that have historically controlled the processor market through x86 architecture.
Marketing CPUs in China may face fewer regulatory obstacles compared to graphics processing units, which encounter stricter American export restrictions. While Washington has authorized approximately 10 Chinese companies to purchase the H200 GPU, no actual deliveries have occurred as Chinese authorities, focused on supporting domestic suppliers, have not granted approval.
Chinese customers plan to initially test Vera chips exclusively in their international data centers, one source revealed.
