Nvidia is attempting to rebuild its position in China, but not through the graphics chips that have fuelled the artificial intelligence boom in recent years. The company has begun talks with Chinese customers regarding the Vera processor, its first standalone CPU designed specifically for so-called agent-based AI, i.e. systems capable of performing tasks autonomously. According to Reuters, orders could begin as early as August.
This marks a significant shift for the US manufacturer. Sales of advanced GPUs to China remain restricted by US export regulations, and the Chinese authorities are also promoting domestic technology suppliers. As a result, Nvidia’s share of the Chinese AI market has virtually disappeared. However, processors may prove easier to sell than the most advanced AI accelerators.
Vera is intended to be a response to the changes taking place in the industry. Inferential computing is becoming increasingly important – this is the stage at which models respond to user queries. It is in such applications that the role of central processing units, hitherto dominated by Intel and AMD, is growing. Nvidia claims that Vera offers up to 1.8 times higher performance than comparable processors from competitors and has been designed specifically for the needs of AI agents.
Interest in the new chip is already emerging at the testing stage. According to Reuters sources, one of China’s major cloud companies plans to order over 300 servers, each equipped with two Vera processors. However, final purchasing decisions are expected to depend on the results of pilot deployments.
Nvidia makes no secret of its ambitions for this new segment. During a presentation in March, Jensen Huang described Vera as the foundation of the company’s next multi-billion-dollar business. Partners collaborating on the processor’s implementation include Alibaba Cloud and ByteDance.
At the same time, success in China is by no means guaranteed. Local data centre operators are increasingly using domestic chips, and migrating existing AI environments to the new architecture takes time. Nevertheless, Vera shows that Nvidia does not intend to limit itself to the GPU market. The company is increasingly competing for control over the entire AI infrastructure stack – from processors and accelerators to networks and complete systems for data centres.
