Nvidia denies China-specific AI chip on roadmap

On August 20, Nvidia firmly denied a report by The Information that it was planning to launch a language processing unit (LPU) for the Chinese market by the end of the year. The company's spokesperson explicitly stated that it currently has no LPU sales in China and no China-specific LPU product on its roadmap. This denial comes after Reuters had reported in March that Nvidia was developing a China-compatible version of its AI chips to compete in the rapidly growing AI inference market.

The Information had cited two unnamed Nvidia employees in an August 20 report, claiming that the company intended to begin small-batch shipments of an AI chip for Chinese customers by the end of 2026. It was also reported that several customers had already placed orders for this chip, which was described as a version of Nvidia's LPU.

This processor, developed using technology licensed from start-up Groq, was said to work alongside graphics processing units (GPUs) to expedite AI chatbot responses. The report further mentioned that the chip complied with US export-control rules and that Nvidia had modified its software to enable the LPU to operate with processors available in China following its next-generation Vera Rubin AI system's unavailability in the market due to US restrictions.

Nvidia's presence in the Chinese market has been uncertain, with the US approving sales of its H200 AI chips to a limited number of Chinese firms in May, but deliveries only starting recently. Nvidia has also been promoting its Vera CPU to Chinese customers. In May, Nvidia's CEO, Jensen Huang, acknowledged that the company had largely given up on the Chinese AI chip market, with Huawei Technologies becoming the main domestic rival.