Nvidia has developed three new chips tailored for China that aim to meet the region’s growing demand for artificial intelligence technology while complying with US export controls, according to leaked documents and four people familiar with the situation.
The latest effort marks the second time in little more than a year that Silicon Valley-based Nvidia has been forced by new US regulations to reconfigure its products for Chinese customers, as it strives to maintain its foothold in one of its most important markets.
Nvidia is preparing to launch the new chips just weeks after the US restricted sales to China of high-performance chips that can be used to create AI systems, in the Biden administration’s latest salvo in a tit-for-tat tech war between the two superpowers.
The three new Nvidia chips are named the H20, L20 and L2, according to a document distributed by the company to prospective customers that was obtained by the Financial Times.
The overall performance of these chips has been moderated compared with those that Nvidia had previously sold in China. Nonetheless, the new graphics processing units were expected to remain competitive in the Chinese market, said the people familiar with the situation.
“Nvidia is perfectly straddling the line on peak performance and performance density with these new chips to get them through the new US regulations,” wrote analysts at SemiAnalysis, a chip consultancy, in a note to clients on Thursday.
Nvidia did not immediately respond to a request for comment.
Nvidia was co-founded by Jensen Huang, who is also its chief executive. The company’s market value soared to more than $1tn this year driven by investor enthusiasm about its dominant role in the processors needed to develop AI systems. Its A100 and H100 chips have become the most sought-after components for AI companies around the world that want to create large language models, the technology that underpins chatbots such as OpenAI’s breakthrough ChatGPT.
As the US sought to constrain China’s AI development, the Biden administration blocked sales of the A100 and H100 GPUs in October 2022. In response, Nvidia developed two alternative models for China, the A800 and H800, which fell below the performance threshold set by US sanctions. But the US last month tightened its restrictions so that they also caught the A800 and H800.
The latest export restrictions took effect immediately as the US government speeded up the deadline, leaving Chinese tech groups dependent on outdated and stockpiled chips to pursue their AI ambitions. The rules were seen as forcing Chinese groups to turn to six-year-old technology to develop AI systems.
But Nvidia, which has held a dominant share of China’s AI chip market, is moving quickly too. The manufacturing process of its latest chips for China was less complex than the development of the A800 and H800, said a person familiar with the situation. Nvidia has already sent samples of the chips for customers to test, suggesting it expects mass production to begin very soon, according to two people close to the company.
In the interim, Chinese companies have redoubled their efforts to source AI chips from domestic suppliers, reducing the risk of relying on Nvidia and accommodating the escalating AI chip ban. Prominent Chinese Nvidia competitors include Huawei, Cambricon and Biren. The founder of Chinese AI company iFlytek said in August that Huawei’s Ascend AI chip could achieve performance comparable to Nvidia’s A100.
However, Nvidia’s Chinese rivals are all constrained by geopolitical conflicts that prevent them from producing chips outside of China, while international sanctions have also sought to limit their access to advanced chipmaking equipment from suppliers such as Netherlands-based ASML.