Nvidia leads sell-off after report of more US AI chip curbs
Sign up now: Get ST's newsletters delivered to your inbox
An AI accelerator is a graphics processor, or GPU, tailored to train AI models by feeding them tons of data.
PHOTO: REUTERS
CALIFORNIA – Nvidia led declines in tech stocks after a report that Washington could close loopholes in the sale to China of powerful chips used to train artificial intelligence (AI),
The world’s most valuable chipmaker, which gets about a fifth of its revenue from China, slid as much as 3.2 per cent in after-hours trading in New York.
Rival Advanced Micro Devices (AMD) fell about 3 per cent.
The two lead the market for chips vital to the development of generative AI models such as ChatGPT.
In China, traders unloaded a slew of AI-related stocks, pushing Inspur Electronic Information Industry Co and Unisplendour – both key hardware suppliers – down 10 per cent.
Nvidia in 2023 designed less-capable chips that fall under thresholds that require a licence from the US Commerce Department before export to China or other countries of concern.
Washington is now weighing action as soon as July to expand the curbs to include those lower-powered semiconductors, The Wall Street Journal reported, citing anonymous sources.
Such a move underscores the Biden administration’s determination to contain China’s technological rise and could stoke tensions between the two countries.
The United States is increasingly concerned about Beijing’s technological ambitions, including around the use of AI in military and scientific advances that could tilt the geopolitical balance.
While that is likely to hurt Nvidia’s and AMD’s business with the world’s No. 2 economy, the two chipmakers remain at the forefront of a surge in AI development that is driving investment from the US to Europe and China.
From Microsoft to Baidu and ChatGPT developer OpenAI, companies around the world are buying their products to train the next generation of AI services.
US officials are also weighing whether to restrict the leasing of cloud services to Chinese AI companies, which have employed such platforms in training their models, the Journal reported, citing people familiar with the discussions.
Amazon.com and Microsoft are among the world’s biggest cloud service providers.
Nvidia declined to comment on the Journal report.
“Chinese AI firms may also be able to source dedicated AI chips from third-party countries. So I think it will be hard for (the) US to enforce the regulations,” said Bloomberg Intelligence analyst Robert Lea.
“While further restrictions could delay AI developments by Chinese firms, I don’t see a major long-term impact as Chinese firms have taken an increasingly innovative approach to workarounds.”
Nvidia has a more than 80 per cent share of the market for so-called data centre accelerator chips, and had been operating under rules that required approval for shipments to China of its A100 and new H100 parts.
It was able to partially alleviate the impact on its finances by selling a modified version of the A100 that is slower at accessing data, and therefore did not trigger the restriction.
An AI accelerator is a graphics processor, or GPU, tailored to train AI models by feeding them tonnes of data.
It is better suited for such tasks than a general-purpose central processing unit, or CPU, because its architecture can do parallel work in huge volumes.
Nvidia was the first to come up with a language to make GPUs do AI tasks, giving it a huge head start over rivals like AMD and Intel.
AMD was also subject to the initial rules announced last August.
While Nvidia’s much smaller rival emerged largely unscathed by the clampdown, Nvidia said last September that the restrictions would cost it US$400 million (S$540 million) in that quarter.
The report “may have hurt sentiment today”, said Union Bancaire Privee managing director Vey-Sern Ling. BLOOMBERG
