Tech war: Nvidia sampling alternative AI chips to China as US restrictions take a toll there

US chip giant Nvidia said it has started shipping alternative samples of permissible artificial intelligence (AI) chips to Chinese customers, as Washington’s restrictions on sales of advanced semiconductors to the mainland begin to take a toll on its business in China.

Nvidia reported upbeat results for its fourth quarter ended January 28 on the back of robust demand worldwide for its data centre graphics processing units (GPUs), with revenue up 265 per cent year on year to US$22.1 billion, including US$18.4 billion generated from the data center segment.

But China was an exception. “Growth was strong across all regions except for China, where our data centre revenue declined significantly, following the US government export control regulations imposed in October,” Nvidia’s chief financial officer Colette Kress said in an earnings conference call with analysts on Tuesday.

Semiconductor giant TSMC to build second plant in Japan with new investor Toyota

Kress said Nvidia has started shipping alternative GPUs that do not require a licence for the Chinese market under the tighter export regime, adding that the firm has not received licences from the US government to ship restricted products to the mainland. 

She expects China to represent a mid-single-digit percentage of its data centre revenue in the next quarter, flat with the fourth quarter.

Nvidia has been among the most-affected American tech companies after the US Commerce Department tightened export controls in October, citing national security concerns.

While Nvidia’s net profit skyrocketed 581 per cent to US$29.8 billion in the 12 months ended January 28, its prospects in China remain under a cloud. The trade sanctions prevent Nvidia from exporting to China its advanced GPUs, such as the A100 and H100, which have become sought-after for training AI systems. Its tailor-made A800 and H800 GPUs, developed as workarounds for Chinese clients in 2022, have also been blocked by the updated US controls imposed last October, leading to new alternatives such as the H20, L20 and L2 GPUs.

“At the core, the US government wants to limit the latest capabilities of Nvidia’s accelerated computing and AI to the Chinese market,” Jensen Huang, the company’s billionaire chief executive, said on the call.

He added that following the October restrictions, Nvidia has “reconfigured its products in a way that is not software hackable in any way”, which took some time. He added that Nvidia is sampling products with customers in China.

Nvidia has started to take pre-orders for its H20, a GPU that can be used by China-based customers for AI training, with a single card price range between US$12,000 and US$15,000, the Post reported previously.

China, including Hong Kong, was Nvidia’s third-largest single market after the US and Taiwan in its financial year 2023. Greater China, which includes Taiwan, accounted for 21.5 per cent of the company’s total revenue, down from 26 per cent in the previous year.