Unlock Editor’s Digest for free
Roula Khalaf, editor-in-chief of the FT, selects her favourite stories in this weekly newsletter.
Nvidia is on track to sell $12 billion worth of artificial intelligence chips in China this year, despite U.S. export restrictions that limit trade in one of the world’s largest semiconductor markets.
The $3 trillion Silicon Valley group is expected to ship more than 1 million of its new H20 chips in the coming months, according to analyst forecasts. The chips are designed to avoid U.S. restrictions on the sale of AI processors to Chinese customers.
That number is nearly twice what Huawei expects to sell of its China-made competitor, the Ascend 910B, according to estimates from SemiAnalysis, a chip consulting firm.
Nvidia is the latest Silicon Valley company to find itself caught in the middle of tensions between Washington and Beijing. The Biden administration wants to stem the flow of the world’s most powerful chips to China, fearing Beijing could use them to create more powerful AI systems with military applications.
The resulting shortage of AI chips is negatively affecting the competitiveness of Chinese technology companies such as ByteDance, Tencent and Alibaba, which want to compete with American companies OpenAI, Microsoft, Meta and Google. This technology is radically changing the sector.
Each H20 chip will cost between $12,000 and $13,000, suggesting Nvidia will likely generate more than $12 billion in sales. That would be more than the $10.3 billion in revenue it generated from its entire China business — which includes sales of graphics chips to PC gamers and other products — in the fiscal year ending January 2024.
Nvidia declined to comment on the predictions. Huawei did not respond to a request for comment.
Since the Biden administration first imposed restrictions on sales of Nvidia’s most powerful AI chips in China in 2022, the US company has warned that its business will suffer if cloud computing providers and AI startups there turn to local alternatives such as Huawei.
“Our business in China is substantially lower than historical levels,” Jensen Huang, Nvidia’s CEO, said during the company’s most recent earnings call in May. “And it’s much more competitive in China now, because of the limitations of our technology… But we continue to do our best to serve customers in the markets there.”
Nvidia CFO Colette Kress said on the same call that revenue from the data center segment — which includes AI chips — in China in the last quarter was “significantly lower than levels before the new export restrictions were implemented in October.”
As recently as 2021, before the U.S. began imposing export controls, China accounted for more than a quarter of Nvidia’s total revenue. Even if the H20 chip sells as well as analysts expect, China could still account for closer to 10 percent of sales this year. But that also reflects the massive growth Nvidia is seeing among U.S. tech companies building ever-larger AI systems.
While Nvidia’s sales in China were down ahead of the new H20’s introduction this spring, analysts at both Morgan Stanley and SemiAnalysis say the chip is now shipping in large numbers and is proving popular with Chinese customers, despite its poorer performance compared to the chips Nvidia is able to sell in the U.S.
“Buyers report positive feedback on the potential competitiveness of H2O clusters,” Morgan Stanley wrote in a research note to clients this week, pointing to “strong Chinese demand.”
SemiAnalysis’ Dylan Patel said the H20’s capabilities were “on paper” below Huawei’s 910B, but that Nvidia’s chip performed “a lot better” in practice, thanks to superior memory performance.
He estimated that Huawei would sell about 550,000 910B chips in the same period, as the Shenzhen-based company and its manufacturing partners struggle to produce the complex processors in large enough quantities to meet demand.
Most Chinese AI companies have also built their AI models on Nvidia’s ecosystem and software. Switching to Huawei’s infrastructure would be time-consuming and expensive.
The Biden administration introduced restrictions on Nvidia’s ability to sell its most powerful chips, including the A100 and H100, to China in October 2022. Late last year, those controls were further tightened to exclude Nvidia’s newer chips as well. In November, Nvidia had begun rolling out a new set of chips customized for China, the most powerful of which is the H20.
Including chips for PC gamers, data centers and other customers, China accounted for about 9 percent of Nvidia’s total revenue in the most recent quarter ended in April, down from 22 percent in the same period a year earlier. Total revenue from China, including Hong Kong, still grew in the period, though, up more than 50 percent year over year to $2.5 billion.