Nvidia’s $5 Billion
of China Orders in Limbo After Latest U.S. Curbs
Tech company had been pushing
to make chip shipments for next year before new restrictions came into effect
New
U.S. export controls may compel artificial-intelligence giant Nvidia to cancel billions of dollars in next-year orders
for its advanced chips to China, a move that could deprive Chinese tech companies
of crucial AI resources.
The
Santa Clara, Calif.-based company had already finished delivering orders of its
advanced AI chips to China for this year, according to people familiar with the
matter, and was pushing to deliver some 2024 orders in advance before the new rules
were scheduled to come into effect in mid-November.
Then
the U.S. government told Nvidia in a letter last week that the new export restrictions
on the sale of high-end chips to countries including China were instead effective
immediately.
China’s
biggest AI and cloud-computing companies including Alibaba Group, TikTok owner ByteDance
and Baidu had made large orders for delivery next year, the people said. Orders
from major Chinese companies for 2024 exceeded $5 billion, one of the people said.
A
spokesman for Nvidia said the company has been working to allocate its advanced
AI computing systems, which use graphics chips affected by the rules, to customers
in the U.S. and elsewhere and is pursuing additional supply. “These new export controls
will not have a meaningful impact in the near term,” the spokesman said.
Earlier
this year, Colette Kress, Nvidia’s chief financial officer, said that in the longer
term prohibiting sales of AI chips to China would result in a permanent loss of
opportunities for the U.S. chip industry.
The
restrictions on AI chips are part of a wider effort by the Biden administration
to curtail China’s access to advanced chips, AI tools and other technology that
the U.S. believes China could use to advance its military and cyberwarfare capabilities.
The
latest rule, announced on Oct. 17, requires any company whose AI chips exceed a
performance benchmark to seek a license from the U.S. Commerce Department before
exporting them to China and other countries of concern.
Nvidia’s
China stakes
Nvidia
bore the brunt of the restriction because its AI chips, which are essential in creating
popular AI tools such as OpenAI’s ChatGPT, are the most advanced and widely deployed
in the world. Surging interest in AI has sent Nvidia’s sales and stock price skyward,
crossing a $1 trillion valuation earlier this year.
In
its most recent four quarters, Nvidia has reported around $22 billion of total revenue
in the data-center division that houses its AI chips.
After
the U.S. government imposed a less-stringent set of chip restrictions late last
year, Chinese companies raced to order Nvidia’s A800 and H800, AI chips modified
for the Chinese market to comply with the regulations. Under the tougher rules announced
on Oct. 17, however, Nvidia must cancel the orders unless it gets export licenses
from the U.S. Commerce Department.
Nvidia
stopped taking new orders from China for its advanced AI chips in the wake of the
new restrictions, according to people familiar with the matter. The company planned
to rush some deliveries of previously placed orders during the rule’s 30-day grace
period, the people said.
The
Commerce Department’s decision to make the new curbs effective immediately meant
deliveries were no longer possible.
Nvidia
didn’t give priority to production of A800 and H800 chips this year, according to
people familiar with the matter. Instead it planned to increase sales of the China-targeted
chips in 2024, they said.
Nvidia
said its L40S, an AI chip it made available in August, is also affected by the U.S.
export curb. Some Chinese companies had talked with Nvidia about increasing purchases
of L40S GPUs weeks before the Oct. 17 new rules, as the industry speculated that
A800 and H800 were likely to be banned by the U.S. government, the people said.
The L40S is designed to be better at operating AI systems than building them, making
it a less suitable alternative for those other two chips.
Chip ban may slow China’s AI advances
The
latest chip restrictions are likely to slow China’s pace in developing advanced
AI capabilities and force Chinese developers to use homegrown alternatives, industry
executives and analysts say.
The
rules now cover most of the high-performance AI and data-center chips from Nvidia,
Intel and Advanced Micro Devices. Chinese companies will have to rely on inventory,
use a larger number of less advanced chips or find other workarounds.
Analysts
at Bernstein Research estimated in a report last week that using Nvidia’s V100,
a less-capable AI chip launched in 2017, would lead to 30% higher costs to train
AI systems as it requires more chips and thus consumes more energy.
Chinese
companies have started sourcing homegrown chips, including Huawei Technologies’
Ascend 910 and Cambricon Technologies’ Siyuan 590.
The
founder of Chinese speech-recognition company iFlytek said in August that Ascend
had achieved capabilities and performance comparable to Nvidia’s A100. However,
U.S. rules have barred leading foundries such as Taiwan Semiconductor Manufacturing,
or TSMC, from manufacturing many of these chips.
Other
Chinese companies such as Baidu and Alibaba have also developed their own AI chips
and are counting on better algorithms and software to achieve higher performance
with less advanced chips.