US microchip export controls imposed last year to freeze China’s development of supercomputers used to develop nuclear weapons and artificial intelligence (AI) systems such as ChatGPT are having only minimal effects on China’s tech sector.
The rules restricted shipments of Nvidia Corp and Advanced Micro Devices Inc chips that have become the global technology industry’s standard for developing chatbots and other AI systems.
However, Nvidia has created variants of its chips for the Chinese market that are slowed down to meet US rules.
Photo: AFP
Industry experts said the newest one — the Nvidia H800, announced in March — is likely to take 10 percent to 30 percent longer to carry out some AI tasks and could double some costs compared with Nvidia’s fastest US chips.
Even the slowed Nvidia chips represent an improvement for Chinese firms. Tencent Holdings Ltd (騰訊), one of China’s largest tech companies, last month estimated that systems using Nvidia’s H800 would cut the time it takes to train its largest AI system by more than half, from 11 days to four days.
“The AI companies that we talk to seem to see the handicap as relatively small and manageable,” said Charlie Chai, a Shanghai-based analyst with 86Research.
The back-and-forth between government and industry exposes the US challenge of slowing China’s progress in the high-tech sector without hurting US companies.
Part of the US strategy in setting the rules was to avoid such a shock that the Chinese would ditch US chips altogether and redouble their own chip development efforts.
“They had to draw the line somewhere, and wherever they drew it, they were going to run into the challenge of how to not be immediately disruptive, but how to also over time degrade China’s capability,” said one chip industry executive who asked to remain anonymous as they were discussing private discussions with regulators.
The export restrictions have two parts. The first puts a ceiling on a chip’s ability to calculate extremely precise numbers, a measure designed to limit supercomputers that can be used in military research.
Chip industry sources said that was an effective action.
However, calculating extremely precise numbers is less relevant in AI work such as large language models where the amount of data the chip can chew through is more important.
Nvidia is selling the H800 to China’s largest technology firms, including Tencent, Alibaba Group Holding Ltd (阿里巴巴) and Baidu Inc (百度), for use in such work, although it has not yet started shipping the chips in high volumes.
“The government isn’t seeking to harm competition or US industry, and allows US firms to supply products for commercial activities, such as providing cloud services for consumers,” Nvidia said in a statement last week.
China is an important customer for US technology, it added.
“The October export controls require that we create products with an expanding gap between the two markets,” Nvidia said last week. “We comply with the regulation while offering as competitive as possible products in each market.”
Nvidia chief scientist Bill Dally said in a separate statement this week that “this gap will grow quickly over time as training requirements continue to double every six to 12 months.”
A spokesperson for the Bureau of Industry and Security, the arm of the US Department of Commerce that oversees the rules, did not return a request for comment.
The second US limit is on chip-to-chip transfer speeds, which does affect AI. The models behind technologies such as ChatGPT are too large to fit onto a single chip. Instead, they must be spread over many chips — often thousands at a time — which all need to communicate with one another.
Nvidia has not disclosed the China-only H800 chip’s performance details, but a specification sheet seen by Reuters shows a chip-to-chip speed of 400 gigabytes per second, less than half the peak speed of 900 gigabytes per second for Nvidia’s flagship H100 chip available outside China.
Some in the AI industry believe that is still plenty of speed.
Naveen Rao, CEO of the start-up MosaicML, which specializes in helping AI models to run better on limited hardware, estimated a 10 to 30 percent system slowdown.
“There are ways to get around all this algorithmically,” he said. “I don’t see this being a boundary for a very long time — like 10 years.”
Money helps. A chip in China that takes twice as long to finish an AI training task than a faster US chip can still get the work done.
“At that point, you’ve got to spend US$20 million instead of US$10 million to train it,” said one industry source, who asked to remain anonymous because of agreements with partners.
“Does that suck? Yes it does, but does that mean this is impossible for Alibaba or Baidu? No, that’s not a problem,” they said.
Moreover, AI researchers are trying to slim down the massive systems they have built to cut the cost of training products similar to ChatGPT and other processes. They would require fewer chips, reducing chip-to-chip communications and lessening the effect of the US speed limits.
Two years ago, the industry was thinking AI models would get bigger and bigger, said Cade Daniel, a software engineer at Anyscale, a San Francisco start-up that provides software to help companies perform AI work.
“If that were still true today, this export restriction would have a lot more impact,” Daniel said. “This export restriction is noticeable, but it’s not quite as devastating as it could have been.”
Semiconductor shares in China surged yesterday after Reuters reported the US had ordered chipmaking giant Taiwan Semiconductor Manufacturing Co (TSMC, 台積電) to halt shipments of advanced chips to Chinese customers, which investors believe could accelerate Beijing’s self-reliance efforts. TSMC yesterday started to suspend shipments of certain sophisticated chips to some Chinese clients after receiving a letter from the US Department of Commerce imposing export restrictions on those products, Reuters reported on Sunday, citing an unnamed source. The US imposed export restrictions on TSMC’s 7-nanometer or more advanced designs, Reuters reported. Investors figured that would encourage authorities to support China’s industry and bought shares
FLEXIBLE: Taiwan can develop its own ground station equipment, and has highly competitive manufacturers and suppliers with diversified production, the MOEA said The Ministry of Economic Affairs (MOEA) yesterday disputed reports that suppliers to US-based Space Exploration Technologies Corp (SpaceX) had been asked to move production out of Taiwan. Reuters had reported on Tuesday last week that Elon Musk-owned SpaceX had asked their manufacturers to produce outside of Taiwan given geopolitical risks and that at least one Taiwanese supplier had been pushed to relocate production to Vietnam. SpaceX’s requests place a renewed focus on the contentious relationship Musk has had with Taiwan, especially after he said last year that Taiwan is an “integral part” of China, sparking sharp criticism from Taiwanese authorities. The ministry said
US President Joe Biden’s administration is racing to complete CHIPS and Science Act agreements with companies such as Intel Corp and Samsung Electronics Co, aiming to shore up one of its signature initiatives before US president-elect Donald Trump enters the White House. The US Department of Commerce has allocated more than 90 percent of the US$39 billion in grants under the act, a landmark law enacted in 2022 designed to rebuild the domestic chip industry. However, the agency has only announced one binding agreement so far. The next two months would prove critical for more than 20 companies still in the process
CHANGING JAPAN: Nvidia-powered AI services over cellular networks ‘will result in an artificial intelligence grid that runs across Japan,’ Nvidia’s Jensen Huang said Softbank Group Corp would be the first to build a supercomputer with chips using Nvidia Corp’s new Blackwell design, a demonstration of the Japanese company’s ambitions to catch up on artificial intelligence (AI). The group’s telecom unit, Softbank Corp, plans to build Japan’s most powerful AI supercomputer to support local services, it said. That computer would be based on Nvidia’s DGX B200 product, which combines computer processors with so-called AI accelerator chips. A follow-up effort will feature Grace Blackwell, a more advanced version, the company said. The announcement indicates that Softbank Group, which until early 2019 owned 4.9 percent of Nvidia, has secured a
RSS