Beijing is reportedly set to give the green light for Chinese technology firms to import Nvidia’s H200 artificial intelligence (AI) chips, even as an industry heavyweight warns companies not to become overly dependent on foreign semiconductors.
Both Alibaba Group Holding and ByteDance have told Nvidia privately that they are interested in ordering more than 200,000 units of its H200 chips, Bloomberg reported. With Nvidia currently holding about 700,000 H200 units in inventory, the combined demand could be met swiftly once Beijing grants approval for the purchases.
Nvidia told its Chinese customers last month that it aims to begin shipping H200 chips to China before the Lunar New Year holiday in February, with an initial batch of 5,000 to 10,000 modules, equivalent to roughly 40,000 to 80,000 H200 AI chips, Reuters reported. The Year of the Horse under the Chinese zodiac calendar begins on February 17.
“From the perspective of industry development, the rational flow of high-end computing resources helps promote the application and exploration of cutting-edge technologies such as AI,” Wei Shaojun, vice president of the China Semiconductor Industry Association (CSIA) and a professor at Tsinghua University, told the Global Times in an interview.
“We have always believed that technology has no borders and that cooperation drives progress,” the industry expert said. “As long as it complies with the regulatory requirements of both sides, any positive interaction that benefits technological innovation and industrial development is welcome.”
Wei said Washington has an inconsistent approach, shifting between tightening pressure and easing curbs on high-end chips. He questioned whether the latest so-called relaxation of controls reflects a genuine effort to promote constructive engagement, or a calculated tactic aimed at slowing China’s development momentum and encouraging complacency.
“China’s semiconductor industry must remain highly vigilant toward this, resolutely refuse to be deceived by superficial appearances and, even more importantly, must not waver in its confidence and determination to persist on the path of indigenous development in advanced process technologies and other key areas,” Wei said.
Although Wei is not a government official, he is widely regarded as one of China’s most influential industry voices commenting on Nvidia’s H200 imports. Beijing has not clearly indicated whether it will approve H200 imports since US President Donald Trump said on December 8 that Washington would allow Nvidia to supply the chips to approved Chinese customers.
Wei’s remarks followed comments by Nvidia chief executive Jensen Huang on January 6 at the Consumer Electronics Show in Las Vegas, where he said demand from Chinese customers for the H200 remains strong. Huang has previously said the AI chip market in China could generate about US$50 billion annually for Nvidia in the coming years.
Media reports said that Nvidia has submitted export license applications to Washington and that final details of US government approval are still being completed.
Play by China’s rules
Last November, the US Department of Commerce reportedly considered allowing the export of Nvidia’s H200 chips to China. At the time, some Chinese commentators cautioned that such a move could act as “sugar-coated bullets,” offering short-term relief while potentially slowing the long-term development of China’s domestic AI chip industry.
Following a phone call between US President Donald Trump and Chinese President Xi Jinping on November 24, Beijing began signaling more positively about H200 imports.
Foreign ministry spokesperson Guo Jiakun said on December 9 that China supports pursuing mutual benefit with the US through cooperation.
Some Chinese commentators have described this as a “two-track” approach, in which Chinese firms use H200 chips for high-end tasks such as AI model training while relying on domestic chips for AI inference – the term refers to a model answering people’s questions.
Citing people familiar with the matter, Reuters reported on December 31 that Chinese technology companies have placed orders for more than 2 million H200 chips, to be delivered in 2026, far exceeding Nvidia’s current inventory of about 700,000 units. The report said Taiwan Semiconductor Manufacturing Company (TSMC) will begin to produce more in the second quarter of 2026.
A Shanghai-based columnist writing under the pen name Yan Ge says the core message of Wei Shaojun’s comments is that, as long as China continues to develop and improve its own AI chips, it will achieve AI chip self-sufficiency.
“Foreign chips are welcome, but they must play by China’s rules,” he says. “Chinese firms may buy foreign chips, but they must continue supporting domestic chipmakers and their research and development. Otherwise, when the next round of US export controls arrives, it will be too late to react.”
“Wei has repeatedly stressed in public speeches that around 90% of AI use cases do not actually need to use 7nm or 5nm chips – chips with a combination of 28nm processes, advanced packaging and chiplet design are good enough,” Yan says, adding that such an approach can cut power consumption by about 30% and costs by roughly 40%.
Upfront payment
The Information reported on January 7 that Beijing has asked some Chinese technology firms to halt new orders for H200 chips and is expected to mandate purchases of domestically produced artificial intelligence chips, citing people familiar with the situation.
Reuters also reported on January 8 that Nvidia is requiring full upfront payment from Chinese customers seeking H200 chips, a shift from its past practice of accepting deposits. They said the change reflects Nvidia’s effort to hedge against uncertainty over Beijing’s approval of the shipments.
Over the long run, some Chinese pundits say China’s strategy is to shift the global AI competition away from an exclusive focus on cutting-edge process nodes and toward system-level engineering built on mature chips.
“From 28nm to 7nm and now 3nm or 2nm, each step packs in more transistors in a chip. But this race is nearing its physical limits, making cost a huge bottleneck,” says Geng Biao, a Liaoning-based columnist.
“A single 3nm chip production line can cost more than 200 billion yuan (US$28.7 billion) in investment, meaning that a one percentage point gain in yield can be worth several billion yuan,” he says. “Only a handful of global giants, such as Samsung and TSMC, can bet on this.”
He says advanced packaging is changing how performance gains are achieved, as stacking multiple chiplets together can deliver sufficient computing power for most AI tasks. He points to AMD’s MI300X as an example of how combining several chips can significantly boost performance while keeping costs down.
Read: China aims for twin-track use of Nvidia H200 amid back door fears
Follow Jeff Pao on Twitter at @jeffpao3