NextFin

China Reported to Convene Emergency Meetings on Nvidia H200 Chip Purchases as ByteDance, Alibaba Express Interest

Chinese regulators held emergency talks with major technology companies on Wednesday to assess demand for Nvidia Corp.'s advanced H200 chips, a day after reports emerged that Beijing plans to restrict domestic access to the semiconductors despite U.S. approval for exports.

AI Generated Image

AI Generated Image

Officials from the National Development and Reform Commission (NDRC) and the Ministry of Industry and Information Technology (MIIT) convened meetings with representatives from Alibaba Group, ByteDance and Tencent Holdings, asking them to evaluate their requirements for the H200, according to The Information. The companies were told they would soon be informed of the government's decision.

ByteDance and Alibaba have approached Nvidia about placing large orders for the H200 chip following U.S. President Donald Trump's announcement that he would permit exports to China, Reuters reported, citing four sources briefed on the matter. The two companies are eager to make substantial purchases should Beijing grant approval, though they remain concerned about supply constraints.

The developments underscore Beijing's dilemma between supporting artificial intelligence (AI) advancement that requires powerful chips China cannot yet manufacture and pushing adoption of domestic alternatives to achieve technological self-sufficiency. Trump said Monday that the U.S. would allow Nvidia to ship H200 products to "approved customers" in China, with 25% of sales going to the federal government.

Beijing Considers Access Restrictions Despite U.S. Green Light

Chinese regulators are discussing ways to permit only limited access to the H200, Nvidia's second-most advanced generation of artificial intelligence chips, according to the Financial Times. Buyers would likely need to undergo an approval process, submitting purchase requests and explaining why domestic providers cannot meet their needs, the newspaper reported, citing two people with knowledge of the matter.

No final decision has been made. The regulators may apply additional measures to ensure competitiveness of domestic chips, including potentially banning China's public sector from purchasing the H200, according to the Financial Times sources.

The H200 is significantly more powerful than the H20 chip Nvidia previously sold to China and had been banned under the Biden administration over concerns about potential military applications. The H200 is almost six times as powerful as the H20, making it capable of training AI models at levels currently unmatched by Chinese alternatives.

Tech Giants Assess Demand in Government Consultations

Wednesday's emergency meetings followed letters sent to technology companies a day earlier requesting information on how many H200 chips they require for AI model development, according to The Information, citing two people involved in the correspondence.

Officials told the companies during the meetings that Beijing would communicate its decision after compiling their responses. The consultations highlight Chinese policymakers' challenge in balancing immediate AI development needs against long-term goals of technological independence.

Chinese companies have been training AI models abroad to access Nvidia chips banned domestically. The return of more advanced Nvidia chips would be welcomed by tech giants that have been using more Chinese semiconductors for basic AI functions but still prefer Nvidia products for their higher performance and easier maintenance, according to the Financial Times.

ByteDance and Alibaba Signal Strong Purchase Intent

ByteDance and Alibaba are keen to place large orders for the H200 should Beijing grant approval, two Reuters sources said. However, the companies are seeking clarity from Nvidia on supply issues, with very limited quantities of H200 currently in production as Nvidia focuses on its most advanced Blackwell and upcoming Rubin lines, two people familiar with the supply chain told Reuters.

Chinese companies anticipate authorities may need to review purchase requests and require them to provide use cases as Beijing weighs costs and benefits of allowing H200 imports while encouraging sales of AI chips manufactured domestically by companies like Huawei and Cambricon, Reuters sources said.

"The training of leading Chinese AI models still relies on Nvidia cards," Zhang Yuchun, general manager at Chinese cloud service provider SuperCloud's solution and ecology units, told Reuters. "I expect the leading Chinese tech companies to buy a lot although in a low-key manner."

Self-Sufficiency Goals Shape Approval Calculus

The bottom line for allowing H200 purchases is that it would not derail China's long-term goal of technological self-sufficiency, according to three people familiar with the Chinese government's thinking cited by The Information. Officials view buying high-end American chips as a temporary measure to address the shortage of powerful training chips while Chinese manufacturers catch up.

One option under consideration involves imposing caps on Nvidia chip purchases relative to the amount of domestic processors companies already have and plan to buy, though whether the ratio would be measured by dollar value or chip count remains unclear, according to two people close to the Chinese government cited by The Information.

China has used export restrictions to push domestic chipmakers to compete with Nvidia through measures including stepped-up customs checks of chip imports and energy subsidies to data centers using domestic chips. Over recent months, Beijing has barred government-funded data centers and Chinese tech companies from buying Nvidia's AI chips, pummeling the company's market share in China, Reuters has reported.

Nvidia's revenue from China plunged 63% to about $3 billion in the three months ending October from the year-earlier period, largely due to its inability to sell advanced chips there, according to the Information. Thecompany took in just $50 million in H20 revenue in the October quarter, compared to $4.6 billion between February and April.

Explore more exclusive insights at nextfin.ai.