Nvidia weighs expanding H200 output after China rush

Nvidia is weighing a plan to expand output for its H200 chips after Chinese buyers pushed demand far beyond what the company can produce right now. According to Reuters, the company told clients in China that it is studying ways to add capacity because orders have gone past its current limits.

The push began after U.S. President Donald Trump said the government would let Nvidia export the H200 into China as long as the company pays a 25% fee on those sales. That rule turned the H200 into the one chip every major Chinese firm now wants to secure.

Chinese companies jumped on the news fast. Firms like Alibaba and ByteDance contacted Nvidia this week to ask for large batches and requested details on how soon they can get them.

Nvidia also allegedly said that it is working to “manage our supply chain to ensure that licensed sales of the H200 to authorized customers in China will have no impact on our ability to supply customers in the United States.”

But Chinese approval is still missing. Three people aware of the situation said officials in Beijing held emergency meetings on Wednesday to decide whether the chip will be allowed into the country at all.

China weighs approvals and demand pressures

Chinese buyers are stressed because supplies are tight. Reuters reported that only very small amounts of H200 chips are being made right now.

Nvidia is busy building its Blackwell line and preparing its Rubin line, which has pulled resources away from the H200. China wants answers, and Nvidia has been briefing clients on current supply levels, though it did not give them a hard number.

The H200 started mass use last year. It is the fastest chip in the older Hopper family and is built by TSMC using its 4nm process. TSMC did not comment on how much capacity is allocated to Nvidia but pointed back to comments from its chairman C.C. Wei about rising AI demand and how the company plans long-term production.

Chinese firms want the H200 because it is the strongest chip they can legally buy. It delivers about six times the performance of the H20, a chip Nvidia designed for China in late 2023 with limits to meet U.S. rules.

China is pushing its own chip ecosystem at the same time. Domestic firms have not reached H200-level performance, and that gap has triggered worries that letting the H200 into the country could slow local progress.

Nori Chiou, investment director at White Oak Capital Partners, said the H200’s compute power is “approximately 2-3 times that of the most advanced domestically produced accelerators.”

He added that cloud providers and enterprise users are placing huge orders and lobbying the government to relax restrictions. Demand in China is much larger than what local production can cover.

Officials also discussed tying H200 imports to purchases of domestic chips. Two people aware of the talks said the idea is to require buyers to purchase a set amount of local chips for every H200 they import.

Nvidia studies new capacity as industry tensions rise

Any capacity increase is tricky for Nvidia. The company is shifting to Rubin and competing with firms like Google for limited high-end manufacturing slots at TSMC.

Even if Nvidia wants more H200 units, it must fight for room on the same production lines everyone else wants.

China is also trying to boost its own chip capabilities through SMIC. The Kirin 9030 chip, used in Huawei’s Mate 80 phones, is built with SMIC’s updated 7nm process. Research firm TechInsights said the chip uses SMIC’s N+3 process, which is a scaled extension of its N+2 node.

But it also said the N+3 process is still far behind the 5nm technology used by TSMC and Samsung. China even placed TechInsights on its unreliable entity list in October after the firm published reports on Huawei and SMIC’s progress.

Source