TLDR
- Nvidia is considering increasing production capacity for its H200 AI chips after orders from Chinese clients exceeded current output levels
- Trump’s administration approved H200 exports to China with a 25% fee, marking a shift from previous restrictions that limited shipments to the less powerful H20 model
- Major Chinese companies including Alibaba and ByteDance have reached out to Nvidia about placing large H200 orders
- The H200 is roughly six times more powerful than the H20 and 2-3 times more powerful than China’s most advanced domestic chips
- Chinese officials held emergency meetings to discuss whether to allow H200 imports, with potential requirements to bundle purchases with domestic chips
Nvidia shares edged higher in premarket trading Friday following reports that the chipmaker is evaluating additional production capacity for its H200 AI processors. The move comes in response to strong demand from Chinese clients that has exceeded current supply levels.
The company has informed Chinese customers about the potential capacity expansion. Sources familiar with the discussions said Nvidia is leaning toward adding new production capability.
Earlier this month, President Trump announced the U.S. would permit Nvidia to export H200 processors to China. The sales would include a 25% fee collected by Washington.
The decision represented a change from previous export rules. Those restrictions limited Nvidia to shipping only the H20 model to China, a less capable chip designed specifically for the Chinese market.
Major Chinese technology companies moved quickly after the announcement. Alibaba and ByteDance contacted Nvidia this week about purchasing H200 chips and are interested in placing large orders.
Supply Concerns Drive Customer Inquiries
Chinese clients have expressed concerns about H200 availability. Multiple companies reached out to Nvidia seeking information about supply levels.
The company has provided guidance on current production capacity during these briefings. Specific numbers were not disclosed.
Very limited quantities of H200 chips are currently in production. Nvidia has focused its manufacturing efforts on the more advanced Blackwell and upcoming Rubin chip lines.
The H200 entered mass deployment last year as the fastest AI chip in Nvidia’s Hopper generation. TSMC manufactures the processor using its 4nm technology.
Performance Gap Creates Strong Demand
The H200’s superior performance explains the intense interest from Chinese buyers. The chip delivers roughly six times more power than the H20 model released in late 2023.
Bernstein analysts noted that while Chinese companies can now produce AI chips better than the H20, domestic processors still lag behind the H200. The chip offers approximately 2-3 times the performance of China’s most advanced accelerators, according to White Oak Capital Partners.
Cloud service providers in China would likely welcome the opportunity to purchase H200 chips. This could present challenges for local AI chip adoption.
Some analysts question whether Chinese buyers will embrace the H200. The chip remains far less powerful than Nvidia’s newest models like Blackwell and Rubin.
Chinese officials have not yet approved H200 purchases. Emergency meetings took place Wednesday to discuss the matter and determine whether to allow shipments into the country.
One proposal under consideration would require buyers to pair each H200 purchase with a certain ratio of domestic chips. This approach could support China’s push to develop its local AI chip industry.
China continues pursuing greater self-sufficiency in chip production. This effort persists regardless of U.S. policy changes, analysts say.
Cloud service providers and enterprise customers are already placing large orders and lobbying the government to relax restrictions conditionally. Chinese AI demand exceeds what local production can supply.
Adding new H200 capacity presents challenges for Nvidia. The company is transitioning to Rubin production while competing with companies like Google for limited advanced chipmaking capacity from TSMC.


