Nvidia (NVDA) is considering increasing production of its H200 artificial intelligence processors after demand from Chinese tech giants surged well beyond current supply capacity, according to sources familiar with the matter.

The discussions come after U.S. President Donald Trump approved conditional H200 exports to China, imposing a 25% fee on each sale. The policy shift immediately triggered a wave of inquiries from major Chinese firms such as Alibaba and ByteDance, both eager to place large orders for Nvidia’s second-fastest AI chip.

However, any shipments remain on hold as Beijing has yet to authorize purchases. Chinese officials convened urgent meetings this week to assess whether to allow the chips into the country, given ongoing efforts to accelerate domestic AI semiconductor development.

H200 Supply Strained Amid Limited Production

Only small quantities of H200 chips are currently being produced as Nvidia prioritizes capacity for its next-generation Blackwell and upcoming Rubin architectures. Chinese customers—concerned about access—requested clarity from Nvidia, which provided preliminary guidance but no exact figures.

The H200, built on TSMC’s 4nm process, is the most advanced chip China can legally obtain. Its performance is roughly six times that of the H20, a downgraded and export-compliant variant.

China Balances Demand With Domestic Chip Goals

Beijing is wary of allowing unrestricted H200 imports that could slow the momentum of its local AI chip sector. During emergency meetings, officials discussed requiring buyers to pair each H200 purchase with a proportional order of domestic accelerators.

Despite this, Chinese cloud providers and enterprises are lobbying strongly to move forward, citing the massive performance gap. As one analyst noted, the H200 delivers 2–3x more compute power than China’s best locally made accelerators, and domestic production cannot meet current demand levels.

Capacity Constraints at TSMC Add Pressure

Even if Nvidia moves to scale up H200 output, increasing supply won’t be easy. Nvidia is battling for advanced manufacturing capacity at TSMC, where it competes with major clients including Google. The company must also manage its production shift toward Rubin in 2026.