Nvidia is rushing to meet strong demand for its H200 AI chips from Chinese tech companies and has approached Taiwan Semiconductor Manufacturing Co (TSMC) to increase production, according to sources. Chinese firms have reportedly placed orders exceeding 2 million H200 chips for 2026, while Nvidia currently has just 700,000 units in stock. The precise number of additional chips Nvidia plans to order from TSMC is unclear, though production is expected to begin in the second quarter of 2026.

This surge in demand raises concerns about potential tightening in global AI chip supplies, as Nvidia must balance Chinese orders with limited availability elsewhere. Risks also remain because Beijing has not yet approved H200 shipments, though U.S. export restrictions were recently eased. Nvidia has set prices for Chinese clients at roughly $27,000 per chip, with variations depending on volume and arrangements.

The H200, part of Nvidia’s older Hopper architecture and built on TSMC’s 4-nanometer process, includes both standalone H200 chips and GH200 superchips combining the Grace CPU with Hopper GPUs. Initial orders are expected to be fulfilled from existing stock, with deliveries planned before the Lunar New Year.

Most of the Chinese demand comes from major internet companies seeking a significant performance boost over previous chips. Eight-chip modules are priced at around 1.5 million yuan, offering better value than the discontinued H20 module and grey-market alternatives. For example, ByteDance could spend roughly 100 billion yuan on Nvidia chips in 2026 if sales are approved. Regulatory uncertainty remains, as Chinese authorities weigh allowing H200 imports while promoting domestic AI chip development. One potential condition under consideration would require bundling H200 purchases with a portion of domestically produced chips.