Nvidia has approached Taiwan Semiconductor Manufacturing Co to increase production of its H200 artificial intelligence chips as demand from Chinese technology companies rises sharply, according to people familiar with the discussions.

Chinese firms have placed orders for more than 2 million H200 chips for delivery in 2026, while Nvidia currently has about 700,000 units available, two of the sources said. The exact volume of any additional order has not been finalized.

One source said Nvidia has asked TSMC to begin preparations for expanded production, with manufacturing expected to start in the second quarter of 2026.

The talks raise concerns about further tightening in global AI chip supplies, as Nvidia balances strong Chinese demand with constrained availability in other markets. They also highlight regulatory risks, as Beijing has not yet approved imports of the H200 chips, despite recent U.S. authorization allowing limited exports.

The discussions with TSMC and details of the Chinese orders have not been previously reported. NVIDIA has decided which H200 variants it will offer to Chinese customers and has priced the chips at about $27,000 each, the sources said.

Nvidia said it continuously manages its supply chain and that licensed H200 sales to authorized Chinese customers would not affect its ability to supply clients in the United States.

TSMC declined to comment. China’s Ministry of Industry and Information Technology did not immediately respond to a request for comment.

If finalized, the additional order would mark a significant expansion of production for the H200, part of Nvidia’s Hopper architecture and manufactured using TSMC’s 4-nanometer process. Nvidia has recently focused on ramping up output of its newer Blackwell chips and its upcoming Rubin platform.

Nvidia plans to fulfill initial H200 orders from existing inventory, with the first shipments expected to arrive before the Lunar New Year holiday in mid-February.

Most of the Chinese demand comes from major internet companies that view the H200 as a substantial upgrade over chips currently available to them, the sources said.

Of Nvidia’s current inventory, about 100,000 units are GH200 Grace Hopper superchips, which combine Nvidia’s Grace CPU with its Hopper-based GPU, while the remainder are standalone H200 processors. Both versions will be offered to Chinese customers.

An eight-chip H200 module is expected to cost about 1.5 million yuan, slightly higher than the now-unavailable H20 module, which previously sold for about 1.2 million yuan. However, sources said Chinese buyers view the pricing as attractive given the H200’s higher performance and a roughly 15% discount compared with grey-market alternatives.

The planned shipments follow a recent U.S. decision to allow H200 exports to China subject to a 25% fee, reversing earlier restrictions. Chinese authorities are still assessing whether to approve imports amid concerns that reliance on advanced foreign chips could slow development of the domestic semiconductor industry.