American company Nvidia, known for its powerful AI chips, is trying to meet the enormous demand from Chinese companies for its H200 model. According to information from late December 2025, Chinese technology companies ordered more than 2 million of these chips for 2026. Nvidia currently has only 700,000 units in stock, far fewer than needed. Nvidia has therefore approached Taiwanese company TSMC, which manufactures the chips, to expand production. Work on the new orders is expected to begin in the second quarter of 2026.
This demand comes at a time when U.S. policy is changing. In December 2025, President Donald Trump decided that sales of H200 chips to China would be permitted, but subject to a 25% fee. This is a major change from the previous administration of Joe Biden, which banned such exports due to national security concerns. Nvidia plans to send the first shipments from its existing inventory to China before the Lunar New Year in mid-February 2026. These shipments are expected to include 5,000 to 10,000 modules, equivalent to approximately 40,000 to 80,000 individual H200 chips.
Why is the H200 in such high demand?
The H200 belongs to Nvidia's older Hopper series, but it remains very popular for artificial intelligence workloads. It has been superseded by newer models such as Blackwell and the upcoming Rubin, but supplies of these newer chips are limited. In China, the H200 is attractive because it is approximately six times more powerful than the H20 model, which Nvidia designed specifically for the Chinese market but which was later blocked. Chinese companies such as Alibaba Group and ByteDance see the H200 as an opportunity to significantly improve their models.
Nvidia's inventory includes about 100,000 units of the GH200 Grace Hopper variant, which combines a Grace CPU with a Hopper GPU, while the rest consists of standalone H200 chips. Nvidia has already indicated prices to Chinese clients: around $27,000 for a standalone chip (approximately CZK 621,000), while a module with eight chips is expected to cost around 1.5 million yuan (about CZK 4,800,000). That is slightly more than the previous price of an H20 module, which sold for 1.2 million yuan (approximately CZK 3,840,000), but Chinese companies consider it worthwhile due to the higher performance. Moreover, it is cheaper than on the gray market, where similar modules sell for more than 1.75 million yuan (around CZK 5,600,000).
Regulation and domestic competition
Everything depends on approval from the Chinese government. Beijing has not yet approved any H200 purchases, and the timeline could be delayed. Chinese officials are concerned that importing advanced chips will slow the development of domestic alternatives. Chinese companies have already developed chips comparable to the H20, but nothing that matches the H200. One proposed condition is that every H200 purchase would have to be tied to a certain quantity of domestically produced chips.
Nvidia said in a statement that it manages its supply chain so that sales to China do not affect deliveries to the United States. Company spokespeople emphasized that China is a competitive market with rapidly growing local chip suppliers, and that a complete export ban would only benefit foreign competitors.
Plans of major players in China
The main orders are coming from large internet companies in China. For example, ByteDance plans to spend about 100 billion yuan (approximately CZK 320 billion) on Nvidia chips in 2026, up from 85 billion yuan (around CZK 272 billion) in 2025—assuming, of course, that the sales are approved. These companies see the H200 as key to accelerating the development of AI applications.
Nvidia now has to manage the risks: while trying to meet Chinese demand, it must ensure that this does not affect chip availability elsewhere in the world. Talks with TSMC about new orders suggest that Nvidia is taking the situation seriously and preparing a longer-term solution.
Source: reuters.com
