Source: Tencent Technology
By Su Yang
Edited by Xu Qingyang
On December 24, market sources reported that$NVIDIA (NVDA.US)$NVIDIA has informed its Chinese customers of plans to deliver the H200 chips by mid-February 2026. The total shipment volume is expected to be 5,000–10,000 modules, equivalent to approximately 40,000 to 80,000 H200 chips.
In response, NVIDIA stated, “We are continuously managing our supply chain. Selling H200 to authorized customers in China will not affect our ability to supply global customers.”
According to NVIDIA’s statements, the plan to sell H200 in the Chinese market has been largely confirmed.
Meanwhile, a source familiar with the matter revealed that NVIDIA CEO Jensen Huang is also scheduled to visit China in January 2026. By comparison, throughout 2025, Huang will visit mainland China three times, with his annual trip during the Spring Festival to attend year-end events in Beijing, Shanghai, and Shenzhen, as well as supplier appreciation meetings, having become a regular occurrence.
“The unit price for the H200 8-GPU module provided to channels by NVIDIA is RMB 1.4 million, slightly higher than the H20,” an unnamed industry insider told Tencent Technology.
NVIDIA’s early version of the H20 featured 96GB of HBM3 memory, while an upgraded version with 141GB of HBM3e memory was launched earlier this year. The price of the 8-GPU module increased from RMB 880,000 to RMB 1.05 million at one point.
Wu Zihao, CEO of Ronghe Semiconductor Consulting, revealed that based on performance density (TPP), the computing power value of the new ‘special supply’ product H200 is 15,832, approximately 6.7 times that of the H20.
The physical layer comparison of TPP performance density is more objective, while TFLOPS floating-point operations often lead to comparison errors due to different calculation methods used by manufacturers.
With a performance difference of more than six times and a price difference of 1.3 times, from this perspective, the cost-performance ratio and attractiveness of H200 will significantly increase.
Precisely because of this, as a product with performance exceeding regulations (the new rule of October 17, 2023), H200 cannot go through the normal export process but requires special export licensing, from which the U.S. government will obtain 25% of the revenue.
According to market disclosures, the first batch of orders for H200 is expected to be between 40,000 and 80,000 chips, corresponding to 5,000–10,000 8-GPU modules, with a total value estimated at RMB 7 billion to RMB 14 billion.
If NVIDIA delivers 1 million H200 chips to China in 2026, equivalent to 125,000 H200 8-GPU modules, the corresponding sales revenue will reach RMB 175 billion (approximately USD 25 billion), of which the U.S. government will collect over USD 6 billion in revenue.
Previously, Tencent Technology emphasized in 'Three Key Questions on NVIDIA’s H200 Being Approved for Export to China' that relevant matters require procedural handling.
“It doesn’t mean sales can start immediately after saying it’s okay to sell; going through normal regulatory procedures also takes time, and directly approving licenses would not comply with export control regulations,” emphasized a researcher who has long been following technology policies.
Given that there has been much groundwork laid for the export of H200, it is possible that NVIDIA has already completed the preliminary procedural work.
“Application materials can be prepared in advance, and the U.S. can approve very quickly, even within one or two days,” said Han Lijie, a partner at Katten Law Firm.
Han Lijie emphasized that it is possible NVIDIA had already submitted the application systematically, but it remains in a Pending status.
Although the performance of the H200 has significantly improved compared to previous products, it is still not the most advanced product under NVIDIA. Therefore, there are industry rumors suggesting that it could be a way to 'clear inventory' for customers in mainland China while maintaining a generational lead over local Chinese products using older equipment.
This can be compared to the '1017 New Rules' and '1007 New Rules' introduced by the U.S. Department of Commerce’s Bureau of Industry and Security (BIS) in 2023 and 2022, respectively, which restricted China's semiconductor industry through various regulations.
If the aforementioned regulations were also adjusted and domestic chips could be designed normally and$Taiwan Semiconductor (TSM.US)$manufacture products with specifications equivalent to the H200, it would signify that the U.S. had completely shifted its approach to 'controlling' China's chip and even artificial intelligence industries.
However, in reality, the BIS has not made any adjustments to the aforementioned regulations, and the sale of the H200 still requires special permission. To put it more bluntly, I may allow you to purchase the H200, but I will not permit you to produce a product with equivalent specifications.
"The approval of the H200 represents a clear strategic signal: relying on external parties to provide 'suboptimal solutions' will never achieve true autonomy," said Wu Zihao.
Editor/Rocky