US Sanctions Halt Chinese AI Giant DeepSeek

Chinese company DeepSeek, which has joined the ranks of the world’s leading artificial intelligence developers, is facing serious problems. The development of its latest model, DeepSeek-R2, has been significantly delayed due to US export restrictions on Nvidia H20 graphics processing units. These measures, which took effect in April 2025, require Nvidia to obtain a special license to sell these chips to China.

DeepSeek is not the first Chinese company to feel the impact of these restrictions. To train its previous R1 model, the company used 50,000 Nvidia Hopper graphics processing units, 30,000 of which were H20 chips. These processors were originally designed specifically for the Chinese market to comply with earlier export controls, but even that ultimately did not protect them from stricter measures.

Chip Shortage Paralyzes the Entire Sector

The situation is further complicated by the fact that many Chinese artificial intelligence companies run their models on cloud infrastructure powered by these H20 processors. The current restrictions have exhausted available capacity, preventing companies from scaling or training new large-scale models.

Although Chinese companies purchased billions of dollars’ worth of H20 graphics cards before the ban, these inventories have now largely been used up. Domestic alternatives, such as the Huawei Ascend 910B, exist, but they are less powerful and lack compatibility with Nvidia’s CUDA software, which is essential for most Chinese artificial intelligence development. Export restrictions therefore pose a serious obstacle to the further development of China’s AI sector.