Chinese firms acquired significant amounts of AI chips, worth billions of dollars, prior to the implementation of the US embargo.
Article:
China's AI sector is gearing up for a significant shift following reports of U.S. export restrictions on Nvidia chips. By the end of May 2025, Chinese tech giants are said to have requested millions of Nvidia's H20 chips, with ByteDance being one of the most active buyers.
However, the U.S. administration under President Donald Trump prohibited the export of these chips without a license since early April, causing the actual delivery of graphics processors to be lower than requested. Analysts estimate that the value of these orders exceeds $12 billion, with Chinese companies receiving graphics processors worth "several billion dollars" before the restrictions came into effect.
The market response to these restrictions has been immediate, with increased demand for domestic alternatives. Huawei and Cambricon are seeing a surge in orders from cloud providers like Alibaba and Tencent, as well as AI startups seeking replacements for Nvidia's top-tier AI chips. This temporary demand has boosted the margins for domestic chip manufacturers.
In the long run, these restrictions have prompted strategic shifts. Huawei is prioritizing the development of its Ascend 910B, a chip comparable to Nvidia’s A100, and next-gen chips using in-house Da Vinci architecture. Cambricon, on the other hand, is focusing on its 3rd-gen MLU370 and edge-AI chips for applications such as autonomous driving and smart cities.
The U.S. restrictions have also led to increased government support. China's "High-Tech Import Substitution" policy offers subsidies, tax breaks, and procurement mandates, reducing reliance on foreign technology. Chinese chip manufacturers are also partnering with domestic foundries like SMIC and Yangtze Memory to fast-track advanced packaging technology and bypass lithography limitations.
However, the Chinese chip sector faces challenges. The software stacks (such as CUDA alternatives) are not yet mature, limiting developer adoption. Advanced node production still lags TSMC and Samsung due to EUV restrictions, and U.S. sanctions on SMIC and Primarius could potentially cripple production.
Nvidia has also implemented workarounds, such as offering specialized China-market chips with lower performance, and providing custom cloud solutions through partnerships. Chinese companies are also attempting to mimic Nvidia's CUDA ecosystem by integrating their chips with indigenous AI frameworks and CPUs.
Overall, while the U.S. restrictions initially strain China's AI infrastructure, they act as a catalyst for domestic innovation. Huawei and Cambricon now have unprecedented access to capital and policy support but must solve critical software/hardware gaps to truly rival Nvidia. The next 2-3 years will determine whether China’s chip sector can pivot from “substitution” to “leadership” in AI hardware.
For more news on this topic, follow us at @expert_mag #China #Artificial Intelligence #USA.
- Alibaba, being a cloud provider, has experienced a surge in orders for domestic alternative chips due to the U.S. export restrictions on Nvidia's H20 chips.
- The Chinese tech giant, ByteDance, is one of the most active buyers of Nvidia's H20 chips, with the value of these orders estimated to exceed $12 billion.
- In response to the U.S. export restrictions, Chinese companies like Huawei and Cambricon are focusing on developing their next-gen chips, such as Huawei's Ascend 910B and Cambricon's 3rd-gen MLU370, to rival Nvidia's technology in artificial intelligence.
- The Chinese government's "High-Tech Import Substitution" policy offers support to domestic chip manufacturers, including subsidies, tax breaks, and procurement mandates, aiming to reduce reliance on foreign technology.
