Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
The writer is professor of law at the Gould School of Law, University of Southern California, and author of ‘High Wire: How China Regulates Big Tech and Governs Its Economy’
Last week, the US further tightened its restrictions on semiconductor sales to China, triggering sharp declines in the stock prices of US chipmakers Nvidia and AMD. Washington seems determined to double down on export controls in its quest for supremacy in an AI arms race — no matter the cost to itself. But what if China isn’t trying to win?
American policymakers warn that whoever dominates AI will gain decisive economic, national security and military advantages. Beijing, however, may have quietly concluded that being a close second will serve its interests better.
The US dominates expensive, cutting edge AI models. But the rise of open-source, low cost AI models like China’s DeepSeek has already shifted the global industry’s demand for computing power from training AI models to using existing models to respond to user queries — a process known as inference. Barclays estimates that inference will account for up to 70 per cent of total AI compute demand by 2026.
That explains why the White House moved to restrict sales of Nvidia’s H20 chip, the most advanced chip that Chinese customers were able to purchase from Nvidia and one that is optimised for inference.
Yet the policy could end up helping the Chinese government’s push for technological self-sufficiency and at a lower cost. With Nvidia chips off the table, Chinese companies are likely to use domestic substitutes, particularly those offered by Huawei and Cambricon.
At a symposium between President Xi Jinping and Chinese tech entrepreneurs in February, Huawei founder Ren Zhengfei was seated directly opposite Xi — a symbolic gesture that signalled its importance.
It helps that inferencing poses lower technical barriers for Chinese hardware makers. According to research from DeepSeek, Huawei’s Ascend 910c chip already delivers roughly 60 per cent of the inference performance of Nvidia’s H100. In fact, DeepSeek is reportedly already running some of its inference workloads on Huawei’s newest chips.
The US chip ban extends beyond China. Towards the end of the Biden administration, the US introduced an “AI Diffusion Framework,” which categorises countries into three tiers and extends export restrictions to Tier 2 countries including India, Malaysia, Singapore and the United Arab Emirates. This is expected to be implemented in mid-May, unless the Trump administration withdraws the rules, which seems unlikely.
Tier 2 countries will not sit still. Many harbour ambitious plans to build their own world-class data centres. Faced with US chip restrictions, they may turn to China as an alternative source.
In its effort to stifle China’s progress, Washington may therefore be doing Beijing yet another favour by spurring both domestic and international demand for Chinese chips.
Demand is also a powerful driver of innovation. Huawei now leads a coalition of more than 2,000 companies, all working towards the goal of making China more than 70 per cent self-sufficient across the semiconductor supply chain by 2028. Just last week, it unveiled its CloudMatrix 384 AI supernode, which reportedly outperforms the computing power of Nvidia’s market leading NVL72 AI infrastructure system and is designed to ease congestion in the computing segment of AI data centres. The breakthrough could help China overcome a key bottleneck in building large-scale AI computing infrastructure.
China still lags behind the US in both AI hardware and software. It is likely to remain so for the foreseeable future. But this is no barrier to the country pushing ahead with its AI agenda. It may be perfectly content with remaining number two.
