Export restrictions have become a central part of the technology rivalry between the United States and China, as both countries seek to promote their own approaches to AI models and chips. Washington has focused on controlling access to advanced hardware, while Beijing is preparing rules that could limit exports of AI software and technologies.
US restrictions on AI chip exports, introduced in late 2022 and strengthened under the Trump administration, have supported American companies’ lead in developing frontier models. Nvidia’s position in the chip market has been an important part of that advantage.
But DeepSeek’s arrival has highlighted how Chinese startups are narrowing the gap with open-source models. The restrictions could become less effective if Chinese developers adopt chips being developed by DeepSeek and Huawei, designed to better suit the needs of China’s AI models. The companies are also working on adapting Huawei’s CANN software as an alternative to Nvidia’s CUDA platform.
If those hardware and software efforts succeed, Washington’s chip controls may lose influence as the two countries’ technology ecosystems develop more independently. China’s broader push is to reduce its reliance on foreign components across the physical infrastructure that underpins AI.
The contest is also shifting toward rules governing AI models. Beijing is drafting export restrictions covering as many as 23 technologies, including models, agents and chip-design technology. The proposal is intended to protect Chinese AI capabilities and keep training data for the country’s most advanced models under regulatory supervision. It remains a draft and would not apply to models already released; access to those models would be limited to cloud-based use.
The proposed Chinese measures come as the Trump administration has tightened restrictions on access to DeepSeek models, extending them to civilian agencies in the federal executive branch and requiring the removal of third-party software developed using DeepSeek technology.
Energy is another factor in the competition. China’s limited access to Nvidia’s most advanced chips means it may need to use large numbers of less powerful processors, which consume more electricity. Industry estimates cited in the report put energy at as much as 70% of data centers’ operating costs, adding to the importance of China’s efforts to build data-center infrastructure using relatively inexpensive electricity.
The article also identifies China’s dominance in rare-earth refining as a vulnerability for the global chip industry and a focus for Washington. US efforts to strengthen supply-chain security include work with TSMC to find alternatives. Taiwanese officials have expressed optimism that a pilot production line not dependent on rare earths could be available within three years.
The rivalry is increasingly about control of the full AI value chain and the standards that govern it. The US is concentrating on the physical infrastructure, while China is seeking greater control over exports of AI software and technology. Unless new competitors emerge, restrictions imposed by both sides are likely to shape how AI develops globally.
Sources: El País Economía.