Chinese artificial intelligence firms are increasingly adopting domestically produced chips as U.S. export restrictions limit their access to advanced American technology. Last month, Chinese start-up DeepSeek announced that its latest AI model had been optimized to run on chips made by Huawei, marking a notable step in China’s efforts to reduce reliance on U.S.-supplied semiconductors. While most leading AI systems globally continue to use Nvidia technology, Beijing’s push for technological self-sufficiency has accelerated amid ongoing trade tensions.

DeepSeek’s announcement, which came just before a scheduled summit between U.S. President Donald Trump and Chinese President Xi Jinping, highlights Beijing’s confidence that American controls on Nvidia’s chips have not halted its AI development. U.S. export controls, aimed at restricting China’s access to high-performance AI chips, have instead compelled Chinese companies to adapt, designing systems compatible with local processors rather than waiting for restrictions to ease.

The new model by DeepSeek uses Huawei chips for inference—the stage where AI systems process input and generate responses—while still relying on Nvidia hardware for the more intensive training phase, according to semiconductor industry sources who spoke anonymously. Details remain unclear on how DeepSeek obtained Nvidia chips for training; some Chinese firms remotely access Nvidia’s hardware located in overseas data centers. Huawei has stated plans to release a training-capable chip this year but acknowledged it may take an additional year for its products to match Nvidia’s performance.

The growing divergence in AI hardware between China and the United States has been anticipated by Nvidia CEO Jensen Huang, who has warned that stringent U.S. export controls could lead to a bifurcated market. Under this scenario, Chinese AI systems would operate on domestically produced chips, while Western systems would continue to use American technology. Huang has also cautioned that these restrictions may ultimately weaken U.S. influence over China’s AI sector.

Despite a temporary easing last year when Trump authorized Nvidia to sell its H200 chip to China, regulatory scrutiny in Washington and Beijing’s directive for companies to prioritize local chips have limited actual shipments. Commerce Secretary Howard Lutnick told a Senate committee last month that no H200 chips had yet been delivered to China, and Nvidia’s filings confirm no revenue from H200 sales there to date. The availability of Nvidia chips remains a contentious issue likely to surface during ongoing U.S.-China trade discussions.

While Beijing continues to express frustration over U.S. export controls, some analysts suggest that China is also keen to foster domestic innovation to lessen dependence on foreign technologies. However, significant challenges remain for Chinese semiconductor manufacturers. For instance, Semiconductor Manufacturing International Corporation (SMIC), which produces some Huawei chips, faces difficulties scaling production with chips that are less efficient and more prone to defects compared to foreign counterparts.

To compensate, Huawei has employed a strategy of combining large numbers of these less-powerful chips to approximate the performance of more advanced processors, contingent on SMIC’s ability to meet volume demands. Chinese chip production is expected to represent only a small fraction of the global output of sophisticated semiconductors this year, particularly compared to companies like Nvidia and Taiwan Semiconductor Manufacturing Company, which formerly produced many Huawei chips before being affected by export controls.

The restrictions have not only limited China’s chip imports but also spurred a shift in AI development philosophy among Chinese companies. Rather than focusing solely on accumulating vast computing power, firms like Huawei and DeepSeek aim to build integrated systems of hardware and AI models that deliver adequate performance for most practical applications. This approach allows close collaboration between chip makers and AI developers to tailor hardware to software needs, potentially offsetting some of the limitations imposed by export controls.

DeepSeek’s public disclosure of specific enhancements requested for chip performance reflects this cooperative model development. Analysts note that such targeted optimization efforts point to China’s evolving strategy of building a self-sufficient AI ecosystem despite international setbacks. As trade negotiations continue, the trajectory of China’s AI ambitions remains a critical point of focus for both countries.