DeepSeek plans to make its own AI chips as US export controls tighten

DeepSeek, the Chinese company whose AI models drew global attention for their efficiency, is planning to design its own chips to power future systems, according to a report described by Ars Technica. The stated goal is to reduce dependence on outside suppliers, specifically Nvidia and Huawei, at a time when US export controls have made access to the most advanced processors uncertain for Chinese firms.

The context is a years-long effort by Washington to limit China's access to cutting-edge semiconductors. US export controls restrict the sale of the most powerful AI chips, made largely by Nvidia, to Chinese customers, on the reasoning that the same processors that train commercial AI can also underpin military and surveillance capabilities. Those rules have repeatedly tightened, closing loopholes as they appear.

For a company like DeepSeek, that policy creates a supply problem. Training and running large AI models requires vast quantities of specialised chips, and the best of them have been the hardest to obtain. Building its own processors is a way to insulate itself from both export controls and reliance on a small number of suppliers, whether foreign or domestic.

The report is careful to note that the plan is early, and that qualifier deserves emphasis. Designing a competitive AI chip is one of the most difficult undertakings in modern technology, requiring specialised engineering talent, years of development and access to advanced manufacturing that itself is subject to export restrictions. An intention to build chips is a long way from having them in hand.

The manufacturing dimension is the deepest challenge. Even a company that designs an excellent chip must have it fabricated at a foundry capable of producing at the leading edge, and the most advanced fabrication is concentrated in a handful of facilities, primarily in Taiwan and South Korea, whose equipment and processes are themselves entangled in export controls. This is the bottleneck that makes chip self-sufficiency so hard to achieve quickly.

DeepSeek's move fits a broader pattern across the Chinese technology sector. Faced with restrictions, Chinese companies have accelerated efforts to develop domestic alternatives across the semiconductor supply chain, from chip design to manufacturing equipment. The strategy is one of long-term self-reliance, accepting near-term inefficiency in exchange for reduced vulnerability to external decisions.

The reference to reducing reliance on Huawei is notable, because Huawei has become China's most prominent domestic chip champion, developing AI processors positioned as alternatives to Nvidia's. DeepSeek's apparent desire to reduce dependence even on Huawei suggests a preference for controlling its own hardware destiny rather than swapping one supplier for another, and hints at the concentration risks within China's own market.

The episode illustrates a central dynamic of the current technology landscape: export controls intended to slow a rival can also spur it toward self-sufficiency. By restricting access to foreign chips, the controls raise the incentive for Chinese firms to build their own, which could, over a long enough horizon, reduce the leverage the controls were designed to provide. Analysts debate whether the net effect is to slow China down or to accelerate its independence.

For the global industry, DeepSeek's ambition is a data point in a larger fragmentation. The AI supply chain has been remarkably global, with design, manufacturing and materials spread across many countries. Efforts by multiple governments to secure domestic capacity, driven by both economic and security concerns, are pushing toward a more divided landscape in which capability is duplicated rather than shared.

Whether DeepSeek succeeds is an open question that will take years to answer. Designing a first chip is achievable for a well-resourced company; matching the performance of the industry leaders, and manufacturing at scale, is far harder. What the plan signals clearly, regardless of outcome, is how thoroughly the politics of technology now shapes the strategies of the companies building artificial intelligence.