Chinese artificial intelligence (AI) company DeepSeek is moving to reduce its dependence on Nvidia as the United States tightens semiconductor export controls to China.

On April 27, the South China Morning Post (SCMP) reported that DeepSeek has begun restructuring its supply chain by considering alternatives to Nvidia as it secures computing resources needed for training and inference.

The core issue is a shift in its chip procurement strategy. As the United States continues to restrict exports of high-performance AI semiconductors to China, Chinese AI companies have found it difficult to secure a stable supply of Nvidia chips. DeepSeek also appears to have entered a phase of reducing reliance on a specific supplier and looking for other options suited to running its models to avoid such constraints.

The change is not limited to simply replacing components. For AI companies, the choice of chips affects model training costs, inference efficiency and the speed of service expansion. DeepSeek's adjustments are closer to an infrastructure strategy aimed at sustaining future service operations while responding to a procurement environment narrowed by U.S. regulations.

Such moves around DeepSeek align with broader trends across China's AI industry. Chinese companies have used Nvidia products as much as possible even after U.S. regulations, but as restrictions tighten, securing replacement chips and in-house optimisation capabilities has become more important. The industry sees Nvidia as still having strengths in software ecosystems and performance, but in a regulatory environment, that advantage alone makes it difficult to resolve supply uncertainty.

In practice, the issue cannot be solved by simply switching chips. Training frameworks, server configurations and inference infrastructure must be reworked to fit new hardware. Performance degradation or development delays can occur in the process, but over the long term, reducing dependence on a specific U.S. supplier has emerged as an important task for Chinese companies.

DeepSeek's steps also show that Chinese AI companies are beginning to place procurement viability itself at the centre of their business strategies, not only technology competition. Nvidia chips have effectively become a standard in China's AI market, but after export controls were strengthened, the cost of maintaining that standard rose. Companies are now in a situation where they must recalculate the balance between model performance and costs within the limits of uninterrupted supply.

Amid this trend, market attention is focused on what alternatives DeepSeek will actually adopt. Another point to watch is whether introducing replacement chips will be applied only to some tasks in a limited way or expanded across training and inference. If software optimisation and data centre design also change together, it could have a significant impact on the infrastructure structure of China's AI industry.

Ultimately, DeepSeek's effort to move away from Nvidia is seen as an example showing that U.S. export controls are changing not only technology choices but also how Chinese AI companies operate their businesses. Going forward, an AI company's competitiveness in China is expected to hinge not only on what models it releases but also on what chips it can run them on.