China's AI self-reliance process: industrial restructuring from chips to large language models

Industry Context

In recent years, artificial intelligence has become the core battleground of U.S.-China technological competition. The release of ChatGPT at the end of 2022 pushed the global AI race to a new peak. The United States restricted China's access to advanced GPUs and manufacturing equipment through export controls, prompting Beijing to establish "self-reliance and controllability" as a national strategy. In April 2025, Chinese leaders explicitly called for achieving "self-reliance and self-strengthening" in AI technology during a Politburo group study session, marking China's shift from early international cooperation to comprehensive self-reliance.

A report released by MERICS (Mercator Institute for China Studies) in July 2025, *China's drive toward self-reliance in artificial intelligence*, systematically reviewed China's progress at various levels of the AI technology stack. The report noted that China is the first country to attempt to build a "national AI technology stack," and its experience offers reference value for Europe's construction of digital sovereignty.

Market Impact

China's AI self-reliance process is reshaping the global AI industry landscape. First, for U.S. chip companies such as NVIDIA, China's domestic substitution efforts mean that their long-term market share in China faces challenges. Although the performance of domestic chips such as Huawei's has not yet reached NVIDIA's top level, under geopolitical pressure, Chinese customers are being forced to accelerate the adoption of domestic solutions, forming a de facto "dual-track" market.

Second, China's large language model ecosystem is growing rapidly in a closed market. Local models such as DeepSeek have gained widespread attention domestically and internationally thanks to their open-source strategy and cost advantages. This has weakened the penetration of U.S. models in the Chinese market, while also providing global developers with alternative options.

For investors, the localization trend of China's AI industry chain has created new investment opportunities, especially in semiconductor equipment, domestic computing infrastructure, and the model application layer. However, policy uncertainty and technological bottlenecks remain major risks.

Competitive Landscape

At the bottom layer of the AI technology stack (chips), Huawei is the absolute leader. The report emphasizes that Huawei, together with domestic chip manufacturers, has still achieved mass production of AI chips under the sanctioned environment, although performance still lags behind NVIDIA. This layer receives the most national support because it is the foundation of the entire AI ecosystem.

In the middle layer (machine learning frameworks), Chinese internet giants such as Baidu, Tencent, and Alibaba dominate the local ecosystem. However, the report notes that users still prefer mainstream global frameworks such as TensorFlow and PyTorch, and the ecosystem maturity of domestic frameworks still needs improvement.At the top level (models and applications), competition in the Chinese market is exceptionally fierce. Numerous players including Baidu, Alibaba, ByteDance, and DeepSeek are vying for market share. DeepSeek's rise demonstrates China's deep engagement in the open-source community, with its models approaching top U.S. levels on multiple benchmarks. The beneficiaries are Chinese AI application companies, while those under pressure are U.S. model suppliers, especially in the Chinese market.

Enterprise Implications

For Chinese enterprises, self-reliance and controllability are no longer an option but a necessity for survival. Companies should actively evaluate domestic computing solutions such as Huawei Ascend and establish a "dual supply chain" strategy to hedge against geopolitical risks. Meanwhile, the rapid iteration of domestic large models offers enterprises an increasing number of application options, but attention must be paid to model performance, ecosystem maturity, and integration costs with existing systems.

For multinational enterprises, the regulatory environment and data policies of China's AI market require them to adopt localized solutions. The MERICS report suggests that European companies should monitor the progress of China's AI technology stack, as similar "digital sovereignty" requirements may emerge in the European market in the future.

In addition, when planning AI strategies, companies should incorporate supply chain resilience into their considerations and avoid over-reliance on the technology of a single country or company. Participation strategies in the open-source community also need to be reassessed, as China is increasing its contributions to global open-source projects, which may influence the direction of technical standards in the future.

Outlook

Within 12 months, China will accelerate the deployment of domestic chips in training and inference scenarios, with companies like Huawei striving to narrow the performance gap with NVIDIA. Meanwhile, competition at the model layer will intensify, leading players such as DeepSeek may receive more investment, and more vertical solutions will emerge at the application layer.

Within 24 months, China is expected to make significant progress in specialized AI chips and inference efficiency, but breakthroughs in high-end training chips will remain constrained by equipment limitations such as lithography machines. The fragmentation of the global open-source community may intensify, and China will more actively lead AI standards that align with its own needs.

Within 3 years, China's AI industry may form a relatively independent ecosystem, developing in parallel with the U.S. ecosystem. For Europe, it will be necessary to find a balance between China and the U.S., drawing on China's autonomy experience while avoiding the "de-globalization" trap. Corporate decision-makers should plan ahead and strike a dynamic balance among technological innovation, supply chain resilience, and geopolitical risks.