China is striving to achieve comprehensive self-reliance in Artificial Intelligence (AI), positioning the technology as crucial to its national and economic security. Against a backdrop of escalating geopolitical tensions and U.S.-led export control restrictions, Beijing has refocused its AI policy toward building an “independent and controllable” ecosystem encompassing AI chips, machine learning software, and applications such as large language models (LLMs). This ambition marks a significant shift away from previous strategies that emphasized international cooperation.
The Chinese government employs a layered approach to AI development. The lowest level—the semiconductor industry—receives the most intensive state support via funding and policy frameworks, such as the multi-phase “Big Fund” initiative, which aims to bolster domestic production. Huawei has emerged as the de facto leader in AI chip design, collaborating closely with China’s top foundry, SMIC. While these chips are improving in performance, they still trail behind U.S. designs in key aspects like efficiency and ecosystem maturity.
One major bottleneck is advanced manufacturing. Most Chinese chip designers rely on SMIC’s limited 7nm fabrication capabilities, leading to high competition for capacity. Compounding this issue is China’s restricted access to high-bandwidth memory and advanced chip packaging. These are essential for training and deploying cutting-edge AI models. Although Huawei’s hardware, such as its Ascend series, boasts competitive specifications, software immaturity—particularly issues with its CUDA alternative, CANN—significantly hampers real-world performance.
In the middle layer of the AI stack, Chinese firms have rolled out domestic alternatives to globally dominant machine learning frameworks like TensorFlow and PyTorch. Baidu’s PaddlePaddle and Huawei’s MindSpore are notable efforts. However, adoption lags behind Western frameworks, with most developers continuing to use global open-source tools, though contributions to these ecosystems are growing.
At the application level, LLM development has become a dynamic and fast-growing sector in China. The release of proprietary, high-performing models like DeepSeek-R1 has placed China on the global map of frontier AI development. Protected from direct competition with leading Western LLMs due to government restrictions, Chinese developers have thrived within a vibrant open-source ecosystem. Local governments are further incentivizing innovation through compute subsidies and industrial support.
However, despite marked progress, significant vulnerabilities remain. U.S. export controls continue to restrict Chinese access to high-performance chips and manufacturing equipment. While workarounds, such as accessing cloud services or mixed-chip training techniques, have temporarily offset these constraints, the future availability of advanced computing resources is uncertain.
China’s AI ambitions also rely heavily on domestic talent, data, capital, and infrastructure. The country has successfully cultivated a large pool of AI researchers and engineers, many trained domestically. Public and private investments in AI totaled over $7.3 billion in 2024, despite broader economic slowdowns. In terms of data, the government is consolidating its fragmented data environment through initiatives led by the National Data Administration, alongside policies to scale up the data labeling industry.
Computing infrastructure projects, part of the ‘East Data, West Computing’ initiative, aim to create a network of data centers powered by renewable energy. While capable of supporting AI development, inefficiencies remain, with many computing resources underutilized.
Externally, China’s AI trajectory is highly sensitive to international dynamics. U.S. technology export restrictions, particularly under the Trump administration’s recent tightening, are reshaping the landscape. While China adapts by leaning on domestic capacity and exploring cooperation with other countries, further curbs could limit progress.
For Europe, China’s evolving AI ecosystem poses both challenges and lessons. The continent will need to weigh the implications of integrating Chinese-origin AI technologies, particularly regarding compliance with regulations such as the GDPR and the upcoming AI Act. Meanwhile, Europe’s own “EuroStack” ambitions signal a growing desire to achieve digital sovereignty.
China’s path—though state-driven—is not free of risk. Centralized planning and subsidy allocation often lack efficiency, and mounting censorship and surveillance pressures threaten to undermine open collaboration, a critical ingredient for innovation.
In conclusion, while China has made impressive strides in climbing the AI technology stack, the interplay between internal capabilities and external constraints—particularly access to advanced chips and open-source collaboration—will determine its ultimate success. The nation’s pivot toward specialized AI applications may offer a more sustainable pathway forward, aligning with its core strengths in rapid commercialization and large-scale deployment. For global stakeholders including Europe, understanding China’s AI rise is essential for shaping strategic responses and ensuring technological competitiveness in a rapidly bifurcating world.
Source: https:// – Courtesy of the original publisher.
