China Accelerates AI Chip Independence With State-Backed Ecosystem

Chinese Firms Accelerate In-House AI Chip Designs
Gains Also Seen in Manufacturing and Packaging — Will a Self-Sufficient Ecosystem Emerge?
Massive State Investment Fuels Industry Growth

China is accelerating efforts to achieve self-sufficiency in artificial intelligence (AI) semiconductors. Beyond chip design, Chinese firms are also gaining ground in manufacturing and packaging, signaling the emergence of a homegrown industrial ecosystem. Analysts note that this progress reflects the payoff from more than a decade of sustained, large-scale government investment.

China Begins Using Self-Designed AI Chips

According to U.S. tech outlet The Information on the 11th, Chinese tech giants Alibaba and Baidu have started training AI models with semiconductors they designed themselves. Alibaba has reportedly been using its in-house “Zhenwu” chip since early this year to train smaller AI models, while Baidu is experimenting with its Kunlun P800 to train the latest version of its Ernie model. Following the “DeepSeek shock” earlier this year, China is now moving beyond AI model development to strengthen self-reliance in the semiconductors that power and train them.

Until recently, Chinese firms depended heavily on NVIDIA’s AI chips, constrained by limited domestic capabilities and U.S. export restrictions. They were forced to use the lower-performance H20 chips rather than the flagship B100 or H100. In April, the Trump administration further blocked H20 exports to China on national security grounds. Although Washington partially lifted the ban in July, China’s domestically produced chips have since advanced enough to compete with NVIDIA’s offerings.

Huawei, through its HiSilicon subsidiary, launched the Ascend 910 series, achieving up to 800 TFLOPS (FP16) performance—comparable to NVIDIA’s H100. Cambricon has positioned itself as a general supplier, leading optimization efforts with FP8-based low-precision computing. GPU startup Birentech is targeting the high-performance market with its BR100 series, while Shanghai-based MetaX unveiled its C600 chip, boasting greater memory capacity than NVIDIA’s H20.

These developments highlight China’s accelerating push to establish an independent AI semiconductor ecosystem, reducing reliance on U.S. technology and reshaping global competition in the field.

AI Semiconductor Independence Becomes Reality

What stands out is that China is no longer focused solely on chip design but is advancing across the entire AI semiconductor ecosystem.

In manufacturing, the lead player is SMIC (Semiconductor Manufacturing International Corp.), which has successfully produced Huawei’s Ascend 910B and Kirin 9000S chips using a 7nm process. Remarkably, SMIC achieved this without access to cutting-edge EUV lithography from ASML, relying instead on advanced DUV multi-patterning techniques to fabricate AI chips.

In the packaging and testing (OSAT) segment, Chinese firms are also making significant progress. Companies such as JCET, TFME, and Huatian Technology now rank among the world’s top ten in capability. They have developed 2.5D/3D stacking and TSV-based 3D packaging, reaching the point where Huawei’s Ascend chips can be integrated through dual-chip packaging without outside assistance.

Progress is also evident in lithography equipment development. Last month, China unveiled its domestically built “Xizhi” electron-beam lithography machine, capable of etching circuits at the 8nm level with positional accuracy of 0.6nm, meeting international standards. However, its throughput remains slower than that of conventional DUV or EUV tools, making it less practical for mass production. Huawei is therefore pushing forward with its own EUV lithography project, aiming for mass production capability by 2026.

Taken together, these advances signal that China is steadily transforming AI semiconductor independence from a long-term ambition into an emerging reality.

State Funding Pushes China’s AI Chip Drive

China’s ability to build competitiveness across the AI semiconductor ecosystem is rooted in massive government investment. The effort began in 2014 with the launch of the “National IC Industry Development Guidelines” and the establishment of the National Integrated Circuit Industry Investment Fund (ICF, or “Big Fund”). The first-phase fund injected 138.7 billion yuan (about $28 billion), followed by a second phase in 2019 worth 204.1 billion yuan (about $41.2 billion), and a third phase last year totaling 344 billion yuan (about $69.5 billion).

Local governments have followed suit with their own initiatives. In March, Shanghai created a 530 million yuan ($104 million) third-phase semiconductor fund, bringing the city’s total across three phases to 40 billion yuan ($7.8 billion). In May, Shenzhen launched a 5 billion yuan ($980 million) semiconductor fund, co-financed by the city and Longgang District, and managed by the state-owned Shenzhen Capital Group.

At the same time, Beijing is reinforcing domestic dominance through procurement and preference policies. Both state-owned enterprises and private tech giants such as ByteDance (the parent of TikTok) are being encouraged to adopt China-made chips over foreign alternatives. The strategy is already paying off: China Mobile, the state telecom giant, will source its entire 19.1 billion yuan ($3.7 billion) worth of AI server chips for 2024–2025 from Huawei.

This sustained flow of public and local capital—coupled with preferential demand policies—is rapidly reshaping China’s semiconductor landscape, displacing foreign suppliers like NVIDIA and accelerating the rise of a self-sufficient AI chip ecosystem.