China's AI Chip Industry Shifts Toward Custom ASICs as Export Controls Reshape Data Center Supply Chains

Key Takeaways

• Chinese technology companies are shifting investment toward custom AI ASICs as U.S. export controls continue restricting access to advanced Nvidia GPUs.
• China has formally incorporated domestic AI processors into its "secure and reliable" technology certification framework for the first time.
• The trend is reshaping demand across the broader semiconductor supply chain — from AI accelerators and memory to packaging, power delivery, thermal management, and interconnects.

Chinese technology companies are increasingly investing in custom artificial intelligence chips as U.S. export controls continue to restrict access to advanced processors from Nvidia, accelerating efforts to build a domestic AI computing ecosystem.

Rather than developing direct alternatives to Nvidia’s general-purpose graphics processing units (GPUs), many Chinese firms are focusing on application-specific integrated circuits (ASICs) optimized for targeted AI workloads such as inference, recommendation systems, cloud services, and video processing.

The shift reflects a broader strategy aimed at reducing reliance on foreign semiconductor technologies while improving performance and efficiency for specific AI applications.

Custom ASICs typically offer lower power consumption and higher workload-specific efficiency than general-purpose GPUs, although they lack the flexibility required for a wide range of AI training tasks. Industry analysts expect demand for AI inference hardware to grow as companies move from training large language models toward deploying them at scale, creating opportunities for specialized processors.

The move comes as Washington maintains restrictions on exports of advanced AI accelerators and related technologies to China. The U.S. Department of Commerce’s Bureau of Industry and Security has also clarified licensing requirements for certain advanced computing products supplied to entities headquartered in China or Macau, even when transactions involve overseas subsidiaries.

The restrictions have prompted Chinese technology companies to accelerate investment in domestic processors, software platforms, and computing infrastructure designed to reduce dependence on foreign suppliers.

In late May, Chinese authorities formally included AI training and inference chips within the country’s “secure and reliable” technology assessment framework for the first time. The certification program helps determine which products are eligible for procurement by government agencies, state-owned enterprises, and operators of critical infrastructure.

Nine domestically developed AI chips were included in the initial approved list, according to Chinese authorities. Approved products came from companies affiliated with Huawei’s HiSilicon, Alibaba’s T-Head Semiconductor, Biren Technology, Hygon Information Technology, Iluvatar CoreX, MetaX, and Moore Threads.

The certifications are valid for three years and provide a formal pathway for adoption across sectors including government, telecommunications, cloud computing, finance, and industrial infrastructure.

The development mirrors a broader global trend toward custom AI silicon. Major cloud providers and hyperscale data center operators are increasingly developing or deploying specialized AI accelerators to improve efficiency and reduce dependence on third-party GPU suppliers.

Research firm TrendForce forecasts global AI server shipments will grow more than 28% in 2026, with ASIC-based systems accounting for an increasing share of deployments as cloud providers expand custom hardware programs.

The growing adoption of custom AI processors is expected to affect demand across the broader semiconductor supply chain, including advanced packaging, substrates, high-bandwidth memory, power management devices, networking components, and thermal management systems.

Industry participants say the trend highlights how competition in AI infrastructure is expanding beyond processors themselves, creating new sourcing and capacity challenges across memory, power, interconnect, cooling, and server integration markets.

For semiconductor manufacturers, distributors, and procurement teams, the emergence of multiple AI hardware architectures could lead to more fragmented demand patterns and greater emphasis on supply-chain flexibility, regional compliance requirements, and supplier diversification.