China has formally brought AI chips into its national security and reliability evaluation framework for the first time, marking a further expansion of the country's Information Technology Application Innovation procurement policy into AI computing infrastructure.
The China Information Technology Security Evaluation Center and the China National Security Technology Evaluation Center, on May 26, released the "Announcement of Safety and Reliability Evaluation Results (No. 2, 2026)," listing nine China-designed AI training and inference chips that passed Level I security and reliability certification.
The approved chips include HiSilicon Ascend 310, Ascend 910, Alibaba T-Head Zhenwu M530 and M890, Biren Bilian 166, Hygon DCU-3G, Iluvatar CoreX KCC-V100X, MetaX MXC600 and Moore Threads PH100.
All certified products were developed by Chinese chip designers and received Level I qualification under the evaluation system.
According to the two evaluation agencies, the certification results will remain valid for three years.
A new frontier for AI procurement
The latest announcement marks the first time AI training and inference chips have been classified as a standalone category under China's security and reliability evaluation system.
The framework has in recent years become an increasingly important reference standard for procurement by Chinese government agencies, state-owned enterprises, and critical infrastructure operators. Market observers, therefore, view the certification as potentially significant for vendors seeking qualification advantages in public-sector and enterprise AI projects.
The move also signals that China's Information Technology Application Innovation policy, previously focused on CPUs, operating systems, and general-purpose servers, is extending further into AI computing hardware.
Diverse approaches, one direction
The first batch of certified products reflects multiple technology approaches within China's domestic AI chip ecosystem.
Huawei's Ascend series covers both AI training and inference workloads, while Hygon has established a dual-platform strategy combining CPUs and data center GPUs, or DCUs. GPU-focused companies, including Moore Threads, Biren, and MetaX, continue to push the deployment of domestic AI computing platforms.
All products approved in the current round received only Level I certification, with no higher-level AI chip certifications announced so far. The latest evaluation results nonetheless underscore China's broader efforts to establish long-term standards for supply chain autonomy, security mechanisms, and controllability across AI infrastructure, while gradually building a more complete domestic framework spanning chips, software, and procurement standards.
Article translated by Levi Li and edited by Jerry Chen
