In the first four months of 2026, China’s chip exports reached US$103.5 billion, a year-on-year surge of 83.7%, but the export volume fell by 20.5%.
Behind this “shrinking and increasing value” data is the structural transition of China’s chip industry from low-end volume to high-end computing power.
The technological breakthroughs and national-level safety certification of domestic AI chips have become the core engine for shaping new export competitiveness.
On May 26, 2026, the China Information Security Evaluation Center and the National Security Technology Evaluation Center jointly issued an announcement to include the “artificial intelligence training inference chip” into the national security and reliability evaluation system for the first time.
All nine domestic AI chips have been rated as Level I, the highest security level, and are valid for three years.
This list covers the first-tier domestic companies, including Huawei Shengteng 310/910, Alibaba Pingtou Ge Zhenwu M530/M890, Biren Technology Bili™ 166, etc.
This move marks the official inclusion of domestic AI computing power infrastructure into the national Xinchuang security certification system, and its security and controllability have been elevated to a strategic level as important as databases and operating systems.
Certification provides authoritative endorsement for chips to enter Xinchuang Procurement in key fields such as finance, government affairs, and energy, and also adds hard-core “national team” credibility for domestic chips to go overseas.
The assessment adopts the method of voluntary application by enterprises and is conducted in batches. Enterprises not included in the first batch of lists, such as Cambrian and Suiyuan Technology, still have opportunities.
Behind the policy recognition are the real technological breakthroughs made by leading companies.
Domestic chips are building a full-stack independent computing ecosystem from the cloud to the car.
In terms of cloud computing power, Huawei’s Ascend series based on its self-developed DaVinci architecture has achieved overperformance.
The single-card inference performance of Ascend 950PR is 2.87 times that of NVIDIA’s special version H20 for China, and the purchase price is only 1/3 to 1/4 of it.
Vimicro’s XPU multi-core heterogeneous architecture supports device-edge-cloud collaboration, and its “Star Intelligence” based on the “Starlight Intelligent No. 5” chip supports 8 joint deployments to run the 671 billion parameter DeepSeek large model.
Muxi’s revenue in 2025 will increase by 121% year-on-year, and its full-process localized high-end GPU has completed more than 1,000 model deployments in more than 20 industries such as finance and medical care.
In terms of car-grade chips, BYD released China’s first mass-produced 4nm car-grade smart driving chip “Xuanji A3”.
The chip uses a car-grade 4nm process. The total computing power of the vehicle exceeds 2100TOPS and has passed ASIL-D certification, the highest level of automotive functional safety.
BYD Chairman Wang Chuanfu pointed out that the technical difficulty of automotive-grade 4nm is “equal to the 2nm level in the consumer electronics field.”
In terms of cutting-edge exploration, the technical routes are diversified.
Shanghai has launched the country’s first key laboratory for optical computing, co-constructed by Shanghai Jiao Tong University and Xizhi Technology.
Optical computing chips can achieve chip performance equivalent to less than 7 nanometers using a 40-nanometer process, which is expected to break through the power consumption bottleneck of traditional electrical computing.
At the same time, the DSA field computing laboratory focusing on RISC-V and dedicated computing architecture innovation was also opened in Zhangjiang, Shanghai.
The structural changes in chip exports are essentially the spillover of the competitiveness of China’s computing power industry, and its motivation comes from three levels.
Mature processes have become the “stabilizer” of the global supply chain.
Against the background of restrictions on advanced processes, demand for mature process chips of 28 nanometers and above has surged in fields such as automotive electronics and industrial control. These fields place greater emphasis on supply stability and cost advantages.
China has accumulated profound experience in manufacturing, packaging and testing in this field. SMIC and Hua Hong Semiconductor have maintained high production capacity utilization and accepted more international orders.
The whole machine goes overseas to drive the “package output” of chips.
New energy vehicles, servers and other complete products are going overseas, and the self-developed chips are embedded into the overall solutions and output simultaneously.
For example, the overseas expansion of BYD’s new energy vehicles will inevitably lead to the spillover of its “Xuanji A3” and other chips.
Breakthroughs in supporting industrial chain support high-end manufacturing.
The improvement of chip export competitiveness cannot be separated from the guarantee of upstream materials.
Taking electronic special gas as an example, the leading company CSSC Special Gas has the world’s largest tungsten hexafluoride production capacity. Its average export price rose from US$68.75/kg in January 2026 to US$149.79/kg in April, and its gross profit margin continued to improve.
From policy certification to technological breakthroughs to export market verification, the domestic chip industry is completing a profound transformation.
The core of this transformation is computing power – it is not only the basis of productivity in the AI era, but also becoming China’s new business card in shaping its foreign trade competitiveness.
With the improvement of Xinchuang ecology and the iteration of computing power technology, domestic chips are expected to move from “running parallel” to “leading” in some fields in the global market by virtue of the combination of “safety and controllability + excellent performance + cost advantage”.
This marks that China is upgrading from the “world factory” to an innovation-driven “computing power engine”.
