Southern Old Talisman

The Ministry of Industry and Information Technology recently held a national conference on the high-quality development of the electronic information manufacturing industry in Wuhan, and made it clear that "accelerating the construction of an efficient and unified artificial intelligence chip computing interconnection ecosystem" was the core task, setting the direction for the development of AI embedded and chip industries. During the same period, Cambrian, Horizon and other companies released a new generation of AI chips intensively, and the market share of domestic chips continued to rise. AI embedded technology in automobiles, industries, homes and other scenarios accelerated penetration. With the breakthrough of end-to-side computing power, the deepening of ecological coordination, and the promotion of localization substitution, the AI embedded and chip industries are bidding farewell to "single-point competition", entering a new stage of high-quality development of "efficient computing power, open ecology, and inclusive scenarios", and becoming the core engine driving the upgrading of the digital economy.

Policy precision empowers, cracks the pain points of the industry, and builds a solid foundation for development. The MIIT meeting pointed directly to the core pain points of the current AI chip industry "hash island" and "ecological closure", and clearly proposed to promote the high-quality development of the advanced computing industry, pull the industrial chain to jump to higher value links, and help achieve the goal of 12.5% of GDP in the added value of the core industries of the digital economy by 2030. In order to break down ecological barriers, the conference promoted the industry to speed up the formulation of unified agreements. ZTE, Beijing Zhiyuan Research Institute and other enterprises and institutions responded positively. The former launched the "super node" architecture to achieve multi-brand GPU compatibility, and the latter released the Zhongzhi FlagOS 2.0 open-source system to build a unified technology stack for different AI chips, which greatly reduced development and deployment costs. All localities also made simultaneous efforts. The domestic computing power of the artificial intelligence public computing power center in Haidian District, Beijing accounted for more than 80%. Qingdao listed advanced computing as a key area, and planned to break through 12 billion yuan in industrial revenue in 2026, forming an industrial development pattern linked up and down. TAG5}}

technology continues to break through, the rise of domestic chips, and the cost-effectiveness advantage is highlighted. Since 2026, domestic AI chips have ushered in explosive iteration. Cambrian, Biren, Muxi, and Horizon four companies have released a new generation of products at the same time. The performance is benchmarked against international flagships, but the price is only 40% to 70% of international products. When Huawei's 950PR chip runs a 70B parameter large model, the response speed is nearly three times faster than that of Nvidia H20, and the delay is reduced by 60%; the Horizon "Journey" series chips support local operation of 70 billion parameter large models, and the voice interaction response time is less than 100ms. In the field of embedded technology, model lightweight, dedicated chip upgrade, and edge computing empower three major breakthroughs to cooperate. Through pruning, quantization and other technologies, the memory occupation of the 100 billion parameter large model is compressed to less than 16GB, and the accuracy is retained in the cloud by 95%. Low-power embedded devices realize real-time reasoning, and solve the problems of power walls, memory walls, and interconnection walls faced by traditional chips.

Embedded scenes are fully penetrated, and global coverage is achieved from consumer level to industrial level. In the field of smart cars, the car-grade RISC-V chip integrates NPU and supports end-to-end voice interaction. Geely "Xingrui AI Cloud Power" optimizes the global distribution of oil and electricity through embedded AI, and the energy efficiency is improved by 18%; in the smart home scenario, Xiaomi, Aqara and other brands use edge AI to achieve device linkage. The human body sensor recognizes the elderly falling and alarms within 0.1 seconds, and the false alarm rate is reduced to 0.3%; in the industrial field, Lipper Technology's solid crystal inspection system uses edge AI to achieve sub-micron defect recognition, with a detection speed of 15K +/hour. TSMC's smart factory uses edge AI to optimize manufacturing parameters, and the yield of key processes is increased by 3.5%. In the medical field, Apple Watch Series 9 realizes atrial fibrillation detection through embedded AI, with a sensitivity of 98.3% and data throughout the process Processing on the device side to ensure user privacy.

The market scale continues to expand, and ecological collaboration has become an industry consensus. Industry data show that in 2025, our country's computer, communication and other electronic equipment manufacturing revenue reached 17.40 trillion yuan, accounting for 12.5% of industrial revenue; in March 2026, the average daily amount of AI word element calls exceeded 140 trillion, an increase of more than 1,000 times compared with the beginning of 2024. With the explosion of demand, the domestic market share of domestic AI chips has exceeded 50%, occupying half of the country for the first time. Huawei Atlas 950 SuperPod, Cambrian Siyuan 590 and other products have realized 10,000-card interconnection, effectively relieving high-end chip dependence. Enterprises have accelerated the construction of an open ecosystem. Edao Digital, SenseTime Jueying and other enterprises have promoted the deep integration of embedded AI and end point devices. ZTE has opened hardware interface specifications, promoted industry standardization, and formed a full-link collaborative system of "chip-embedded system-end point application".