The logic of the AI chip industry is driven by two core forces: the explosive demand for computing power and the push for domestic substitution. It relies on the synergy of the “design – manufacturing – packaging” industrial chain, achieving technological breakthroughs and commercialization under the dual drivers of policy and market.
The demand side constitutes the underlying support for industry growth. The global demand for AI computing power is growing exponentially, with the domestic AI computing power demand expected to exceed 1800 EFLOPS by 2025, achieving a compound annual growth rate of 62%. On one hand, restrictions on the export of high-end chips abroad, such as Nvidia’s A100 and H100 products, have accelerated the push for domestic substitution, with the self-sufficiency rate of domestic AI chips rising to 40% in the first half of 2025, doubling from three years ago; on the other hand, the construction of intelligent computing centers, new energy vehicles, industrial robots, and other scenarios are opening up incremental space, with the demand for AI servers from domestic internet giants alone reaching hundreds of thousands of units, corresponding to a market worth hundreds of billions.
A differentiated competitive landscape has formed across various links of the industrial chain. In the design phase, leading companies in niche markets are more resilient. For instance, Lanqi Technology’s DDR5 chips meet the high bandwidth requirements of AI servers and have achieved profit growth through deep cooperation with Nvidia, while Cambrian faces customer concentration risks despite explosive revenue growth. In the manufacturing phase, breakthroughs have been made to overcome “bottlenecks”; SMIC’s 14nm process yield has reached 98%, and the 7nm process yield exceeds 60%, supporting the capacity needs of companies like Huawei HiSilicon. Packaging and testing have become key to technological upgrades, with Changdian Technology’s 3D packaging and Chiplet technology achieving a doubling of computing power and cost reduction, with advanced packaging accounting for over 40% of revenue.
The health of the industry relies on core evaluation criteria. Companies need to have “real orders” to support them; for example, Haiguang Information’s revenue reached 2.4 billion in the first half of 2025, with a domestic x86 server penetration rate of 25%; R&D investment must be precisely focused, with Lanqi Technology’s breakthroughs in the DDR5 field forming competitive barriers; at the same time, stable cash flow must be ensured to avoid the risk of Cambrian’s “profitability but negative operating cash flow”.
Policies and ecosystem construction provide development guarantees. The National Big Fund’s third phase of investment supports advanced process research and development, with the Yangtze River Delta computing power cluster accounting for 38% of the national total, forming an industrial agglomeration effect. Companies like Huawei Ascend are reducing inference costs to 1/30 of international levels through ecosystem cooperation, accelerating commercialization. This logic essentially reflects the transformation of China’s technology industry from import dependence to self-control, highlighting the core demand for technological independence.