Under a push to localize capacity, China's leading domestic foundries are well-positioned to leverage their home market and policy advantages to replicate TSMC's growth trajectory. This momentum is fueled by a virtuous cycle of massive orders supporting substantial R&D and capital expenditure, which is accelerating domestic substitution toward advanced process nodes.
Geopolitical factors, AI chip demand, and regional subsidy policies are reshaping the global semiconductor manufacturing landscape into three distinct camps: China suppliers serving Chinese customers, US suppliers serving US customers, and non-China suppliers serving non-China customers.
According to a Chinese brokerage report, the AI compute chip market in China is projected to reach RMB800.4 billion (approximately US$115.8 billion) by 2027, corresponding to a demand of 8 million compute chips, equivalent to Huawei's Ascend 910B. The wafer demand for these compute chips will require about 21,000 monthly wafers. However, given constraints on China's advanced process capacity and estimated yields between 20% and 33%, the advanced node foundry market size in 2027 is expected to be around US$3.5 billion — still far behind the scale of the US and non-China supply chain camps.
Credit: DIGITIMES
From mature to advanced: the next frontier
In semiconductor manufacturing, China has largely achieved self-sufficiency at mature nodes of 28nm and above in recent years. Domestic IC design firms' reliance on local foundry leaders is also growing, pushing domestic Chinese capabilities from mature to advanced process nodes. Domestic substitution for advanced node processes is set to become the next major development.
The semiconductor industry currently faces an insatiable appetite for AI compute capacity. SMIC's sustained high capex through 2026 reflects this shift — from competition in consumer electronics to explosive AI compute demand growth. According to SMIC's latest 2025 annual report, SMIC South completed a capital increase in December 2025, raising registered capital from US$6.5 billion to US$10.1 billion, with participation from multiple national-level funds, including China Integrated Circuit Industry Investment Fund phases one through three and the Shanghai Integrated Circuit Industry Investment Fund.
Credit: DIGITIMES
State capital backs SMIC's advanced push
While this capital injection aims to reduce SMIC South's debt ratio and optimize its financial structure, the unit remains under the long-term leadership of co-CEO Liang Mong Song and is regarded as the core R&D and volume production base for SMIC's advanced processes. The infusion of state-level funding signals a continued strategic emphasis on advancing domestic capabilities.
As the AI era takes hold, whether for cloud-based large model training or edge devices like AI smartphones and computers, wafer consumption is growing dramatically. In this context, production capacity directly determines bargaining power within the supply chain.
The capex arms race
Globally, the AI capacity arms race is intensifying. SMIC's planned capex exceeding US$8 billion in 2026 underscores its significant commitment to strengthening China's domestic supply chain for advanced processes.
Meanwhile, US semiconductor export controls have tightened in recent years, restricting exports of equipment for advanced nodes at 14nm/16nm and below, as well as limiting foundries such as TSMC from providing 7nm-and-below high-end AI chip foundry services to China. These measures aim to block China's access to overseas advanced process technologies.
Despite this, China's domestic end-market demand for advanced nodes continues to steadily expand. On the mobile front, exemplified by Huawei, the return of high-end domestic smartphones generates roughly tens of millions of advanced node SoC foundry orders annually.
Compute capacity on a steep climb
Advanced node foundry capability forms the foundation for AI chips. International Data Corporation (IDC) forecasts that China's intelligent compute capacity will reach 1,460.3 EFLOPS in 2026 — double the 2024 level — and grow to 2,781.9 EFLOPS by 2028. From 2023 to 2028, China's intelligent compute and general compute capacities are expected to achieve five-year compound annual growth rates (CAGR) of 46.2% and 18.8%, respectively.
Credit: DIGITIMES
Article translated by Lily Hess and edited by Jerry Chen
