Tracking China’s Semiconductor Self-Sufficiency Acceleration Driven by Local AI GPU and Memory

Core Conclusion

China's semiconductor self-sufficiency ratio is on a faster-than-anticipated growth trajectory, forecast to rise from 24% in 2025 to 32% by 2028. This acceleration is structurally driven by aggressive local memory capacity expansion, a booming domestic AI GPU ecosystem with tangible manufacturing progress, and steady gains in mature-node segments. The investment implication is a multi-year opportunity for foundries, semicap equipment vendors, and select chip designers positioned as key enablers of this localization cycle.

Evidence Chain

Self-sufficiency is accelerating with concrete, multi-pronged drivers. The overall ratio is projected to reach 32% by 2028, up from 24.3% in 2025, supported by a 22% revenue growth for local semiconductor firms in 2025. This trend is not broad-based but concentrated in high-impact segments: domestic DRAM/NAND production shares rose sharply to 8.5%/12.6% in 2025, and locally produced GPU sales doubled to US$13bn. The investment meaning is that aggregate top-line growth for the local industry understates the profound market share shifts occurring within strategic sub-sectors like memory and compute.

Memory and AI GPU are the dual engines of market share gain. Memory vendor capacity plans are clear and aggressive; CXMT and YMTC are expected to add 60kwpm and 25kwpm of new capacity, respectively, in 2026. Concurrently, the domestic AI GPU TAM expanded to US$32bn in 2025, with local foundries ramping 12nm-and-below capacity to support production. The fact that a leading AI firm like MiniMax has begun adopting local AI GPUs signals early ecosystem viability. This creates a self-reinforcing loop: GPU demand pulls for advanced node capacity, which in turn enables further design iteration and adoption. The investment takeaway is to focus on companies along this specific capability build-out, from memory makers to leading-edge foundries like SMIC and Hua Hong.

Plunging equipment imports confirm the localization shift, benefiting domestic toolmakers. Semiconductor equipment imports into China fell 24% Y/Y in Feb 2026, with double-digit declines from all major supplier nations. This trend starkly illustrates the substitution occurring at the equipment layer. Against this backdrop of a still-growing China wafer fab equipment market, local vendors are positioned to capture an expanding share of a large, captive TAM. The investment implication is direct: domestic semicap companies (e.g., NAURA, AMEC, ACMR) are fundamental beneficiaries of this import substitution trend, with revenue visibility tied to the broader capacity expansion cycle.

Key Risks

• Technological hurdles and ROI challenges in advancing to leading-edge nodes (e.g., sub-7nm).
• Escalation of geopolitical friction leading to further supply chain restrictions.
• Pace of local technology iteration failing to keep up with global innovation.
• Cyclical downturns in end-demand from key sectors like AI infrastructure or automotive.

Investment Implications for the Localization Cycle

We favor capital allocators and tool providers in this cycle. Foundries (SMIC, Hua Hong) are critical infrastructure enabling AI and memory production. Semicap equipment vendors (NAURA, AMEC, ACMR) have the clearest structural tailwind from import substitution. Within design, companies leveraged to high-growth niches like HPC (Montage Technology) or serving as a memory proxy (GigaDevice) offer targeted exposure. The thematic is less about broad sector exposure and more about identifying companies integral to building specific, strategic domestic capabilities.

Appendix Data Summary