China's semiconductor localization accelerates, with AI GPU and memory as the main drivers of higher self-sufficiency

Overview

This report is Morgan Stanley's industry research and conference notes on China's semiconductor localization. Its core conclusion is that China's semiconductor self-sufficiency will continue to improve. The latest upgrade is mainly driven by domestic AI GPU demand and capacity ramp-up, memory maker expansion, higher penetration of domestic equipment in mature nodes, and advanced-node capacity expansion. The report also tracks semiconductor equipment imports, monthly stock performance, and upcoming catalysts.

Core views

The report believes China's semiconductor localization has entered a new phase driven by AI compute and memory. China's semiconductor self-sufficiency rose slightly to 24.3% in 2025 and is expected to reach 32% in 2028e. AI GPU companies are working with Chinese foundries to improve 7nm-10nm yields, and AI application players such as MiniMax have already started using Chinese AI GPUs to meet inference demand. On memory, CXMT and YMTC capacity expansion is key to improving DRAM and NAND self-sufficiency. For mature-node products, MCUs and power discretes benefit from local foundry capacity expansion and automotive demand, while analog chips may pause in 2026 due to price competition from overseas vendors.

Analysis framework

The report combines Gartner revenue and market demand data, semiconductor self-sufficiency estimates by segment, equipment import trends, WFE capex expectations, company meeting feedback, and monthly stock performance to assess the progress and investment implications of China's semiconductor localization.

Methodology notes

• Industry self-sufficiency estimationSemiconductor revenue vs. local market demand

  Estimate self-sufficiency by comparing Chinese semiconductor company revenue with China's semiconductor market size

  Using Gartner data, the report estimates that Chinese semiconductor companies generated about US$53bn of revenue in 2025 versus a China semiconductor market of about US$217bn, implying a self-sufficiency rate of about 24.3%, and uses this to forecast 32% in 2028e.

• Segment-by-segment driver decompositionAI GPU, memory, mature nodes, equipment localization

  Break down localization contributions by product and supply-chain segment

  The report attributes the improvement in self-sufficiency to growth in domestic AI GPU sales and TAM, CXMT/YMTC capacity additions, domestic equipment penetration at mature-node fabs and memory makers, and advanced-node capacity expansion.

• Investment mappingScreening localization beneficiaries

  Identify benefiting assets across foundry, equipment, chip design, and materials

  The report positions SMIC and Hua Hong as key enablers of China's AI localization, Naura, AMEC, and ACMR as beneficiaries of the equipment localization cycle, and mentions Montage and GigaDevice as proxies for HPC chips and memory, respectively.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• SMIC

  Key foundry enabler for China's AI localization

  Strengths

  Listed by the report as OW; benefits from advanced-node capacity expansion and AI GPU mass-production demand.

  Weaknesses

  Returns on advanced-node investment and process yields remain challenges.

  Comparison

  Compared with mature-node companies, SMIC has more direct exposure to AI GPU and advanced-node localization.

  Risks

  Export controls, equipment supply constraints, advanced-node yields, and uncertainty over capex returns.

• Hua Hong

  Beneficiary of specialty processes and mature-node localization

  Strengths

  Listed as EW by the report and described as one of the important enablers of China's AI localization.

  Weaknesses

  Its leverage to the advanced-node AI GPU chain may be weaker than SMIC's.

  Comparison

  More focused on mature-node and specialty processes, with a different benefit path from SMIC.

  Risks

  Mature-node competition, pricing pressure, and demand-cycle volatility.

• Naura、AMEC、ACMR

  Beneficiaries of China's semiconductor equipment localization cycle

  Strengths

  The report is constructive on Chinese semiconductor equipment, expecting domestic equipment makers to continue gaining share as China's WFE TAM increases.

  Weaknesses

  Equipment demand is affected by the pace of fab capex.

  Comparison

  Compared with imported equipment makers, domestic vendors benefit from localization substitution and declining imports.

  Risks

  Speed of technological breakthroughs, customer validation cycles, export controls, and order volatility.

• Montage

  A name linked to HPC chips and China's AI compute localization

  Strengths

  The report expresses a preference for Montage in chip design as a representative HPC chip name.

  Weaknesses

  Demand is tied to servers and AI compute capex, which may be cyclical.

  Comparison

  Compared with general mature-node names, it is closer to the AI/HPC compute chain.

  Risks

  Customer concentration, technology iteration, and changes in the competitive landscape.

• GigaDevice

  A proxy for memory localization

  Strengths

  The report lists it among preferred chip-design names, positioning it as a memory proxy.

  Weaknesses

  Memory pricing and end-demand volatility may affect performance.

  Comparison

  Compared with equipment and foundries, GigaDevice has more design-side and storage-cycle leverage.

  Risks

  Storage cycles, overseas competition, and pricing pressure.

• CXMT、YMTC

  Core capacity sources for China's DRAM and NAND self-sufficiency improvement

  Strengths

  Their global output shares rose to 8.5% and 12.6% in 2025, respectively, and the report expects further expansion in 2026.

  Weaknesses

  Advanced memory technology and mass-production yields still require continued ramp-up.

  Comparison

  Compared with mature-node chips, memory contributes more directly to self-sufficiency improvement.

  Risks

  Equipment constraints, export controls, price cycles, and capex execution risk.

Key data

• China semiconductor self-sufficiency rate24.3% in 2025, 23.7% in 2024, forecast 32% in 2028eUp 0.6 percentage points from 2024 to 2025; the report raised its medium-term forecast.
• China semiconductor company revenueUS$53bn in 2025, +22% YoY; US$43bn in 2024Based on Gartner data.
• China semiconductor market sizeAbout US$217bn, 27% of global demandUsed to estimate China's semiconductor self-sufficiency rate.
• Domestic GPU sales in ChinaUS$13bn in 2025, more than double 2024AI GPU demand is supported by higher CSP capex and post-DeepSeek-moment demand.
• China GPU TAMUS$32bn in 2025, US$19bn in 2024Reflects the expansion of domestic AI GPU market space.
• CXMT global output share8.5% in 2025, 5.6% in 2024An important source of DRAM self-sufficiency improvement.
• YMTC global output share12.6% in 2025, 7.4% in 2024An important source of NAND self-sufficiency improvement.
• Expected new memory capacity in 2026CXMT +60kwpm; YMTC +25kwpmThe report's charts indicate that memory makers still have significant capacity-expansion potential.
• China semiconductor equipment importsUS$1.2bn in Feb 2026, -24% YoY; three-month moving average -15% YoYThe decline in imports is related to domestic equipment substitution and cyclical factors.
• China WFE forecastUS$46.5bn in 2026, +12% YoYMorgan Stanley's U.S. team expects this to be driven mainly by memory and leading-node expansion.
• YTD change in major import sourcesU.S. -54%, Netherlands -13%, South Korea -70%, Japan -39%, Singapore -9%The report says imports from all major sources declined YoY.
• Monthly outperformersSICC +11.6%, SG Micro +9.4%, Maxscend +6.9%SICC was supported by 8-inch SiC substrate share and LPU cooling optionality; SG Micro benefited from price increases in the analog industry.
• Monthly underperformersShanghai Fudan -26.4%, Omnivision -21.1%Affected by sentiment normalization in low-Earth-orbit satellites and weak smartphones, respectively.

Impact & implications

The investment implication tilts toward China's semiconductor localization chain: foundries and semiconductor equipment are critical infrastructure for AI GPU mass production and advanced-node expansion, memory maker capacity additions strengthen DRAM/NAND localization optionality, and mature-node products benefit from local foundry capacity and automotive semiconductor demand. The report is more constructive on Chinese foundries and semiconductor equipment companies, while noting that export controls, executive orders, investment-banking conflicts of interest, and stock-specific cyclicality should be incorporated into investment decisions.

Risks

• Export controls and restrictions related to U.S. Executive Order 14032 may affect trading in the relevant securities and industry-chain activities.
• Morgan Stanley discloses investment-banking, market-making, service, or potential fee relationships with many covered companies, which may affect research objectivity due to conflicts of interest.
• Localization of AI GPUs still depends on 7nm-10nm yield improvements and advanced-node capacity expansion; execution shortfalls would weaken the self-sufficiency uplift path.
• Analog chips may see improvement pause in 2026 due to overseas price competition, and mature-node products also face pricing and cyclical pressure.
• Memory expansion is affected by equipment availability, yields, price cycles, and capex pace; delays in CXMT/YMTC expansion would affect the contribution to self-sufficiency.
• Weak smartphone demand and rising memory prices may weigh on the performance of some semiconductor design companies.

What to watch

• Whether China's semiconductor self-sufficiency follows the path from 24.3% in 2025 to 32% in 2028e.
• Progress by domestic AI GPU companies and Chinese foundries such as HLMC on 7nm-10nm yield improvements.
• Execution of CXMT and YMTC's planned additions of 60kwpm and 25kwpm capacity in 2026, respectively.
• Whether China's WFE reaches US$46.5bn in 2026, growing 12% YoY, and whether domestic equipment makers continue to gain share.
• Whether the downtrend in semiconductor equipment imports continues, especially from the U.S., the Netherlands, South Korea, Japan, and Singapore.
• Whether the scope of Chinese AI GPU adoption by AI application players such as MiniMax expands.
• Changes in pricing, share, and automotive demand in mature-node subsegments such as analog, MCUs, power semiconductors, and SiC.
• The impact of subsequent China AI industry seminars and GPU product events on market expectations.