AI and Localization Demand Drive China to Fill Semiconductor Supply-Chain Gaps, with Capex Expected to Reach US$82bn by 2030

Overview

This report is the fourth installment in Goldman Sachs' series on China's semiconductor self-sufficiency. It systematically updates product progress across the supply chain, advanced-node supply and demand, capital expenditure, implications for global equipment suppliers, and the market opportunities for GPUs and DRAM in China. The core conclusion is that generative AI, localized procurement, and the self-sufficiency strategy will drive China toward advanced nodes, high-end memory, and advanced packaging, although lithography, R&D investment, and global competitiveness remain key gaps.

Core views

The report first notes that China's semiconductor supply chain is shifting from “expanding existing capabilities” to “filling gaps in high-end products and missing categories.” Domestic equipment capabilities have expanded from etching and deposition to high-end models such as high-aspect-ratio etching and atomic layer deposition, while covering more categories including ion implantation, inspection, and metrology. EDA has also expanded from analog design into digital design, place-and-route, physical verification, and memory tools. In terms of product roadmaps, wafer fabrication is advancing toward 7nm and below, while chiplets and advanced packaging are being used to improve yields. AI chip computing-performance targets are rising from 320 TFLOPS (FP16) in 2023 to 4 PFLOPS (FP8) in the fourth quarter of 2028. Memory is migrating from LPDDR4(X) in 2024 toward LPDDR6, HBM3, and HBM3E in the fourth quarter of 2026. The three forces driving these upgrades are generative AI, the “local for local” supply strategy, and the continued pursuit of self-sufficiency. AI training and inference require advanced nodes, advanced packaging, and high-end memory, while also carrying greater capital intensity per chip. As demand-side evidence, the number of AI chips connected by Huawei Supernode is expected to rise from 384 in 2025 to 1,024 by mid-2026, 8,192 in the fourth quarter of 2026, and 15,488 in the fourth quarter of 2027. Global fabless companies are using local production to reduce the impact of exchange rates, pandemics, and geopolitics. At the same time, a greater share of leading global capacity is being occupied by AI products, creating opportunities for Chinese suppliers to meet demand for conventional ICs and memory. China accounts for approximately 30% of global semiconductor demand, while domestic smartphone, PC, television, and automobile brands further support the adoption of locally produced chips. Progress toward self-sufficiency differs between production volume and value. China's IC production self-sufficiency rate has risen from 38% in January 2010 to 70% in June 2026, but the value-based supply-demand gap remains significant due to insufficient lithography capabilities. Domestic leaders also continue to lag global leaders materially in R&D and capital expenditure. The report therefore believes that sustained increases in capital investment will remain necessary to catch up with global leaders and expand further into overseas markets. For wafers at 7nm and below, Goldman Sachs separately estimates equipment shipments, advanced-node penetration, and die area across eight end markets. It expects Chinese demand to grow at a 17% CAGR during 2025—2035, reaching 619k wpm in 2035, with AI-server demand expected to grow at a 42% CAGR over the same period. The supply forecast is based on capacity-expansion plans and yield improvements: advanced-wafer supply is expected to grow at a 46% CAGR, reaching 410k wpm in 2035. Key assumptions include SMIC adding 30—50k wpm annually during 2026—2031 and 20k wpm annually during 2032—2035, while yields rise from 23% in 2026 to 50% in 2030 and 75% in 2035. Under this scenario, the domestic supply-demand gap will narrow from 92% in 2025 to 34% in 2035, although full supply-demand balance will still not have been achieved by then. Capital expenditure directly supports supply expansion. Following double-digit growth in 2023—2024, China's semiconductor capex was flat in 2025. The report expects year-over-year growth of 13%, 15%, 15%, 10%, and 10% during 2026—2030, reaching US$82bn in 2030. Compared with its prior August 2025 forecasts, Goldman Sachs raised its 2026—2030 capex estimates by 15%, 32%, 51%, 64%, and 79%, respectively. Foundry and memory companies are expected to contribute 80%—85% of China's semiconductor capex over the next several years. On an 8-inch-equivalent basis, China's 8-inch and 12-inch capacity is expected to rise from approximately 6m wpm in 2025 to 14m wpm in 2030. Incremental capacity will primarily come from advanced nodes and high-end memory, requiring more sophisticated manufacturing equipment with higher average selling prices. WFE spending is expected to grow by 13%, 20%, and 15% in 2026—2028, respectively, with China's WFE revenue increasing 20% to US$53bn in 2027. China is expected to account for 29%—31% of global WFE spending during 2026—2028, up from 22% in 2022. Revenue for domestic Chinese equipment suppliers is expected to rise from US$10.2bn in 2025 to US$13.5bn in 2026, US$17.8bn in 2027, and US$22.9bn in 2028, while their value share of the domestic market increases from 26% in 2025 to 31%, 34%, and 38% in 2026—2028. Nevertheless, overseas suppliers will remain key participants due to their leading technologies and intellectual property. Capacity expansion will therefore benefit both domestic and global equipment supply chains. The implications for global equipment suppliers are not entirely uniform. US suppliers are expected to retain a strong but gradually declining share of China's WFE market due to incremental export controls that began affecting sales in the second half of 2025 and Chinese customers' greater emphasis on domestic equipment. However, assuming continued growth in China's capex, Goldman Sachs expects long-term shipments to China to remain at the high end of the mid-20% range of revenue indicated in company guidance. Goldman Sachs also expects global WFE to grow by 36% and 45% in 2026 and 2027, respectively. GAA logic structures, stacked DRAM/NAND, and advanced packaging should drive faster growth in etching and deposition equipment. Within its US coverage, it therefore relatively prefers AMAT, LRCX, and ONTO, all rated Buy. The four Japanese front-end equipment companies Tokyo Electron, SCREEN, Kokusai, and Ebara generated ¥1.21tn of revenue from China in FY2025, accounting for 35% of their SPE revenue, down from ¥1.44tn and 41% in FY2024. The report attributes the decline to earlier advance purchases made in anticipation of potential export restrictions and stronger competition from domestic Chinese equipment. As investment shifts toward advanced nodes and fabs require overseas equipment to ensure yields, Goldman Sachs expects China revenue for all four companies to grow during FY2026—FY2028. However, because non-China operations will grow faster, the share of revenue from China will decline moderately. China revenue for back-end equipment companies Disco, Advantest, and Tokyo Seimitsu has increased steadily over the long term. In the absence of major new export controls, development of China's AI ecosystem and investment in assembly and testing will continue to support demand. Among European equipment companies, Goldman Sachs maintains Buy ratings on ASML, ASMI, and BESI. China is expected to contribute approximately 20% of ASML's group revenue in 2026, near the upper end of its historical 15%—20% range. New fabs, memory demand, and EUV capacity expansion provide support. The 12-month target price is €2,200, based on 32x expected P/E for 2HCY27 plus 1HCY28. ASMI benefits from demand for deposition equipment from Chinese mature-node logic and foundry customers. China orders were strong in the first half of 2026, and related sales can enhance gross margin. Its target price is €955, based on 20x expected EV/EBITDA for 2HCY27 plus 1HCY28. BESI benefits from advanced-packaging capacity expansion related to 2.5D CoWoS, photonics, and pluggable products, as well as a recovery in high-end smartphone modules. Its target price is €325, based on 29x expected EV/EBITDA for the same period. The report estimates China's AI chip market under bull, base, and bear scenarios. TAM is expected to grow at CAGRs of 142%, 69%, and 6%, respectively, during 2025—2030, reaching US$4,123bn, US$678bn, and US$66bn in 2030. Corresponding shipments are 237m, 39m, and 4m chips, while IT power demand is 196GW, 32GW, and 3GW, respectively. Under the base-case AI chip scenario, China's DRAM TAM is expected to grow at a 50% CAGR during 2026—2028 to US$257bn, while HBM grows at a 188% CAGR to US$32bn. This shows that AI not only expands demand for computing chips but also significantly increases demand for high-bandwidth memory. For CXMT, Goldman Sachs initiates coverage with a Buy rating and a 12-month target price of Rmb129, corresponding to 24x 2027E P/E. The report expects average EPS growth of 77% during 2027—2028. CXMT's capacity is expected to more than double from 2026, reaching 665k wpm by 2030. Its conventional DRAM supply is expected to reach 41% and 50% of Samsung's and SK Hynix's supply, respectively, by 2028, up from 28% and 35% in 2025, while covering 50% of China's DRAM demand. The report forecasts CXMT's diluted EPS rising from Rmb2.5 in 2026 to Rmb5.4 in 2027, Rmb7.1 in 2028, and Rmb11.1 in 2030, driven by AI spending, customer supply diversification, capacity expansion, and upgrades toward HBM. Overseas customer opportunities are concentrated primarily in mobile DRAM and conventional DRAM for smartphones and PCs. As global manufacturers shift capacity toward HBM, high prices and limited supply of conventional DRAM are compressing OEM profits and restraining end demand, potentially prompting some US customers to qualify CXMT products. However, CXMT's process technology remains several generations behind global memory peers, and its HBM technology is still at an early stage of maturity. Goldman Sachs does not expect its HBM to enter US customers in the short to medium term. Large-scale shipments will also depend on China-US policies related to semiconductors, GPUs, equipment, and EDA. Aggressive capacity expansion will also become an important variable for investors assessing global DRAM supply and demand in 2028. Nevertheless, Goldman Sachs' base case still projects tight supply that year, albeit less tight than in 2027.

Analysis framework

The report first updates the product and equipment coverage map for each segment of China's semiconductor industry, then explains the drivers of capacity expansion through AI demand, localized procurement, and the self-sufficiency strategy. It subsequently estimates demand based on end-market shipments, advanced-node penetration, and die area, while estimating supply using capacity plans and yield ramp-up assumptions. The report then applies bull, base, and bear scenarios to estimate the GPU and DRAM market opportunities, combines these with bottom-up equipment-supplier revenue forecasts to analyze implications for the global supply chain, and finally derives its rating and target price for CXMT based on capacity, product mix, earnings forecasts, and valuation multiples.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Supply-demand model for wafers at 7nm and below

  On the demand side, wafer demand is estimated using equipment shipments, advanced-node penetration, and die area across eight end markets. On the supply side, effective output is estimated based on capacity-expansion plans and yield ramp-up, which is then used to calculate the path for narrowing the supply-demand gap.

• (Out-of-vocabulary methodology)

  TAM scenario analysis

  The report establishes bull, base, and bear scenarios, using different AI chip growth trajectories to derive the 2030 market size, chip shipments, and IT power demand, and then uses the base scenario to further estimate DRAM and HBM demand.

• Valuation MethodPE/PEG valuation

  Forward P/E valuation

  CXMT's Rmb129 target price corresponds to 24x 2027E P/E; ASML's target price is based on 32x expected P/E for 2HCY27 plus 1HCY28.

• Valuation MethodEV/EBITDA valuation

  Enterprise-value multiple valuation

  The target prices for ASMI and BESI use 20x and 29x expected EV/EBITDA for 2HCY27 plus 1HCY28, respectively, valuing enterprise value relative to operating earnings capacity.

Asset mapping & comparison

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

• CXMT (688825.SS)

  A leading Chinese DRAM company and the core company newly covered in the report, benefiting from AI demand, customer supply diversification, capacity expansion, and upgrades toward HBM.

  Strengths

  Capacity is expected to reach 665k wpm by 2030; in 2028, it is expected to cover 50% of China's DRAM demand, while conventional DRAM supply reaches 41% and 50% of Samsung's and SK Hynix's supply, respectively.

  Weaknesses

  Its process technology is several generations behind global memory peers, and HBM remains at an early stage of technological maturity.

  Comparison

  Conventional DRAM supply relative to Samsung and SK Hynix is expected to rise from 28% and 35% in 2025 to 41% and 50% in 2028.

  Risks

  China-US trade and technology restrictions could affect large-scale overseas shipments, while aggressive capacity expansion could also alter global DRAM supply and demand in 2028.

• Domestic Chinese WFE suppliers (overall)

  Beneficiaries of Chinese fab expansion, localized procurement, and the extension of equipment categories into high-end etching, deposition, ion implantation, inspection, and metrology.

  Strengths

  Their value share of the domestic WFE market is expected to rise from 26% in 2025 to 38% in 2028, while revenue increases from US$10.2bn to US$22.9bn.

  Weaknesses

  They continue to lag global equipment leaders in advanced technologies and intellectual property.

  Comparison

  Domestic share is rising, but global equipment companies remain key suppliers to the Chinese market.

• AMAT, LRCX, ONTO

  Goldman Sachs' relatively preferred Buy-rated names within its US equipment coverage, benefiting from global WFE growth and demand for etching, deposition, and advanced packaging.

  Strengths

  Global WFE is expected to grow by 36% and 45% in 2026 and 2027, while GAA, stacked memory, and advanced packaging support demand for relevant equipment.

  Weaknesses

  Their shares of the Chinese market are expected to remain strong over the long term but gradually decline.

  Comparison

  Relative to other US equipment names under coverage, Goldman Sachs explicitly states a preference for these three companies.

  Risks

  Incremental US export controls and increased purchases of domestic equipment by Chinese customers could reduce their shares of the Chinese market.

• Japanese front-end and back-end equipment companies

  China remains an important demand market for Tokyo Electron, SCREEN, Kokusai, Ebara, Disco, Advantest, and Tokyo Seimitsu.

  Strengths

  Advanced nodes' yield requirements support orders for high-end overseas equipment, while back-end equipment also benefits from China's AI ecosystem and assembly-and-testing investment.

  Weaknesses

  China revenue for the four front-end companies fell to ¥1.21tn in FY2025, accounting for 35% of SPE revenue, down from ¥1.44tn and 41% in FY2024.

  Comparison

  Absolute China revenue is expected to grow during FY2026—FY2028, but its share of total revenue will decline moderately as non-China revenue grows faster.

  Risks

  Future major export controls could affect equipment demand from China.

• ASML (ASML.AS)

  Benefits from the construction of new Chinese fabs and domestic lithography demand; Buy rating maintained.

  Strengths

  China is expected to contribute approximately 20% of group revenue in 2026, while EUV capacity and equipment-performance improvements support long-term growth.

  Weaknesses

  Chinese customers still need to absorb lithography capacity delivered in recent years.

  Comparison

  China is expected to account for approximately 20% of revenue in 2026, at the upper end of its historical 15%—20% range.

  Risks

  EUV delays, fluctuations in the capex cycle, and adverse changes in market share.

• ASMI (ASMI.AS)

  Benefits from demand for deposition equipment from Chinese mature-node logic and foundry customers; Buy rating maintained.

  Strengths

  China orders were strong in the first half of 2026, and sales in China are accretive to gross margin.

  Weaknesses

  Demand visibility for 2027 still depends on customer behavior and the regulatory environment.

  Risks

  A deterioration in the semiconductor cycle, stronger-than-expected competition, and high customer concentration.

• BESI (BESI.AS)

  Benefits from advanced-packaging capacity expansion driven by 2.5D CoWoS, photonics, and pluggable products; Buy rating maintained.

  Strengths

  It is also supported by a recovery in high-end smartphone modules and early signs of recovery in industrial and automotive demand.

  Weaknesses

  Its conventional packaging business still depends on a recovery in industrial, automotive, and end-market demand.

  Risks

  Cyclicality in customer spending, delayed adoption of hybrid bonding, and intensifying competition.

Key data

• China's IC production self-sufficiency rate70%June 2026; 38% in January 2010, although the value-based gap remains significant.
• China semiconductor capexUS$82bn2030E; year-over-year growth during 2026—2030E is expected to be 13%, 15%, 15%, 10%, and 10%.
• Capex forecast upgrades15%/32%/51%/64%/79%Relative to the August 2025 forecasts, corresponding to 2026—2030E in sequence.
• China's 8-inch and 12-inch capacity6m wpm rising to 14m wpmFrom 2025 to 2030E, driven primarily by capacity expansion in advanced nodes and high-end memory.
• Demand for wafers at 7nm and below619k wpm2035E; 17% CAGR during 2025—2035E.
• Supply of wafers at 7nm and below410k wpm2035E; 46% CAGR during 2025—2035E.
• Advanced-node supply-demand gap34%2035E, narrowing significantly from 92% in 2025.
• China WFE marketUS$53bn2027E, up 20% year over year; expected to account for 29%—31% of global WFE spending during 2026—2028E.
• Share of domestic Chinese WFE suppliers26% rising to 38%By value, from 2025 to 2028E.
• China AI chip TAM in 2030US$4,123bn/US$678bn/US$66bnCorresponding to the bull, base, and bear scenarios, respectively.
• China AI chip shipments in 2030237m/39m/4mCorresponding to the bull, base, and bear scenarios, respectively; IT power demand is 196GW, 32GW, and 3GW, respectively.
• China DRAM and HBM TAMUS$257bn/US$32bn2028E; DRAM and HBM CAGRs during 2026—2028E are 50% and 188%, respectively.
• CXMT capacity665k wpm2030E, more than double the 2026E level.
• CXMT diluted EPSRmb2.5/Rmb5.4/Rmb7.1Corresponding to 2026E, 2027E, and 2028E, respectively.
• CXMT target price and valuationRmb129; 24x 2027E P/E12-month target price, initiated with a Buy rating.

Impact & implications

The report believes the focus of China's semiconductor capex is shifting from mature-node expansion toward advanced nodes, memory, and advanced packaging. Domestic equipment suppliers will gain share through category expansion and localized procurement, while overseas equipment suppliers can continue participating in market growth through high-end technologies. AI demand simultaneously expands the GPU, DRAM, and HBM markets and supports CXMT's capacity expansion and product upgrades. However, lithography, R&D investment, yields, and trade restrictions will determine how quickly the self-sufficiency gap can narrow.

Risks

• New China-US import and export restrictions on GPUs, semiconductor equipment, EDA software, and memory could affect sales by CXMT and global equipment suppliers.
• CXMT's process technology remains several generations behind global peers, and its HBM technology has relatively low maturity, limiting its ability to enter US HBM customers in the short to medium term.
• CXMT's aggressive capacity expansion could alter the global DRAM supply-demand balance in 2028.
• ASML faces risks from EUV delays, fluctuations in the capex cycle, and adverse changes in market share.
• ASMI faces risks from deterioration in the semiconductor cycle, stronger-than-expected competition, and high customer concentration.
• BESI faces risks from cyclicality in customer spending, delayed adoption of hybrid bonding, and intensifying competition.

What to watch

• Monitor the upcoming China-US political summit and other trade discussions to assess whether incremental restrictions will be imposed on semiconductors, memory, equipment, and EDA.