AI semiconductors are strengthening, and TSMC remains the core of advanced-node and CoWoS supply

Overview

This report is Morgan Stanley’s company and industry outlook for Greater China semiconductors, centered on TSMC 2Q26 earnings preview, 3Q26 guidance, CoWoS/SoIC capacity expansion, AI GPU and ASIC demand, cloud capex, China AI compute power, and storage supply-demand. The report judges that AI semiconductor demand remains strong, and that TSMC maintains a critical position through advanced nodes, EUV supply, CoWoS and SoIC capabilities, and may raise prices 5%-10% for leading nodes in 2027.

Core views

Core views include: first, TSMC remains the most constrained foundry and packaging platform in the advanced-node and AI semiconductor supply chain; 2nm demand is strong, N3 capacity continues to rise, and N5 capacity is set to decline in 2027. Second, AI training, inference, and agentic workloads are driving expanding demand for GPUs, custom ASICs, CPU orchestration, HBM, CoWoS, SoIC, and test equipment. Third, cloud vendors’ capex remains elevated, and Top 14 listed global CSPs’ cloud capex in 2027 is estimated to approach US$1.3tn. Fourth, China AI chip demand is supported by inference demand from DeepSeek, progress in the domestic supply chain, and cost advantages of local accelerators, while chip capacity, regulation, and export controls remain constraints.

Analysis framework

The report combines company earnings preview, supply-demand modeling, capex tracking, supply chain capacity decomposition, customer demand breakdown, TAM estimation, valuation comparison, and scenario analysis, with a focus on TSMC advanced-node capacity, CoWoS/SoIC packaging capability, CSP cloud capex, NVIDIA GB200/300 rack demand, custom chips such as TPU/Trainium, HBM, and China AI accelerator demand to derive semiconductor value-chain momentum.

Methodology notes

• earnings_previewTSMC quarterly earnings and guidance preview

  Assess short-term earnings momentum through preliminary 2Q26 estimates, 3Q26 guidance, gross margin, operating margin, and EPS forecasts.

  The tables compare TSMC 2Q26 revenue, gross margin, operating margin, and EPS against company guidance and MS and market consensus expectations, and also provide revenue and margin assumptions for 3Q26.

• supply_chain_modelAI semiconductor supply-chain capacity and demand model

  Estimate AI semiconductor revenue and capacity gaps by splitting CoWoS, SoIC, HBM, wafer, rack, and customer demand.

  The report breaks down CoWoS, SoIC, AI wafer consumption, HBM consumption, and NVIDIA GB200/300 rack supply-demand by customer and year to assess the strength of benefits for TSMC and related supply-chain participants.

• tam_forecastAI Semi TAM and China AI GPU TAM forecast

  Estimate global AI semiconductor, China AI GPU, and CPU orchestration market sizes using top-down and bottom-up TAM frameworks.

  The report expects AI semi TAM to reach about US$753bn by 2030 and China AI chip TAM to about US$91bn by 2030, and raises the base-case orchestration CPU TAM to US$79bn.

• capex_trackerCloud capex tracker

  Track Amazon, Google, Microsoft, Meta and other CSP capex to judge the sustainability of AI chip demand.

  The report notes that Top 4 CSP 1Q26CY capex rose 95% year-over-year, and estimates that Top 14 listed global CSPs’ cloud capex in 2027 will be close to US$1.3tn.

Asset mapping & comparison

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

• TAIWAN SEMICONDUCTOR MANUFACTURING CO LTD / US.TSM

  Core covered company and core beneficiary in the AI semiconductor supply chain

  Strengths

  Advanced nodes, logic density, EUV resources, CoWoS and SoIC capacity, 2nm demand, pricing power, and strong AI customer demand are all strengths.

  Weaknesses

  High capex, advanced capacity ramp-up, and dependence on sustained AI demand increase operating volatility.

  Comparison

  The report says TSMC still leads Intel and Samsung foundry in technology roadmap and logic density, and is competitive with Intel EMIB in advanced packaging.

  Risks

  Cloud capex slowdown, AI demand below expectations, EUV or packaging capacity expansion below expectations, export controls, and customer concentration risk.

• AI semiconductor supply chain

  Main thread of cyclical industry momentum

  Strengths

  Cloud capex, GPUs, ASICs, HBM, CoWoS, SoIC, test equipment, and China AI compute demand jointly support growth.

  Weaknesses

  Non-AI semiconductor demand may weaken, while storage and T-glass bottlenecks may raise costs.

  Comparison

  AI semiconductors are expected to contribute about half of incremental global semiconductor market growth by 2030, clearly ahead of non-AI semiconductor growth.

  Risks

  Budget, energy, chip capacity, regulation, and technology roadmap changes could limit growth.

• China AI compute power and domestic AI accelerators

  Regional incremental demand and domestic substitution theme

  Strengths

  DeepSeek-driven inference demand, improving domestic supply-chain capability, and lower TCO with comparable token costs give local chips an edge in Chinese inference scenarios.

  Weaknesses

  Advanced nodes and high-end capacity remain constrained, and some performance and ecosystem aspects still lag NVIDIA.

  Comparison

  The report compares Chinese versus U.S. AI chips, systems, and infrastructure, and notes that China’s infrastructure advantages help narrow parts of the technology gap.

  Risks

  Export controls, capacity constraints, regulation, customer procurement timing, and uncertainty around commercialization pace of domestic chips.

Key data

• TSMC 2026e/2027e capexUS$56bn / US$75bnThe report says TSMC 2026/27e capex is US$56bn and US$75bn, respectively.
• Potential price increase for TSMC leading nodes5%-10% in 2027The report believes TSMC can raise prices for leading nodes in 2027 to reflect the value it provides to customers.
• AI semi TAM 2030eabout US$753bnThe report expects AI semi TAM to reach about US$753bn by 2030.
• Potential global semiconductor market size in 2030about US$1.5tnThe report says the global semiconductor market may reach US$1.5tn by 2030, with AI semiconductors contributing about half.
• Top 4 CSP 1Q26CY capex growth+95% Y/YAmazon, Google, Microsoft, and Meta capex were up 95% year-over-year.
• 2027 global cloud capex estimateclose to US$1.3tnMorgan Stanley’s cloud capex tracker covers Top 14 listed global CSPs and excludes sovereign AI.
• AI computing wafer consumption 2027e>US$46bnThe report says AI computing wafer consumption could exceed US$46bn by 2027, with NVIDIA accounting for the majority.
• China AI chip TAM 2030eUS$91bnThe report expects China AI chip TAM to grow to US$91bn by 2030.
• Base case orchestration CPU TAMUS$79bnThe report increases the base case orchestration CPU TAM from US$60bn to US$79bn; the bull case is US$238bn.
• TSMC 2Q26 EPS MSeNT$25.08The 2Q26 earnings preview table shows MSe EPS at NT$25.08, above consensus NT$23.89.

Impact & implications

From an investment perspective, the report is constructive on firms in the AI semiconductor supply chain with scarce capacity, advanced-node, advanced packaging, HBM, test equipment, and China AI compute power exposure. TSMC is placed at the center due to technology leadership, 2nm demand, CoWoS/SoIC expansion, and pricing power. Companies such as MediaTek, SMIC, Aspeed, Alchip, KYEC, and ASE are also included in the AI chain OW list. At the same time, the report also cautions that non-AI semiconductors may come under pressure in 2026, and higher chip costs, AI crowding out non-AI capacity, storage shortages, and regulatory limits may increase dispersion.

Risks

• AI demand or cloud capex may come in below expectations, causing demand for advanced nodes, CoWoS, SoIC, and HBM to weaken.
• TSMC has very large capex and capacity expansion plans; if client orders fall short, depreciation and utilization pressures may rise.
• Rising costs for chips, OSAT, and storage may compress margins at chip design firms.
• AI semiconductors may crowd out non-AI semiconductors for capacity and resources, leading to weaker growth in non-AI semiconductors in 2026.
• U.S. energy constraints, Chinese chip capacity constraints, regulation, and export controls may limit AI semiconductor growth.
• Advanced packaging competition, including alternatives such as Intel EMIB, may affect TSMC’s CoWoS advantage if supply-chain execution improves.

What to watch

• Whether TSMC 2Q26 actual revenue, gross margin, operating margin, and EPS come in above market consensus.
• Whether TSMC 3Q26 guidance validates AI demand, advanced-node intensity, and advanced-packaging strength.
• Whether TSMC achieves the 5%-10% 2027 price increase for leading nodes.
• Changes in N2, N3, and N5 capacity and progress of new capacity in Arizona, KaoHsiung, and Tainan.
• Whether year-end CoWoS capacity is near the path implied in the report and whether SoIC expansion pace is on track.
• CSP capex and capex-to-EBITDA changes at Amazon, Google, Microsoft, and Meta.
• NVIDIA GB200/300 rack shipments, HBM supply, and actual deployment of AI wafer consumption.
• Orders, token demand, TCO performance, and export-control changes for China AI GPUs and domestic AI accelerators.