Morgan Stanley is bullish on the continuation of the AI semiconductor upcycle, with TSMC remaining a core beneficiary of leading-edge processes and CoWoS

Overview

This is a Morgan Stanley industry outlook report on semiconductors in Greater China. It focuses on TSMC's 2Q26 earnings preview, 3Q26 guidance, advanced-process pricing and capacity, CoWoS/SoIC advanced-packaging expansion, cloud AI capex, AI GPU/ASIC/CPU demand, HBM and memory supply-demand dynamics, and China's AI computing supply chain. The overall tone is positive, viewing AI demand as the most important source of growth for the semiconductor industry, while non-AI semiconductors may come under pressure in 2026.

Core views

The core views include: first, TSMC continues to lead in technology roadmaps and logic density, with pricing power amid tight EUV supply and strong demand for advanced processes; second, AI computing is progressing from training and inference toward agentic AI, increasing demand for CPUs, GPUs, ASICs, HBM, CoWoS, and SoIC; third, global cloud capex remains elevated, supporting expansion of AI semiconductor TAM; fourth, China's AI computing has independent growth drivers in demand, domestic substitution, and infrastructure; fifth, segments including memory, testing equipment, CPO, and advanced packaging present opportunities from tight supply-demand conditions and increasing value content.

Analysis framework

The report uses top-down TAM estimates, supply-chain capacity tracking, cloud capex tracking, customer-demand decomposition, process and packaging roadmap comparisons, product-level supply-demand assumptions, and valuation comparisons to connect TSMC fundamentals with the AI semiconductor cycle. The analysis covers TSMC's own results, global CSP capex, NVIDIA GB200/300 racks, Google TPU, AWS Trainium/Inferentia, Broadcom/MediaTek ASIC projects, Chinese AI accelerator vendors, and memory supply-demand dynamics.

Methodology notes

• Industry cycle and TAM estimationAI semiconductor TAM and cloud capex model

  Derive long-term market potential from global semiconductor market size, AI semiconductor penetration, CSP capex, and the decomposition of GPU/ASIC/CPU demand.

  The report provides key assumptions including a potential global semiconductor market of US$1.5tn by 2030, AI semiconductor TAM of approximately US$753bn by 2030, and major CSP cloud capex approaching US$1.3tn in 2027 to support its view of long-term AI semiconductor demand.

• Supply-chain capacity trackingTSMC advanced-process and CoWoS/SoIC capacity roadmap

  Assess TSMC revenue, gross margin, and customer allocation based on N2/N3/N5 capacity, fab expansion, and year-end CoWoS and SoIC capacity.

  The report emphasizes strong demand for TSMC N2, continued N3 expansion, and a decline in N5 in 2027, while identifying advanced-packaging expansion as a key focus for 2026/27.

• Scenario analysisAgentic CPU TAM bear/base/bull case

  Estimate incremental demand for orchestration CPUs as AI progresses from inference toward action through bear, base, and bull scenarios.

  The report raises its 2030 Orchestration CPU TAM base case from US$60bn to US$79bn, with the bull case reaching US$238bn.

Asset mapping & comparison

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

• TSMC

  Core beneficiary

  Strengths

  Leading technology roadmap and logic density, strong advanced-process demand, CoWoS and SoIC expansion aligned with AI demand, and pricing power in advanced processes.

  Weaknesses

  High capex intensity and dependence on EUV, advanced packaging, and global fab execution for capacity expansion.

  Comparison

  Compared with Intel and Samsung Foundry, the report believes TSMC remains ahead in technology roadmaps and logic density.

  Risks

  EUV supply, CoWoS capacity, fluctuations in customer capex, geopolitics, and export controls.

• MediaTek

  AI Top Pick and beneficiary of the ASIC/AI PC ecosystem

  Strengths

  The report lists MediaTek as an AI Top Pick and notes that it provides a 20-core customized Grace CPU for NVIDIA N1X and participates in successive generations of Google TPU-related programs.

  Weaknesses

  Demand depends on the AI PC replacement cycle, ASIC project execution, and customer adoption pace.

  Comparison

  Compared with other AI ASIC design-service and chip companies, MediaTek is highlighted in the edge AI and custom-chip ecosystem.

  Risks

  Weaker-than-expected AI PC demand, project delays, and intensifying competition.

• Macronix

  Memory Top Pick

  Strengths

  The report lists Macronix as a Memory Top Pick and expects the NOR Flash shortage to continue through 2026.

  Weaknesses

  Memory pricing and demand are highly cyclical.

  Comparison

  It is included in the same Memory watchlist as AP Memory, Nanya Tech, Winbond, and GigaDevice.

  Risks

  Inventory-cycle reversal, limited pricing upside, and weak end demand.

• KYEC

  Beneficiary of the AI testing segment

  Strengths

  The report states that KYEC benefits from AI GPU, TPU, and CPU growth and presents AI revenue forecasts.

  Weaknesses

  Growth depends on AI chip testing demand and customer volume ramp-up.

  Comparison

  The testing equipment and consumables segment also includes Winway, MPI, Hon Precision, and Gudeng.

  Risks

  AI chip shipment timing and volatility in testing duration and equipment utilization.

• China AI accelerator supply chain

  Beneficiary of domestic substitution and local inference demand

  Strengths

  The report believes China's AI chip TAM will reach US$91bn by 2030, while domestic chips offer lower TCO and comparable per-token costs in China's AI LLM inference scenarios.

  Weaknesses

  Advanced processes, chip capacity, and ecosystem maturity remain constraints.

  Comparison

  The report focuses on Cambricon, MetaX, and Iluvatar and discusses the Hygon CPU+GPU computing platform.

  Risks

  Export controls, insufficient chip capacity, regulatory changes, and uncertainty over customer order fulfillment.

Key data

• TSMC 2026/27e capexUS$56bn / US$75bnThe report states that Morgan Stanley estimates TSMC's 2026e and 2027e capex at US$56bn and US$75bn, respectively.
• TSMC 2Q26 earnings preview2Q26 EPS NT$25.08, 3Q26 EPS NT$29.21The table shows Morgan Stanley estimates 2Q26 EPS at NT$25.08 and 3Q26 EPS at NT$29.21.
• TSMC 2Q26 gross margin and operating marginGM 67.4%, OpM 59.8%Morgan Stanley's 2Q26 estimates in the report table are above the midpoint of guidance and consensus.
• TSMC advanced-process pricingLeading-edge process pricing can increase by 5%-10% in 2027The report believes TSMC can reflect the value it provides to customers through price increases.
• TSMC N2 capacity growthApproximately 70% CAGR in 2026-2028eThe report states that TSMC indicated N2 capacity could achieve approximately 70% CAGR in 2026-2028e.
• Global semiconductor market potentialPotentially US$1.5tn by 2030The report believes AI semiconductors may contribute approximately half of this amount.
• AI semiconductor TAMApproximately US$753bn by 2030The report expects AI semiconductor TAM to reach approximately US$753bn by 2030.
• Top 4 CSPs capexUp 95% year over year in 1Q26CYThe Top 4 CSPs are Amazon, Google, Microsoft, and Meta.
• Major CSP cloud capexApproaching US$1.3tn in 2027The Morgan Stanley cloud capex tracker covers the Top 14 listed global CSPs and excludes sovereign AI.
• Orchestration CPU TAM2030 base US$79bn, bull US$238bnThe report raises the base case from US$60bn to US$79bn.
• China AI GPU TAMUS$91bn in 2030The report repeatedly emphasizes that China's AI GPU market will grow to US$91bn by 2030.
• China CPU TAMUS$42bn in 2030The report expects China's CPU TAM to rise to US$42bn by 2030.

Impact & implications

From an investment perspective, the AI semiconductor upcycle will continue to concentrate value in advanced processes, advanced packaging, HBM, ASIC design services, testing equipment, CPO, and China's domestic AI computing supply chain. TSMC is a core beneficiary because its technology leadership, CoWoS/SoIC capabilities, and customer mix directly connect it to AI demand sources including NVIDIA, AMD, Broadcom, AWS, Google, and MediaTek. However, the report also notes that cost inflation, energy constraints, chip capacity, regulation, and export controls may affect the pace of growth.

Risks

• Non-AI semiconductor demand may decline in 2026. The report explicitly states that non-AI semiconductor growth is expected to fall after excluding memory and NVIDIA AI GPU revenue.
• Rising wafer, OSAT, and memory costs may create greater gross-margin pressure for chip design companies.
• The rising priority of the AI semiconductor supply chain may crowd out non-AI semiconductor capacity and create bottlenecks in T-Glass, memory, and other areas.
• US energy constraints, Chinese chip-capacity constraints, budget pressure, and regulation are limiting factors for AI growth.
• Risks related to export controls and restricted entities may affect investment and trade activities.
• If CoWoS, SoIC, HBM, EUV, and advanced-process expansion fall short of expectations, AI hardware shipments will be constrained.
• A slowdown in cloud capex caused by CSP cash flow, capex-to-EBITDA pressure, or weaker-than-expected monetization of AI applications would affect AI semiconductor TAM.

What to watch

• Whether TSMC's actual 2Q26 results and 3Q26 guidance validate Morgan Stanley's above-consensus expectations for GM, OpM, and EPS.
• Whether TSMC's advanced-process pricing increases by 5%-10% in 2027 as expected by the report.
• Changes in TSMC N2, N3, and N5 capacity and the progress of fab expansion in Arizona, Kumamoto, Taichung, Tainan, Kaohsiung, and other locations.
• Whether year-end CoWoS capacity can approach the 200kwpm target and whether SoIC expansion becomes a future focus.
• Whether capex by the Top 4 CSPs and Top 14 global CSPs continues to maintain high growth.
• The realization of rack, chip, and CoWoS demand for platforms including NVIDIA GB200/300, Vera Rubin, AMD MI455, Google TPU, and AWS Trainium3.
• Whether tight supply-demand conditions for HBM, NAND, NOR Flash, and DDR4 continue through 2026-2027.
• Whether China's AI GPU TAM, domestic AI accelerator orders, token demand, inference TCO, and performance/cost advantages continue to improve.