Robust Cloud Capex; Positive AI Semiconductor Outlook Maintained
AI summary card
Robust Cloud Capex; Positive AI Semiconductor Outlook Maintained
Morgan Stanley expects demand for AI semiconductors, advanced packaging, and domestic Chinese computing power to expand continuously from 2026 to 2030, and recommends focusing on segments with technology and supply-chain advantages.
- Cloud AI semiconductor TAM could reach US$485 billion in 2026 under a bull-case scenario, while AI semiconductor TAM could be approximately US$753 billion by 2030.
- Capex by the four largest cloud service providers—Amazon, Google, Microsoft, and Meta—grew 87% year over year in the second quarter of 2026, while capex by 14 listed global cloud service providers could approach US$1.4 trillion in 2027.
- TSMC maintains leadership in advanced process technologies and advanced packaging; the report forecasts its capex at US$62 billion and US$75 billion in 2026 and 2027, respectively.
- China AI chip TAM is projected to increase to US$91 billion by 2030, with domestic chips competitive in total cost of ownership and per-token costs for inference applications in China.
- Memory supply and demand are tightening. The report expects AI memory demand to lead to NAND shortages, tight NOR Flash supply to persist through 2026, and DDR4 shortages potentially to continue into the second half of 2026.
Report interpretation
Overview
This report provides an outlook for global and Chinese cloud computing capex, AI semiconductor demand, and the Greater China semiconductor supply chain. Its core view is that cloud service providers will continue increasing investment in computing power, driving demand for GPUs, ASICs, advanced process technologies, advanced packaging, HBM, memory, and testing equipment; industry growth will concentrate significantly in AI-related segments.
Core views
The report believes the medium-term growth outlook for AI semiconductors is strong, supported by rapid global cloud capex growth. With leading process technologies, logic density, and constrained EUV supply, TSMC remains a key beneficiary in the high-end AI chip and CoWoS supply chains. Custom chips will continue to expand within cloud service provider ecosystems, but NVIDIA remains the primary driver of HBM and high-end AI computing demand. The China market is supported by inference demand, improving domestic supply-chain capabilities, and the cost advantages of domestic chips.
Analysis framework
The report combines cloud service provider capex tracking, supply-chain capacity and shipment data, wafer and advanced packaging consumption, power deployment, chip roadmaps, memory supply-demand models, and valuation comparisons for top-down analysis and supply-chain cross-validation.
Methodology notes
AI Semiconductor Total Addressable Market
Projects the market size of AI semiconductors and chip subsegments based on cloud capex, computing deployment, and supply-chain data.
Advanced Packaging and Memory Supply and Demand
Tracks CoWoS, HBM, wafer, and memory supply expansion against customer demand to assess bottlenecks and beneficiary segments.
Cloud AI Semiconductor Demand
Builds a bull-case scenario using supply-chain data to assess the upside potential of the cloud AI semiconductor market.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- TSMCCore Supplier of Advanced Process Technologies and Advanced Packaging
- Strengths
- Leading technology roadmap and logic density; advanced process technologies, CoWoS, and SoIC are driven by AI demand, providing leading-node pricing power.
- Weaknesses
- High capex and demanding capacity-expansion execution requirements, with relatively high customer concentration.
- Comparison
- The report believes TSMC retains significant advantages due to its technology leadership and EUV supply constraints; Intel EMIB has potential to support large chips, but supply-chain execution remains critical.
- Risks
- AI demand falling short of expectations, delays in advanced packaging capacity expansion, geopolitical risks, and slower customer capex.
- NVIDIACore Driver of AI GPU and HBM Demand
- Strengths
- Dominant in AI computing and HBM consumption; volume ramp-up of GB200/300 racks drives upstream wafer and advanced packaging demand.
- Weaknesses
- Highly dependent on high-end packaging, HBM, and system supply chains.
- Comparison
- Custom ASICs continue to expand, but the report still views NVIDIA as the primary source of AI chip demand.
- Risks
- Cloud service providers shifting toward internally developed ASICs, supply-chain constraints, export controls, and changes in customer capex.
- China Domestic AI Chip Supply ChainBeneficiary of China AI Computing Power Localization
- Strengths
- Offers lower total cost of ownership and per-token costs in inference applications, while domestic capacity and system integration capabilities continue to improve.
- Weaknesses
- Advanced process technologies, key equipment, and high-end memory supply may remain constrained.
- Comparison
- Compared with high-end overseas processors, domestic chips offer more competitive cost performance for local deployment in China, though ecosystem development and technological maturity require continued validation.
- Risks
- Chip capacity constraints, regulatory and export restrictions, software ecosystem adaptation, and intensifying market competition.
- Memory ChipsSupporting Beneficiary Segment of AI Infrastructure Expansion
- Strengths
- AI memory demand supports improving supply-demand conditions for NAND, NOR Flash, and DDR4.
- Weaknesses
- High product price volatility and pronounced cyclicality.
- Comparison
- AI-related memory demand is more resilient than traditional end-market demand, though the degree of benefit varies across memory categories.
- Risks
- Supply expansion outpacing demand, spot and contract price volatility, and weak end-market demand.
Key data
- 2026 Cloud AI Semiconductor TAM, Bull CaseUS$485 billionBased on bull-case assumptions using supply-chain data.
- 2030 AI Semiconductor TAMApproximately US$753 billionReport forecast.
- Year-over-Year Growth in Capex of the Four Largest Cloud Service Providers in 2Q 202687%Covers Amazon, Google, Microsoft, and Meta.
- Global Cloud Capex in 2027Approaching US$1.4 trillionCovers 14 listed global cloud service providers, excluding sovereign AI.
- TSMC 2026/2027 Capex ForecastUS$62 billion/US$75 billionReport forecast.
- AI Semiconductor Revenue as a Share of TSMC 2026 RevenueMore than 30%Report forecast.
- 2030 China AI Chip TAMUS$91 billionReport forecast.
- 2030 China CPU TAMUS$42 billionReport forecast.
- Agentic CPU TAM CAGR from 2026 to 2030251%Estimated by Morgan Stanley's top-down model.
Impact & implications
The investment implication is that rising AI capex will first benefit advanced process technologies, CoWoS/SoIC, HBM, custom ASICs, testing equipment, and key materials. By comparison, if memory and NVIDIA AI GPU revenue are excluded, non-AI semiconductor growth could weaken in 2026. Industry allocation should therefore emphasize AI exposure, technological barriers, and supply-chain pricing power.
Risks
- Cloud service provider capex may fall short of expectations or slow due to budget, energy, and regulatory constraints.
- Capacity expansion for AI chips, CoWoS, HBM, and key materials may underperform expectations, limiting deliveries.
- Weakening non-AI semiconductor demand may weigh on overall industry conditions and supply-chain utilization.
- Rising wafer, packaging and testing, and memory costs may compress chip design company margins.
- China-U.S. technology competition, export controls, and geopolitical risks may affect China AI chip supply and demand.
- The development pace of custom ASICs or changes in the competitive landscape may affect the allocation of demand for GPUs and related supply chains.
What to watch
- Quarterly capex of major cloud service providers, capex as a percentage of EBITDA, and progress in AI infrastructure deployment.
- Rack shipments, supply-demand conditions, and customer mix for NVIDIA GB200/300 and subsequent platforms.
- Monthly capacity expansion, pricing, and utilization for TSMC advanced process technologies, CoWoS, and SoIC.
- Supply-demand conditions and changes in spot and contract prices for HBM, NAND, NOR Flash, and DDR4.
- Volume ramp-up schedules for custom ASIC projects such as Google TPU, AWS Trainium/Inferentia, and Meta.
- Market share of domestic Chinese AI accelerators, inference token demand, cloud capex, and progress in localization.