Morgan Stanley Bullish on AI Semiconductors, Market Size Could Reach $1.5 Trillion by 2030
Taylor Wilson
Morgan Stanley's June 22 deep dive forecasts the global semiconductor market at $1.5 trillion by 2030, with AI chips contributing $753 billion at a 38% CAGR from 2023 — This means AI is becoming the industry's dominant growth engine, not just a side story.
Why are cloud giants spending at record pace?
Amazon, Google, Microsoft, and Meta grew Q1 2026 capex by 95% year-on-year. Morgan Stanley estimates full-year cloud capex at $685 billion — 10% above consensus.
This means → cloud companies are plowing roughly half their operating profit (capex-to-EBITDA ≈ 50%) straight into infrastructure — the highest commitment on record.
In plain terms = for every two dollars these companies earn, one goes to chips and data centers. They are voting with real money: AI is not a bubble.
Why are custom chips gaining ground?
Google's TPU has reached its sixth generation. AWS launched Trainium3. Meta and Tesla are also accelerating in-house chip programs.
This means → cloud hyperscalers no longer rely solely on Nvidia's general-purpose GPUs — they want chips optimized specifically for their own models (ASICs, or custom silicon).
This reflects a deeper shift: AI competition has entered an efficiency war — whoever converts electricity into inference results most cheaply wins.
How much does TSMC stand to gain?
Morgan Stanley forecasts TSMC's AI semiconductor revenue growing at a 60% CAGR, exceeding 30% of total revenue by 2026.
Advanced packaging — CoWoS, a technology that tightly integrates multiple chips on one substrate — is projected to reach 200,000 wafers per month by 2027. AI compute wafer consumption in 2026: $27 billion.
High-bandwidth memory (HBM, ultra-fast memory paired with AI chips) consumption is forecast at 32 billion Gb. Nvidia remains the largest buyer.
In plain terms = no matter who designs the chip, TSMC builds it. TSMC is the shovel-seller in this AI arms race.
AI PCs and edge devices — is a new upgrade cycle coming?
Nvidia and MediaTek jointly developed a next-gen AI PC chip (RTX Spark / N1X) with a 20-core custom Grace CPU. Mass production by Asus and MSI is expected in Q3 2026.
Nvidia's Vera CPU delivers 1.7× the compile performance and 1.9× the Python performance of the top x86 CPU.
This means → AI is moving off the cloud and onto the laptop in your hands. Morgan Stanley estimates the addressable market for Agentic CPUs — processors that run AI agents locally — could reach $238 billion.
Can China's domestic AI chips catch up?
Morgan Stanley projects China's AI GPU market at $91 billion by 2030, with domestic self-sufficiency reaching 70%.
Running inference on large models like DeepSeek R1, domestic AI accelerators may achieve a total cost of ownership (TCO — the all-in cost of chips, power, and maintenance) 30–60% lower than Nvidia's China-export variants.
Top picks: Cambricon (寒武纪), leading on inference performance and customer expansion, and Iluvatar (天数智芯), standing out on supply-chain resilience and strong order visibility.
In plain terms = U.S. export restrictions have inadvertently accelerated a domestic alternative supply chain — one that is already beating Nvidia's restricted chips on cost.
Memory and test equipment — who are the underappreciated winners?
AI demand has already caused a NAND shortage. NOR Flash undersupply is expected through 2026; DDR4 shortages may persist into H2 2026.
The test equipment market is forecast to grow at a 35% CAGR from 2024 to 2027. Morgan Stanley highlights Hon Precision (弘塑), MPI (旺矽), and WinWay (颖崴).
Core overweight picks: MediaTek (top pick), TSMC, SMIC, Macronix (top pick), NAURA, AMEC.
This means → the AI wave benefits far more than chip designers — from memory to test equipment, the entire supply chain is being pulled forward.
Content is for reference only, not financial advice.