Global AI Semiconductor Market to Reach $1.5 Trillion by 2030, Focusing on TSMC and Chinese Chips

Overview

This report deeply analyzes the core components of future artificial intelligence infrastructure, including CPUs, GPUs, ASICs, optical communication, and the Chinese chip supply chain. Morgan Stanley believes that, as AI evolves from training to inference, and cloud capital expenditures continue to grow, the AI semiconductor market will experience a structural boom. The report focuses on TSMC’s monopoly position in advanced packaging, the surge in global cloud service providers’ demand for custom ASIC chips, and China’s accelerating AI chip localization trend amid geopolitical dynamics. The overall outlook is optimistic, recommending attention to leading companies with technological moats and capacity advantages.

Core views

AI semiconductor market poised for explosive growth. The report forecasts the global semiconductor market to reach $1.5 trillion by 2030, with approximately half ($753 billion) from AI semiconductors. This growth is mainly driven by increased cloud service providers’ capital expenditure, expected to reach $685 billion by the top 10 global cloud providers by 2026. Excluding memory and NVIDIA AI GPU revenues, non-AI semiconductor growth is expected to decline by 2026, showing resource concentration in AI. TSMC’s advanced packaging capacity expands continuously. Given strong AI demand, TSMC may expand CoWoS capacity to 165,000 wafers per month by 2027. CoWoS and SolC capacity are expected to double by 2025 and continue expanding into 2026. AI compute wafer consumption could reach $28 billion in 2026, with NVIDIA holding most of the share. Customer demand for TSMC’s 2nm and 3nm nodes is also mainly AI chip related. China’s AI GPU market localization accelerates. The report estimates China’s AI GPU total addressable market (TAM) could grow to $67 billion by 2030. Despite technical constraints, China’s infrastructure strength has narrowed perceived technology gaps. Domestic chips exhibit stronger dollar-performance competitiveness due to significant price reductions. The emergence of low-cost inference models like DeepSeek further drives inference AI demand, benefiting local vendors such as Cambricon, MetaX, and Iluvatar. Storage and testing equipment face structural opportunities. AI storage usage leads to NAND shortages; NOR Flash supply constraints are expected to last until 2026, and DDR4 shortages will continue into late 2026. For testing equipment, increased testing time benefits companies like Hon Precision and MPI as key beneficiaries of the structural trend. WinWay plans to expand its pin capacity from 3.5 million per month in 2025 to 9 million in 2026. New developments in ASIC and optical communication. Beyond NVIDIA GPUs, cloud providers still require custom chips to optimize cost and performance. Giants such as Google, Meta, and Microsoft are advancing TPU and Trainium ASIC projects. In optical communications, CPO (Co-Packaged Optics) technology will scale in large-scale switch applications after 2027, driving demand for related components.

Analysis framework

The report adopts a combination of top-down and bottom-up analytical approaches. First, from a macro perspective, it estimates global cloud capital expenditures and AI semiconductor total addressable market (TAM) to establish the industry growth ceiling. Then, it analyzes supply chain details deeply—TSMC’s CoWoS capacity, wafer consumption, and major customers’ (like NVIDIA, Google) specific demand data—to verify supply-side matching. For the Chinese market, it evaluates localization feasibility and economics by comparing nine factors between Chinese and US AI chips in performance, cost (TCO), and infrastructure. Finally, it segments niche areas like storage, testing equipment, and optical modules to uncover specific investment opportunities through supply-demand gaps and technology evolution paths (such as CPO).

Methodology notes

• Industry/sector analysis frameworkSupply and demand framework

  Judges supply chain bottlenecks and opportunities by analyzing the matching degree between TSMC’s CoWoS capacity expansion and AI chip demand from major cloud vendors.

  The report identifies tightness in the advanced packaging segment by comparing supply (TSMC capacity) against demand (cloud vendors’ orders), which derives investment value for relevant equipment and foundry companies.

• Industry/sector analysis frameworkVolume-price decomposition

  Breaks down the AI semiconductor market into training versus inference, cloud versus edge, general GPU versus custom ASIC dimensions for analysis.

  The report not only examines total market size but also dissects different application scenarios and technological paths for AI chips, pointing out inference demand growth and ASIC customization as key future incremental sources.

• Competition and strategy frameworkMoat / competitive advantage

  Analyzes the competitive advantages of Chinese AI chips in the domestic market, especially the dollar-performance cost advantage.

  Though technologically possibly lagging in absolute terms, the report emphasizes that Chinese chips’ large price cuts create competitiveness in price-performance (TCO), constituting a key moat for survival and expansion in the local market.

Asset mapping & comparison

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

• TSMC (台积电)

  Beneficiary

  Strengths

  Monopoly in CoWoS advanced packaging capacity, leading 2nm/3nm technology

  Comparison

  More mature CoWoS ecosystem compared to Intel's EMIB

  Risks

  Capacity expansion below expectations

• MediaTek (联发科)

  Beneficiary

  Strengths

  AI PC and smartphone chip portfolio, top pick

• Cambricon (寒武纪)

  Beneficiary

  Strengths

  China’s AI GPU leader, benefiting from local substitution

  Comparison

  Peers of MetaX and Iluvatar as focus stocks

  Risks

  Geopolitical risks, technological iteration pressure

• ASMPT

  Beneficiary

  Strengths

  Backend equipment leader, benefiting from advanced packaging demand

  Comparison

  Superior to Chinese OSAT vendors

• WinWay / MPI / Hon Precision

  Beneficiary

  Strengths

  Testing equipment and consumables, benefiting from longer testing times

Key data

• Global semiconductor market size in 2030$1.5 trillionWith about 50% from AI semiconductors
• AI semiconductor TAM in 2030$753 billionOptimistic assumption in the report
• Global cloud capital expenditure in 2026$685 billionSum of top 10 listed cloud service providers
• TSMC CoWoS capacity in 2027165,000 wafers/monthExpected expansion target
• China AI GPU TAM in 2030$67 billionExpected market size
• AI compute wafer consumption value in 2026$28 billionMajority by NVIDIA

Impact & implications

For the industry, rapid AI semiconductor growth will reshape supply chain patterns, with advanced packaging and high-bandwidth memory as key bottlenecks. For equipment and technology leaders like TSMC, this implies long-term capacity utilization and pricing power. For the Chinese semiconductor industry, despite external restrictions, massive local inference demand and policy support provide growth opportunities for companies like Cambricon and Hygon. Investors should focus on irreplaceable AI infrastructure segments with capacity expansion potential, such as testing equipment, optical modules, and local foundry chains.

Risks

• Further tightening of US export control policies
• AI demand growth below expectations and cloud vendors cutting capital expenditures
• Failure in technological iteration or slow yield improvement
• Geopolitical conflicts impacting supply chain stability

What to watch

• Progress of TSMC’s CoWoS capacity expansion
• Shipment volume and customer adoption of Chinese AI GPU vendors
• R&D and mass production progress of cloud service providers’ ASIC chips
• Price fluctuations and supply-demand changes of storage chips