S01.7 · Technology & Digital Infrastructure
A $796B chip and component market that grew 26% on AI and memory demand, dominated by a handful of foundries and IDMs.
Semiconductors & Electronic Components covers chip design, fabrication and the equipment supplying both. WSTS put the market at $796B in 2025, up 26% year over year, with a forecast approaching $1T in 2026. The growth is AI- and memory-led rather than broad-based, and the structural fact underneath it has not changed: leading-edge capacity is scarce, costs $10-20B a fab to build, and sits in very few hands. In this segment, capital is the moat.
Outside the current AI cycle, growth runs high single to low double digits, with memory pricing the swing factor — memory has always been the most cyclical piece of this market, AI demand notwithstanding. Revenue is global; leading-edge fabrication is not. It concentrates in Taiwan (TSMC) and South Korea (Samsung, SK Hynix), with equipment concentrated in the Netherlands (ASML), the US and Japan. The top ten — TSMC, Samsung, Nvidia, Intel, SK Hynix and Broadcom among them — hold most of the value in the chain, and TSMC holds the leading-edge foundry market outright.
The economics bifurcate by business model, and the bifurcation is the analysis: fabless design is high-margin and capital-light; manufacturing is brutally capital-intensive, with margins that swing on the cycle. They are different businesses that happen to share an end product, and they should be read separately. Upstream sit ASML, Applied Materials and Lam Research plus specialty materials suppliers; downstream, OEMs, hyperscalers and systems integrators. Regulatory gating is heavy and compounding — export controls, CHIPS Act incentives with conditions attached, CFIUS-style national-security review — and increasingly determines where capacity goes and who may transact in it.
IDMs and foundries are extending into advanced packaging and design-IP licensing. Packaging is where the new demand lands: AI accelerators need multi-die, high-bandwidth-memory configurations, and the packaging step captures that spend. Reshoring, by contrast, buys insurance against export-control and geopolitical risk; it adds capacity, not demand.
Agentic AI is automating design verification and yield analysis — design-engineering work that has been among the largest fixed costs of bringing a chip to market. It also thins a moat few outside the industry think about: secrecy. AI-assisted reverse engineering makes it easier to infer a competitor's process and design choices, eroding protection that leading-edge nodes have relied on.
Structural change here moves at fab speed, not software speed. Capacity lead times and fab economics put the horizon for AI-driven structural change at five to ten years — double the software norm — even while AI is simultaneously the single largest source of current demand. Both things are true, and the gap between them is where the cycle risk lives.