07 · Outlook & Risks

Toward $1 trillion — and the things that could derail it

The base case is strong: a trillion-dollar market in 2026 and the long-touted $1T+ trajectory toward 2030, powered by AI. But this cycle is unusually concentrated, which makes it unusually fragile. Here's the outlook and the risks.

7.1

The forecast

Analysts broadly agree on direction if not magnitude: 2026 lands near $1T, with the industry widely expected to reach $1.4–1.5T+ by the end of the decade. The composition matters as much as the size — AI logic and memory take an ever-larger share, while mature segments grow steadily.

Path to a trillion-dollar market

Global sales, US$ billions (actual + forecast/illustrative)
Source: WSTS, SIA; longer-horizon points are illustrative of consensus
Base case 2026 ≈ $975B (+25%), clearing or nearing $1T. AI capex stays elevated; memory super-cycle runs hot into 2027.
Bull case AI demand and ASP gains overshoot; HBM/CoWoS capacity additions sell out; logic re-rates further. $1T arrives cleanly in 2026.
Bear case An AI capex digestion / over-build, a memory price reversal, or a Taiwan-related shock pulls the cycle down sharply given the concentration.
7.2

The risk radar

Five risks dominate. Each maps to a concentration the industry has not yet diversified away.

1 · Geographic concentration

~70% of leading-edge logic and the bulk of advanced memory sit in Taiwan and Korea. Any disruption in the Taiwan Strait would be a global systemic event with no near-term substitute.

2 · AI demand digestion

If AI infrastructure spend pauses or efficiency gains cut chip intensity, the narrow, AI-driven up-cycle could reverse quickly — memory most of all.

3 · Policy whiplash

The Information Export rules and tariffs are shifting fast — H200 sales were blocked, then allowed, then tariffed, while Beijing tells its own firms to halt orders. Demand can be rerouted by decree overnight.

4 · Memory cyclicality

Memory has always boomed and busted. A super-cycle that pulls capacity forward sets up a potential glut once HBM/DRAM additions land in 2027–28.

5 · Packaging & power bottlenecks

CoWoS and HBM are the true constraints on AI output, and data-center electricity is an emerging ceiling. Capacity, not design, gates the build-out.

6 · Re-shoring economics

CHIPS-funded U.S. and EU fabs improve resilience but at higher cost; ramp delays (e.g. some 2026→2027 slips) test the return on hundreds of billions in subsidy and capex.

7.3

What to watch in 2026

Methodology & sources

Every figure on this site is sourced. See the full source list and notes on how estimates were assembled.

View sources & methodology