Is the semiconductor industry headed for an almighty crash?
Chip sales grew faster year on year in the first half of 2026 than at any time since the 1980s; TSMC posted a sixth straight quarter of record profit and Samsung a fourth. Yet a leading chip index has fallen more than 10% since 6 October, and Hock Tan said on 12 October that many AI investors “have no clue”. The industry has consolidated sharply: about 25 firms could make leading-edge processors in the early 2000s, four can today, and TSMC makes nearly all high-end ones; Samsung, SK Hynix and Micron make almost all advanced memory. TSMC capex is expected at about $62bn this year and $76bn next, falling as a share of revenue and funded from cash flow. DRAM costs about six times early-2023 levels. Nvidia has committed over $250bn to secure supply to January 2029, and Micron says take-or-pay contracts lock in 35% of revenue to 2030. The Economist’s judgement is that a bust, if it comes, would be milder than the boom because consolidation and long-term contracts shift risk onto buyers—but it still expects prices, profits and share prices would fall and contracts would be renegotiated.
Why it matters
Semiconductors are the bottleneck and bellwether of the AI cycle. Whether the downturn is mild or brutal shapes AI-infrastructure valuations, the exit window for chip and hardware startups, and how much risk sits with buyers such as neoclouds and labs that signed take-or-pay deals.