The new map of chip supply looks nothing like the old one

Semiconductor manufacturing spent three decades concentrating into a handful of locations. The map being drawn now points the other way, with major capacity announced in regions that had no advanced fabrication at all a few years ago.

Subsidies moved first. Public incentive programs across several continents made it economically rational to build in places where the underlying cost structure would otherwise never have supported it, and firms responded by distributing new capacity across multiple jurisdictions rather than concentrating it where it was cheapest.

Talent pipelines have not kept pace. A fab is a building full of extremely specific expertise, and the people who hold that expertise are concentrated where the industry already is. Companies building in new regions describe importing senior staff for the first several years while local university programs spin up, an arrangement that is expensive and, several executives conceded, fragile.

The technical bottleneck has also shifted. Attention for years focused on lithography — who could print the smallest features — but capacity planners increasingly point to advanced packaging as the tighter constraint. The step that stacks and connects finished dies has less spare capacity than the step that makes them, and it is being built out more slowly.

That has knock-on effects for where value sits. A region that builds fabrication without packaging has bought a link in a chain that still runs through somewhere else, which undercuts the resilience argument that justified the subsidy in the first place. Several governments have begun revising incentive programs to weight packaging investment more heavily.

Customers, meanwhile, have adjusted their own behavior. Long-term capacity reservations that were once unusual are now standard for large buyers, and several have taken equity positions or prepaid for output years in advance. The spot market for leading-edge capacity has effectively disappeared.

Analysts covering the buildout are split on how it ends. One camp expects the new capacity to arrive into softening demand and produce a conventional glut. The other argues that the demand curve for compute has changed shape enough that the industry will absorb it. Both agree the answer will not be visible for several years, because a fab announced today produces its first volume long after the conditions that justified it have changed.