Description
This report maps the global AI-infrastructure build-out — a 66× scaling of critical IT capacity from ~2.3 GW to 150 GW by 2030 — and resolves it through two questions: who pays for the silicon, and who runs the workloads?
Every figure is built bottom-up from Structure Research’s proprietary dataset — 14 per-company trackers plus a 38-operator neocloud and sovereign rollup — and held to a hard conservation constraint where owned capacity equals offtaker capacity, so supplier-side and demand-side claims reconcile rather than double-count. The result is a single, consistent owner-versus-offtaker view of where capacity, capital and compute actually concentrate.
What it addresses:
- How fast, how large, and where the power wall binds
- Who owns capacity versus who consumes it
- Which deal structures convert capital into compute — a 45× range
- How Microsoft’s pivot and the labs’ opposing playbooks reshape demand
- Whether neoclouds, sovereigns and custom silicon hold or compress
