Q1 — Where is the money going?

The hyperscalers and frontier labs are spending on the order of three-quarters of a trillion dollars a year on the AI buildout. What does that actually buy — chips, land, power, cooling, networking, labor — and how much of it is the industry paying itself?

This is a live map, not a snapshot: every node and edge below is read straight out of the underlying knowledge graph each time the site publishes, not retyped by hand. It tracks named entities (companies, funds, joint ventures, financing rounds) as nodes, each split by activity where relevant — a company’s capital-raising facet is a different node from its data-center-construction facet, because money and physical buildout are different questions — connected by dated, sourced edges: who paid whom, how much, for what, and how sure we are of it.

Click any node to see everything it connects to — the amount, what kind of transaction it was, what it bought, and a link to the primary source. Search to jump straight to an entity. The core-buildout toggle highlights the named roster this map treats as “the group” for drawing an inside/outside spending boundary (filters/cut-core-buildout.yaml in the underlying repo) — everyone else on the map is a capital provider, supplier, or government sitting outside that boundary by design.

Every dollar figure is graded by how it was sourced — a primary SEC filing or company release fetched directly scores highest; a figure relayed through a news aggregator scores lower and is marked as such in the underlying record, visible in each node’s detail panel.

This page reads live from graph/{atoms,relationships,sources}.jsonl via graph/export_q1.py, as of 2026-08-27 — the first cutover off the hand-maintained snapshot this page used to be. Q2 (who’s buying the inference), Q3 (the datacenter census) and Q4 (governance) are designed but not yet built out on the site.