AI Ledger Hepburn Advisory Hepburn Advisory
Overview / AI Companies / Compute providers

The compute bill is booked twice — once gross, once net of what the apps take back.

Who sells AI compute, and what they actually keep. Hyperscalers report an AI run-rate that includes revenue their own application products generate; the figure that describes the compute business is the one net of that carve-out. Both are shown. Neither is collapsed into the other.

Gross disclosed revenue, revenue net of the apps carve-out, and the gap between them — per provider, per quarter, with the evidence grade attached.

— —

Swipe sideways for the full table — the provider column stays put.

How these figures are built

Directional arithmetic, not a forecast. Every figure is derived at build time from dated claims held against each provider; nothing on this page is typed into the markup.

  • Gross disclosed is the AI run-rate the provider itself states. For the hyperscalers that figure includes revenue their own application products generate on their own infrastructure.
  • Net of apps carve-out removes that application revenue, leaving the compute business. The difference between the two is shown in its own column rather than being folded into either figure.
  • Rows with one figure. Most AI-native providers publish a single run-rate and no carve-out split. Those rows say so; a carve-out is never invented to fill the column.
  • As-of is a quarter, not a date. A run-rate describes a period.
  • Backlog, capacity and customer concentration are shown only where they are held. The coverage line under the table says how many rows carry each, so an empty column reads as a gap in the record rather than a zero.