Generative AI Fundamentals Map
Map/Loops/reinforcing

Capital flywheel

reinforcing loop · 5 steps · Trace on the map →

The cycle

Stronger modelsAdopted at work and athomeRevenueDatacentre capexMore computeR
  1. 01Stronger modelsReasoning models
  2. 02Adopted at work and at homeWorkflow redesign
  3. 03RevenueFrontier lab revenue
  4. 04Datacentre capexDatacentre capex
  5. 05More computeCompute supply

What drives it

Revenue from people and companies using models pays for the next round of compute, which trains stronger models.

What breaks it

Moody’s expects US hyperscaler capex above $700bn in 2026, about six times 2022. Bain estimates around $6tn of annual AI revenue would be needed to justify the buildout. If revenue lags, capex gets cut and the loop runs in reverse. Two balancing loops pull against it: payback and collateral, and commoditisation. Public scepticism slows the adoption step: 52% of Americans are more concerned than excited.

What to watch

AI revenue against AI capex. Enterprise spend on agents rather than seats.

Also feeds it

Sources

  1. 01Moody’s via DCD: US hyperscaler capex to top $700bn in 2026
  2. 02CIO: $6T in annual AI revenue needed to pay off data centres (Bain)
  3. 03Pew: 52% of Americans now more concerned than excited about AI (AI Weekly)