Hyperscaler customer concentration
The data
~70–80% of Microsoft / AWS AI capacity is consumed by just two customers: OpenAI and Anthropic.
The two foundation-model labs are simultaneously:
- The largest customers of hyperscaler AI compute.
- The recipients of the largest hyperscaler equity investments (MSFT → OpenAI; AWS → Anthropic).
- Building downstream products (ChatGPT apps, Claude Code) that cannibalize their own application-layer customers, accelerating those customers’ burn.
Source
Edward Zitron, 2024 reporting, Better Offline / wheresyoured.at. Surfaced through zitron-circular-economics and synthesized in bear-case-synthesis.
What it shows
A market-structure fact that gives the circular-economy thesis its sharpest edge: the “AI revenue” hyperscalers book on earnings calls is, to a striking degree, money they themselves invested via equity, flowing back as compute spend from two labs. The concentration could be a transient market structure or a permanent feature; the bear case treats it as evidence the demand side is hollow.
How to cite
Roughly 70–80% of hyperscaler AI capacity is consumed by just two customers — OpenAI and Anthropic — which are themselves recipients of hyperscaler equity investment (Zitron, 2024).
Connections
- circular-ai-economy — the loop this concentration closes.
- hyperscaler-capex-trajectory — the dollar scale of the capacity being concentrated.
- zitron-circular-economics — primary source.
- bear-case-synthesis — pillar of the bear case.
- distribution-moat — relevant inversion: the labs depend on hyperscaler distribution as much as apps depend on lab distribution.