Hyperscaler AI capex trajectory
The data
~1.7T planned additional capex for 2026–2027.
To break even on the cumulative capex, the AI line item would need to generate:
- ~$3T in net-new AI-specific revenue for break-even.
- ~$6T+ for a normal tech-sector ROI.
For reference: combined annual revenue of MSFT + META + AMZN + GOOG = ~$1.599T across all products. The required AI revenue line would dwarf the entire combined current business of the four largest tech companies.
Source
Edward Zitron, Better Offline podcast and wheresyoured.at essays, 2024–2025. Surfaced through zitron-circular-economics and synthesized in bear-case-synthesis.
What it shows
The financial scale of the capex commitment that the circular-ai-economy thesis claims is propped up by VC-recycled cash. If correct, the gap between booked capex and any realistic enterprise demand path is the most concrete bear-case fact in the system. If the circular thesis is wrong, the capex still represents a generation-defining bet on AI-specific compute as durable infrastructure.
How to cite
Hyperscalers have booked ~1.7T more planned for 2026–2027, requiring ~$3T in net-new AI revenue to break even (Zitron, 2025).
Connections
- circular-ai-economy — the financial-loop concept this data sizes.
- hyperscaler-customer-concentration — who the capex is currently serving.
- perplexity-burn-ratio — the application-layer side of the same loop.
- zitron-circular-economics — primary source.
- bear-case-synthesis — pillar of the bear case.
- ai-factory-huang — the bull-frame interpretation: this capex is the AI factory.
- goldman-sachs-too-much-spend — middle-position skeptic on the same trajectory.
- ai-capex-cycle — the boom-then-bust framing this data sizes.