The AI Demand Bubble

Source: Where’s Your Ed At Date: 2026-08-04 URL: https://www.wheresyoured.at/the-ai-demand-bubble/

Summary

Ed Zitron’s bear-case essay argues that hyperscaler AI revenue is dangerously concentrated in two unprofitable customers, OpenAI and Anthropic, who are simultaneously the hyperscalers’ own investees — Amazon, Google, and Microsoft fund the labs while booking the labs’ compute spend as revenue growth. He cites Anthropic and OpenAI as roughly 70-75% of Amazon/Microsoft’s AI revenue and over 48% of projected 2027 Google Cloud revenue, alongside a roughly 26-to-1 mismatch between Amazon’s 2026 capex and its non-AI-lab AI revenue, and concludes there is no broad third-party demand behind the infrastructure buildout.

Implications

• Feeds the AI-economics / demand-bubble watch thread directly — a sharpened bear case naming specific circular-financing numbers (hyperscaler-invests-in-lab, lab-spends-on-hyperscaler-compute, hyperscaler-books-it-as-revenue) rather than a general “bubble” gesture.
• Complements the token-cost-as-operating-cost thread tracked at the model layer (GPT price cuts, cache-hit economics surfacing in CLI UIs): those cost pressures exist on top of a revenue base this essay argues is structurally thin and concentrated in two counterparties.
• A useful counterweight to keep in the loop’s model whenever a lab announces a large compute commitment or price cut — the underlying demand question this essay raises is independent of any single vendor’s release cadence.