Premium: The AI Compute Demand Story Is A Lie

Source: Where’s Your Ed At Date: 2026-05-04 URL: https://www.wheresyoured.at/premium-the-ai-compute-demand-story-is-a-lie/

Summary

Ed Zitron argues that reported AI compute demand is largely circular: ~70% of Microsoft’s $37B AI revenue traces back to OpenAI’s own infrastructure spend, and Amazon’s $15B AI revenue is mostly Anthropic’s AWS commitment. He estimates genuine third-party AI compute demand outside these two ecosystems at under $1B, meaning hyperscalers have collectively spent ~$78B funding the customers who generate their cloud AI revenue. The “capacity shortage” narrative, he argues, is manufactured scarcity serving infrastructure buildout justification rather than reflecting organic end-user pull.

Implications

• Feeds the sustainability thread: the circular-funding critique is the strongest public case that current inference economics are not self-sustaining. If accurate, consumption pricing models need 10–100x real-demand growth to justify the infrastructure already deployed.
• Reliability-of-demand risk for tooling vendors: tooling and orchestration layers built on the assumption of sustained hyperscaler AI investment face a single-point-of-failure if one major investor relationship unwinds.
• Counter-signal to “agents everywhere” projections: widespread agentic adoption would generate genuine incremental compute demand. The lack of that demand signal to date either means agent deployment is still early or that utilization is far below projected.