Where Are All The Data Centers?

Source: Where’s Your Ed At Date: 2026-05-12 URL: https://www.wheresyoured.at/where-are-all-the-data-centers/

Summary

Ed Zitron’s investigation argues that hyperscaler claims of gigawatt-scale data center buildouts are substantially ahead of delivered capacity. The piece focuses on Microsoft — Satya Nadella’s stated “another gigawatt this quarter” claim — and finds that flagship projects like Fairwater Atlanta and Wisconsin contain far less operational capacity than announced. Construction timelines for large facilities run 18–24+ months minimum; Zitron estimates fewer than 1 million of the ~3 million shipped NVIDIA Blackwell GPUs are deployed in live systems. The broader argument: $800B in capex commitments and $748B in Anthropic/OpenAI forward demand are predicated on infrastructure that may not exist at operational scale for years.

Implications

• AI infrastructure reality-check thread. This is the most detailed public analysis of the gap between announced and operational AI capacity. If the argument holds, the current AI revenue projections are built on phantom capacity — which matters for every product roadmap that assumes price-performance curves will follow current GPU deployment assumptions.
• Hyperscaler credibility thread. Zitron’s specific Microsoft focus is significant given Microsoft’s $13B OpenAI commitment. The piece feeds the broader narrative — visible in the W21 weekly — of a two-tier AI landscape where capability claims outrun physical infrastructure.
• Watch: Quarterly depreciation expense vs. capex announcements across Azure, AWS, and GCP as the sharpest available signal of actual deployment vs. committed spend. If depreciation doesn’t track capex with an 18-month lag, the gap is real.