Source

49-minute interview with Chase Lochmiller (Crusoe CEO) on AI stack unit economics. Recommended by @boringbiz_ as "the deepest practical breakdown of where the bottleneck actually sits in AI infra capex." Full 9701-word transcript at inputs/podcasts/2026-05-02-crusoe-chase-lochmiller-ai-stack-economics.md.

Key signals (already woven into 4 thesis-shift events on 2026-05-02 / 2026-05-03):

• Crusoe is now building its own gas-fired peaker plants to power its data centers — operator-endorsed gen-side bull for power names (GEV, EQT, VST). Drives 2026-05-03-crusoe-gen-side-bull log entry on ai-power-bottleneck perspective.
• Eaton (ETN) thesis hardens as a bear-side stress: operator (Crusoe) builds-around the delivery layer, confirming the parasitic-load thesis. Drives 2026-05-03-crusoe-eaton-bear log entry on ai-power-delivery perspective.
• CPU bottleneck surfaces as the next constraint after GPUs are filled — implications for memory + interconnect names.
• Compute depreciation extends — Crusoe's own modeling shows GPUs holding economic value past traditional 3-year cycles, which softens the "GPUs depreciate to zero" bear case underwriting much of the AI-skeptic position.

This input is load-bearing for citations — the digest at cites the 4 thesis-shift events whose parent_event is this one. Closing the forward-ref makes the input → shift → digest → ticker-feed graph fully resol...