• Type: thesis-shift (subtype: confirms)
• event_id: 2026-05-02-crusoe-genside-bull-confirmed
• Source: Crusoe / Chase Lochmiller AI stack economics (49-min talk hosted by @apoorv03; full transcript at inputs/podcasts/_transcripts/2026-05-02-crusoe-chase-lochmiller.txt)
  • parent_event: 2026-05-02-crusoe-lochmiller-transcript
• Shift: Three operator-side validations of the generation-side thesis from a major hyperscaler customer:
  1. GE Vernova explicit bull. Lochmiller (CEO Crusoe) on record: "It's been good to be a shareholder of GE Vernova." Gas turbine pricing has roughly 3x'd ($1M/MW → ~$2-3M/MW) and GEV is the U.S. champion in that supply.
  2. Gas turbines + onsite generation are the unlocking layer for AI capex. Crusoe's own data center model is built on onsite gas-turbine generation; grid timelines are too long for the buildout cadence the workload demands. Direct validation of the BE / CRWV / CORZ / IREN / APLD "onsite power = first-mover" frame.
  3. Compute depreciation likely longer than 6 years. Public-company AI capex models standardize on 5-6 year compute depreciation. Lochmiller pointed at a Bloomberg chart showing H100 spot pricing has actually risen since launch — counter to the "next gen makes old gen worthless" assumption. Same trend in SemiAnalysis Blackwell pricing data. Agentic-workflow demand keeps older silicon valuable. If depreciation is 8-10 years, AI cloud unit economics re-rate meaningfully across the full cohort.
• Effect on thesis:
  • Strengthens GEV-as-thesis-stock case. Adds GEV explicitly to key_tickers; bumps confidence on existing ai-power watchlist position. Cross-check next refresh against gas-turbine 3x pricing story vs GEV investor day numbers.
  • Reinforces onsite-power vs SMR bifurcation. Already flagged in 2026-05-01 NuttyCLD Bloom side story. Crusoe's operator commentary independently validates: BE / CRWV / CORZ / IREN / APLD are the "now" trade; OKLO / SMR are the "someday" trade.
  • New calibration question: what depreciation schedule do the AI infra cohort use? Worth a coverage-audit pass — pull 10-K language for BE, CRWV, CORZ, IREN, APLD, NBIS, MSFT, GOOGL, ORCL on compute depreciation. If most use 5-6 years and Crusoe's read is right, there's an upside re-rating waiting.
• Per-ticker: GEV ↑ (explicit operator endorsement + 3x gas turbine pricing chokepoint). BE neutral-positive (onsite gen frame validated; already +89% 30D — the price reflects much of this). CRWV, CORZ, IREN, APLD ↑ (onsite-power pure-plays validated by operator-side voice). EQT neutral (nat-gas-supply leg of the thesis still soft per April price action; Crusoe doesn't address fuel commodity directly). HUT, CLSK, RIOT, BTDR ↑ (BTC-miner→AI conversion as onsite-power play continues to validate). PSIX neutral-positive (distributed-power-gen exposure aligns).

Pre-conversion frozen state (informational)

Pre-2026-05-03 thesis, sources, and watch-fors live in production at 2026-05-01-ai-power-bottleneck. That file is frozen at its 2026-05-01 last-refresh state per the research/ isolation rule — no retroactive edits. The README.md here is the current projection (post-Crusoe); the production file is the snapshot as-of conversion.

Original spark (informational, not a log entry):

• 2026-04-15 — Aschenbrenner Q4 2025 13F surfaces watts-per-rack thesis; perspective drafted around BE / CRWV / CORZ / IREN / APLD / EQT cohort.
• 2026-05-01 — NuttyCLD AI Power Crisis 4-part series + 2-part company sequel + Bloom side story (March 30 → May 1). Thesis escalated, layered "30% parasitic loss" sub-thesis, sister perspective ai-power-delivery forked off.

The granular pre-conversion history isn't backfilled here. If we ever decide to mass-backfill (deferred per master brief Stage 5), the prior shifts get standalone event files generated from production prose.

🔄 2026-06-25 full-scan tape check: STALE-RIGHT/mixed — core (electrons+equipment over GPUs) intact, BE strong-up RSI 55 and GEV/VST/TLN/CORZ holding, but the neocloud-miner sub-leg rolled (IREN -18.7% 30D RSI 40, CRWV -42% off high RSI 44) and EQT/CEG/PSIX broke; no watch_for trigger fired. (Data tables spliced fresh to the 2026-06-25 close; full cross-scan read in 2026-06-25-full-scan-market-brief §6.)