Type: thesis-shift (subtype: cross-thesis-confirmation)

event_id: 2026-05-04-cambrian-thesis-ipp-positioning-cross-check

Source: Lonis Hamaili — The Cambrian Thesis (2026-05-03 publish, captured 2026-05-04)

• parent_event: 2026-05-04-ai-supply-chain-2x2-ai-beta-coverage-of-15gw-substitution-risk
• announcement: @Lonis tweet

Shift: Lonis ran a horizontal slice over the entire AI supply chain — three orthogonal axes (AI beta = revenue concentration, coverage % vs 15GW additional demand, substitution risk per layer) yielding a 2x2 of 22 names. Three implications for our generation-side thesis:

1. Independent ratification of GEV — Lonis plots GE Vernova as "priced in but safe" (200%+ off-scale coverage, low substitution). This is the second operator/observer voice in 2 days flagging GEV as the gas-turbine chokepoint pick (Crusoe/Lochmiller 2026-05-02 was the first). Cross-thesis convergence on a single name without prompting.

2. IPP gap in our key_tickers list. Lonis plots CEG, VST, TLN as "priced in but safe" — the IPP cohort that our ai-infrastructure watchlist covers but that this perspective's key_tickers doesn't include. We've been heavily weighted toward the BTC-miner-pivot cohort (CRWV/CORZ/IREN/APLD/CIFR) and the BE/onsite-power layer. The "actually-already-generates-power" cohort (Constellation, Vistra, Talen) is structurally upstream. Adding CEG/VST/TLN to key_tickers — they're already on the watchlist; promoting them reflects the thesis we already implicitly hold but hadn't named.

3. OKLO bifurcation matches. Lonis plots OKLO at ~5% coverage with very high substitution risk (top-left "speculative" quadrant). This matches our existing "someday vs now" framing where we've treated OKLO/SMR-cohort as the "someday" side and BE/CRWV/CORZ as the "now" side. Same conclusion, two framings — useful confirmation that the bifurcation is the right read.

4. BE in moderate-coverage / mid-substitution. Lonis plots BE at ~70% coverage with mid-high substitution risk. Consistent with our existing position that BE is +89% 30D and the price reflects much of the onsite-power story. He's not writing it off — but he's not flagging it as sweet-spot either. The story is realized, not running ahead of itself.

5. NEW candidate ticker: CLF (Cleveland-Cliffs). Lonis includes it as steel infra for data centers + grid build-out. Not in any of our 41 watchlists. Worth tracking as a candidate; not adding to key_tickers without our own due diligence on the AI-attributable revenue %.

6. Quanta Services (PWR) — moderate coverage, mid substitution. Already in our ai-infrastructure watchlist; reflects utility-infra contractor angle. Not promoting to key_tickers; staying in watchlist.

Effect on thesis:

• Promote CEG, VST, TLN to key_tickers alongside the existing BE/CRWV/CORZ/IREN/APLD/CIFR/EQT/GEV cohort. They've been implicitly in the thesis (data center power buyers source from them); naming them makes it explicit. None are new positions — all are already on ai-power/ai-infrastructure watchlists.
• GEV reaffirmed — second independent voice in 2 days. No additional action; already added 2026-05-02.
• OKLO/SMR cohort stays "someday" — Lonis substitution-risk reading aligns. No re-grade.
• BE neutral-positive — story realized at +89% 30D; price-action consistent with Lonis's coverage read. No re-grade.
• CLF flagged for tracking — research candidate, not key_ticker. If it warrants a deep-dive, that triggers the call to add to ai-power or ai-infrastructure watchlist.
• Methodological observation — this is the first time a horizontal-slice researcher has independently corroborated key_ticker assignments we made via vertical (physics-down) analysis. Different methods, converging on similar names. See meta observation 2026-05-04-horizontal-slicing-a-7th-research-paradigm.

Per-ticker: CEG ↑ (key_ticker promotion — already on watchlist; thesis-explicit). VST ↑ (key_ticker promotion). TLN ↑ (key_ticker promotion). GEV → (reaffirm, no change). OKLO → (reaffirm "someday" framing, no change). BE → (price reflects, no change). CLF → (track as research candidate, not yet key). PWR → (in ai-infrastructure watchlist, not promoted).