AI datacenter power & cooling
Board · 1.0 · Jun 2026
STRAT·Infra·vintage Jun 2026

AI datacenter power & cooling

The binding physical ceiling on the AI buildout — grid interconnect queues, transformer/turbine lead times, and the liquid-cooling cutover, priced as toll-booths.

The call

Own the deep-physics bottlenecks of the AI power crunch — firm generation (CEG/VST), turbines (GEV), and grid gear (ETN/PWR/HUBB) — where lead-time scarcity is durable pricing power; rent the compute/cloud layer, don't own it.

The AI power crunch has flipped the value-chain: with US data-center load heading to ~76 GW in 2026 (from ~50 GW in 2024, IEA/Bloom Energy 2026) against a grid that takes 100+ weeks to deliver a transformer and slots gas turbines out to 2030, the binding constraint is no longer compute — it is firm, fast electrons and the gear that moves them. Margin and pricing power are migrating DOWN the stack into the scarce physical layers (generation, turbines, switchgear, transformers) and away from the abundant top (GPU-cloud rental, where supply floods in and depreciation eats the spread). House call: overweight the toll-booth physical layers with multi-year backlogs and sold-out capacity; underweight/avoid the commoditizing neocloud layer where the same AI demand funds the competitors that erode your rent.

11stack layers
10shift points
22opportunities
14name theses
$10Mbook
The power-equipment dominance call

No single layer dominates the way a foundry dominates the chip stack — dominance is distributed across the scarce physical chokepoints, each an oligopoly. GE Vernova effectively dominates heavy-frame gas turbines (with Siemens Energy and Mitsubishi the only alternatives; all sold out), giving it the most durable single-layer pricing power on the board through ~2030. Constellation + Vistra dominate dispatchable nuclear/firm baseload for hyperscaler PPAs, and Eaton/Hubbell/Quanta dominate grid electricals + buildout. This distributed-oligopoly structure is MORE durable than a single chokepoint because each layer is independently capacity-gated by capital, permitting, and skilled labor — none can be bypassed and all are multi-year sold out. The fragile layer is neocloud, where dominance is illusory (capital, not technology, is the moat) and competition is hyperscaler-funded.

Not investment advice — analyst work product for a qualified professional. Bull / base / bear with probabilities, every position carries a falsifier.· src wf:power-map/wf:power-verify/wf:power-book
Phase A · the map

Full-stack value chain

The power & cooling delivery chain from the AI campus load down to grid-scale generation. The marker shows where the AI power crunch is a tailwind, headwind, or mixed for each layer’s margin pool. Tap any layer for the full read.

↑ datacenter loadCrunch: hurt · mixed · helps
↓ generation
AI power-crunch impact
Crunch headwind
Mixed
Crunch tailwind
binding bottleneck
Where power scarcity defends
8of 11 layers

The supply-locked layers — gas turbines, HV transformers, switchgear, liquid cooling — get stronger as the AI power crunch bites, because multi-year backlog and physics, not capital, set the price. Commodity utilities and pure-play SMRs do not.

Sign in, then upgrade to Pro to see the full value chain.

Phase B · 22 names

Opportunity board

22 names sorted into three buckets — durable compounders, undervalued / high-potential, and short / avoid. Each tile shows its conviction; open one for the thesis, catalyst, and falsifier.

Durable compounder

8

Undervalued / high-potential

10

Short / avoid

4

Sign in, then upgrade to Pro to see the full opportunity board.

Members only

Full board analytics

The prediction matrix, scenario bands, regime timeline, book construction, premise tests and the shift-point register — the deep analytical layer behind this board.

  • Prediction matrix & scenario bands
  • Book construction & sizing
  • Premise tests & falsifiers
  • Shift-point register
See plans
QAIFinance