AI needs more power

The bottleneck for AI will be electricity.

By AKA Studio

What’s inside

• CEG Constellation Energy: Nuclear fleet contracted directly to hyperscalers; purest read on power-as-bottleneck
• VST Vistra: Merchant generation repriced upward as data-center demand outruns supply
• NEE NextEra Energy: Largest pipeline of new generation to add the 92 GW Schmidt says the US needs
• EQT EQT Corporation: Gas is the fastest firm fuel for the plants being built to serve AI load
• ETN Eaton: Electrical gear between substation and rack; its order book is the bottleneck
• GEV GE Vernova: Turbines, transformers and grid kit with the multi-year lead times named in stall reports
• HUBB Hubbell: Distribution components quietly gating how fast a site can energize
• PWR Quanta Services: Transmission and substation crews are the scarce input Goldman flags
• FIX Comfort Systems USA: Mechanical and electrical contractor wiring and cooling the actual data halls
• VRT Vertiv: Cooling and power distribution are what let a delivered GPU actually be turned on
• BE Bloom Energy: On-site fuel cells let sites skip the multi-year interconnection queue the articles describe

Why it could work

The bottleneck for AI will be electricity. The basket combines direct participants and broader suppliers.

What could break it

Efficiency gains, project delays, and lower AI adoption could reduce the projected power requirement.

For learning only. Not investment advice.