Electricity Is the Bottleneck
I believe electricity is the real bottleneck for AI, so I want to own the power producers selling into data centers alongside the equipment makers and builders wiring them up.
What’s inside
- CEG Constellation Energy: Google's 3.6 GW deal makes it the clearest listed proxy for AI power contracts
- VST Vistra: 20-year PPAs with Meta tie existing nuclear output directly to AI demand
- NRG NRG Energy: Texas generation plus retail gives leverage to data-center load growth
- NEE NextEra Energy: Largest renewables pipeline, the quickest new megawatts to interconnect
- SO Southern Company: Southeast load growth now dominated by data-center interconnection requests
- GEV GE Vernova: Gas turbines booked to 2028 and $2.4bn of data-center electrification orders
- ETN Eaton: Switchgear and power management sit between the grid and every rack
- VRT Vertiv: Cooling and power distribution inside the building scale with every GW added
- HUBB Hubbell: Electrical distribution components are a bottleneck part of the bottleneck
- PWR Quanta Services: Grid construction labour is scarce; backlog prices the delay risk in new load
- FIX Comfort Systems USA: Mechanical and electrical contracting backlog is now largely data centers
- EME Emcor: Builds the electrical guts of the facilities the power is contracted for
- CCJ Cameco: Uranium supply backs the nuclear PPAs hyperscalers keep signing
- BWXT BWX Technologies: Reactor components and SMR work are the multi-year leg of the power bet
- GNRC Generac: Backup and behind-the-meter generation when grid connection is denied or delayed
Why it could work
If AI compute keeps growing, the scarce input is firm megawatts and the gear and crews to deliver them, and those are priced by contract rather than hype.
What could break it
A step-change in model efficiency or a stalled buildout leaves contracted plants idle and equipment backlogs cancelled, exactly as power stocks showed in the DeepSeek selloff.
For learning only. Not investment advice.