AI Power Scarcity Basket
Bala Kamallakharan@balainiceland
−5.8%
since Oct 6
The idea
I believe the AI build out is bottlenecked by power, not chips, so this basket leans into the pieces that can't be built fast: gas turbines, transformers, switchgear, and firm generation that can sign a data center deal today. It also owns the engineering and electrical contractors who build out the grid and fit out the sites, plus a smaller slice of data center landlords and cooling as the real estate layer. I'm keeping some cash on the side and spreading across many names rather than betting on one. If power stops being the constraint, that's when I'd rethink it.
What it owns
- Turbines, transformers and switchgear have multi-year backlogs, giving makers pricing power regardless of which AI company wins. · GE VernovaIts gas turbines are sold out for years, proving that scarce equipment cannot be built fast enough to meet data center demand., EatonIt makes power management gear that every data center and grid upgrade requires, with demand outrunning its ability to supply., HubbellIt sells grid components, connectors and utility hardware that every transmission upgrade requires to carry new AI loads., Powell IndustriesIt makes switchgear in a market where lead times exceed three years, placing it directly at the equipment chokepoint as backlogs and margins reflect scarcity., BWX TechnologiesIt is one of the few companies that can make heavy nuclear components, so any nuclear expansion for AI power runs through its manufacturing shops., nVent ElectricIt makes electrical enclosures and protection gear that goes into every data center and substation, part of the equipment shortage.29%
- Skilled electrical labor and construction crews are as scarce as the hardware, and someone has to string the lines and wire the buildings. · Quanta ServicesIt is the largest grid builder in the country, engineering and stringing the transmission lines that connect new power to data centers., EMCOR GroupIt performs the electrical work inside data centers, building out the capacity as construction backlogs grow., Comfort Systems USAIt handles mechanical and cooling fit-out inside data centers, getting paid per building regardless of which chip vendor wins., MYR GroupIt builds transmission lines, and new power is useless to a data center until someone wires it to the grid., Sterling InfrastructureIt does site work that precedes every data center, seeing the buildout earliest as new data center site starts keep coming.24%
- Power plants that already exist and can sign data center deals today become more valuable each year the buildout runs ahead of new supply. · Constellation EnergyIt runs the largest nuclear fleet in the country and can sign data center contracts today for firm, round-the-clock carbon-free power., VistraIt owns nuclear and gas plants that can sell firm power to data centers immediately while new capacity remains scarce and years from completion., Talen EnergyIt has already signed power deals directly with data center operators, making it the purest bet on hyperscalers competing for generation that exists today., OkloIf electricity truly gates the AI buildout for years, small reactors built next to data centers become a real answer, requiring regulatory approvals and signed customers.21%
- Data center real estate and cooling infrastructure become scarcer as power, not land, becomes the binding constraint on new capacity. · VertivIt sells cooling and power equipment inside data centers, picks and shovels that every AI buildout requires regardless of chip winners., Digital Realty TrustIts data center real estate becomes more valuable when power, not land, is the binding constraint, as leasing stays tight and rents rise on power-secured capacity., EquinixIt owns interconnection hubs where networks and clouds meet, real estate that becomes scarcer as AI traffic grows and occupancy stays tight., Modine ManufacturingIt is the focused bet on liquid cooling, which AI chips increasingly require, so it pays off if adoption keeps accelerating.17%
- Cash9%