Toll Collectors of Compute
Waylon Jepsen@waylonjepsen
−2.7%
since Sep 30 · after money added
Practiced to 0.0% before going real.
Practice chapter: simulated results; no money was invested.
The idea
I think AI compute demand keeps growing, but building new capacity is slow and expensive, and selling raw inference is a race to the bottom with thin margins. So instead of betting on compute sellers, I own the toll collectors behind them: the financiers funding the buildout, the power producers and grid equipment keeping it running, the chipmaking tolls nobody can bypass, and the data center landlords who own the real estate.
What it owns
- Physical infrastructure like turbines, switchgear, and cooling systems are ordered years before data centers open, creating a durable revenue stream ahead of buildout completion. · EatonIt supplies the electrical distribution gear installed in nearly every data center, generating revenue per building regardless of whose equipment fills it., GE VernovaIt manufactures gas turbines and grid equipment sold out years in advance, with backlogs that confirm ongoing scarcity in the power infrastructure needed for compute., Quanta ServicesIt builds transmission lines and substations connecting data centers to power grids, collecting labor and engineering value as grid construction becomes the constraint., Vertiv HoldingsIt supplies cooling and power distribution systems that dense artificial intelligence server racks require, selling infrastructure that becomes mandatory as density increases.22%
- Every computation consumes electricity, and power near data centers remains scarce enough to support premium pricing for new generation capacity. · Constellation EnergyIt sells nuclear power directly to hyperscalers under long-term contracts, capturing scarcity value as data center electricity demand outpaces available supply., VistraIt sells electricity into markets where data center power demand is outrunning generation, capturing margin from the shortage., Talen EnergyIt supplies power directly to data centers through long-term agreements, benefiting as compute operators pursue captive power supplies to secure electricity., NextEra EnergyIt builds new electric generation faster than peers in the United States, expanding capacity as data center demand pulls forward renewable and storage investment.20%
- The firms lending and leasing capital for data center expansion collect contracted payments regardless of which compute architecture dominates. · Blue Owl CapitalIt finances data center construction through loans and leases, collecting payments as the capital constraint becomes the primary limit on buildout speed., BlackstoneIt deploys infrastructure capital into data centers and owns operating assets in the space, collecting returns whether compute margins compress or remain wide., Brookfield CorporationIt owns power plants, transmission infrastructure, and data centers directly, collecting returns from physical assets independent of whether compute margins sustain., Apollo Global ManagementIt finances chip plants and data center debt through private credit, collecting fees and interest as private lending becomes the standard for multi-billion dollar buildouts.19%
- Every chip that powers artificial intelligence passes through a toll booth, and the companies running those booths get paid regardless of which compute seller wins. · Taiwan SemiconductorIt manufactures every advanced artificial intelligence chip produced globally, collecting a toll on demand that persists across all compute winners., ASML HoldingIt manufactures the only lithography machines capable of printing advanced semiconductor nodes, operating as the deepest toll booth in the chip production stack., BroadcomIt designs custom artificial intelligence chips and networking hardware that hyperscalers use to reduce reliance on consumer chip architectures, capturing value in proprietary inputs.17%
- Real estate operators collect rent from whoever occupies the racks, insulating their margins from competition in compute itself. · EquinixIt operates hyperscale data centers as a landlord, raising rents as finished capacity remains scarce relative to demand for compute facilities., Digital Realty TrustIt leases hyperscale computing space to operators racing to add capacity, collecting rent independent of which artificial intelligence model or vendor prevails., Iron MountainIts data center business is growing on land and power positions, capturing value as a smaller landlord riding the same scarcity in finished compute facilities.14%
- Cash8%