The Capex Collectors
david gabeau@david
−5.8%
since Sep 30
The idea
I think the AI buildout is quietly transferring value from the private labs racing to build models to the public companies selling them the picks and shovels. Whoever wins that race, the chipmakers, networking gear, data center landlords, power providers, and hyperscalers collect the revenue either way. So this basket spreads across every layer of that stack, tilted toward locked-in or recurring revenue over pure cyclical spend, with a small cash buffer and no single name dominating.
What it owns
- The hyperscalers collect cloud rent from AI workloads and everyone building on top of them, providing stable long-term revenue regardless of which models dominate. · Microsoft CorporationIt collects cloud rent from AI workloads through Azure, including from labs and enterprises, as long as AI-driven growth justifies its capex spending., Amazon.comAWS rents capacity to labs and enterprises as a core cloud landlord, collecting rent tied to AI demand without bearing the full capex bill., Alphabet Inc.It collects cloud rent while also building its own AI chips, sitting on both the infrastructure and buyer sides as long as Google Cloud grows on AI demand., Oracle CorporationIt has sold AI compute capacity years in advance to labs, converting their financing directly into contracted backlog as long as customers honor the agreements.26%
- Landlords and power providers sign multi-year contracts with hyperscalers, locking in revenue streams that persist through cycles and competitive shifts. · EquinixIt leases data center space on multi-year contracts to hyperscalers, collecting locked-in rent as long as AI demand keeps occupancy and lease rates strong., Constellation EnergyIt signs long nuclear power contracts with hyperscalers, collecting the purest form of locked-in revenue as long as clouds keep paying for reliable clean power., Digital Realty TrustIt signs hyperscale leases with the largest cloud builders, collecting rent for years regardless of competitive outcomes, as long as AI demand keeps filling its capacity., Vistra CorpIt sells the electricity that data centers run on, benefiting from AI as a scarce input that keeps power prices and contracts firm in its markets., Vertiv HoldingsIt sells the cooling and power equipment inside dense AI data centers, collecting orders as long as cluster buildouts and their thermal demands keep growing., Eaton CorporationIt sells the electrical infrastructure that powers data centers, showing up in order books as every new facility funded by labs' capital gets built.23%
- Lab financing turns into accelerator and memory orders immediately, so chipmakers get paid in cash on delivery no matter which model race outcomes emerge. · NVIDIA CorporationWhen labs raise capital, it becomes NVIDIA revenue in cash today through accelerator orders, as long as AI compute spending flows through its products at scale., Broadcom Inc.It collects from the buildout through custom AI chips and networking silicon designed for hyperscalers, as long as cloud builders keep ordering accelerators alongside traditional suppliers., Advanced Micro DevicesIt collects spend from buyers who need a second source for AI accelerators, holding as long as its AI chip revenue keeps growing on lab and cloud orders., Micron TechnologyEvery AI cluster needs high-bandwidth memory, so it collects from labs' infrastructure spending as long as AI demand keeps memory pricing and volumes strong through cycles.21%
- The machines and factories every chip depends on sit at the deepest picks-and-shovels layer, collecting revenue as capacity expands to meet sustained AI demand. · Taiwan SemiconductorNearly every AI chip gets made at its fabs, so it collects a cut regardless of which designer wins, holding as long as it keeps its advanced manufacturing lead., ASML HoldingIt is the only maker of lithography machines needed for cutting-edge chips, so more AI capex eventually means more orders as long as foundries expand advanced capacity., Applied MaterialsIt sells the equipment that fills new chip fabs, collecting when the buildout reaches the factory floor as long as AI demand keeps driving fab construction.13%
- Every new AI cluster needs switches, optics and interconnects to function, representing a direct but underweighted layer of the infrastructure buildout. · Arista NetworksIt sells the switches that stitch AI clusters together, turning lab-financed data centers into networking orders as long as it stays the go-to supplier for large deployments., Coherent CorpIt sells the optical transceivers that link racks and data centers, collecting cash from lab-financed infrastructure as long as optical demand tracks cluster buildout., Credo TechnologyIt makes the high-speed cables and interconnects that move data inside AI clusters, collecting directly from buildout spend as long as it keeps its design wins.11%
- Cash6%