Neocloud Landlords and Shovels
Christian@miamidegen
−21.6%
since Oct 1
The idea
I believe demand for AI compute keeps outrunning supply over the next few years, and the best way to play it is owning the landlords renting out the GPUs rather than guessing which AI model wins. The core is pure-play neoclouds like CoreWeave, Nebius, and Applied Digital, surrounded by the power, cooling, networking, and server suppliers who get paid regardless of which landlord signs the lease. I'm holding 10% in cash to add on rough days, and I'd rethink this if AI token demand actually stalls out.
What it owns
- Companies that buy GPUs, connect them to power, and rent compute capacity out win directly when AI token demand requires more capacity than hyperscalers can build themselves. · CoreWeaveIt buys GPUs and rents them from power-connected facilities, the direct compute-landlord play, and depends on sustained token demand keeping its leased capacity rented and its debt refinanceable., Nebius GroupIt builds GPU cloud capacity rapidly and signs customers as it scales, so it needs AI workloads to fill new campuses rather than leave them half-leased., Applied DigitalIt converts power-connected sites into AI compute and leases them out, a smaller landlord play that works only if token demand sustains multi-year lease signings on new campuses.30%
- Thousands of GPUs function as a single computer only when connected by fast networks, and these suppliers provide the interconnect infrastructure. · Arista NetworksIt supplies the switches that wire GPU clusters together, so larger clusters and more landlords drive its revenue regardless of which company operates the site., Credo TechnologyIt makes the cabling chips that move data between racks in AI data centers, scaling with cluster size, and depends on continued orders from hyperscalers and neoclouds as clusters expand., Astera LabsIt makes the connectivity chips that enable GPUs to communicate inside servers, riding each new GPU generation, and needs design wins in the next accelerator cycles., Lumentum HoldingsIt supplies lasers and transceivers for AI networks, diversifying the optics exposure, and benefits when data centers shift to faster optical links., CoherentIt makes the optical transceivers that carry data between GPUs at large scale, and its revenue grows as clusters expand and datacom optics demand rises.24%
- AI data centers are fundamentally electrical and thermal projects, and these companies supply the equipment and construction services that make a site operational. · Vertiv HoldingsIt sells the cooling and power gear every new AI data center requires, so its order flow tracks the buildout pace itself across all landlords., Eaton CorporationIt supplies the electrical equipment that connects GPUs to the grid, solving a bottleneck every landlord must address, and needs its data center segment to keep expanding., Quanta ServicesIt builds the power infrastructure and sites themselves, capturing construction value before any GPU powers on, and requires its data center backlog to keep expanding., nVent ElectricIt makes the enclosures and cooling components inside racks, a steady vendor on every buildout, and needs data center orders to remain a growing part of its business., Modine ManufacturingIt specializes in liquid cooling as GPUs run hotter and sites shift from air to liquid, earning revenue on that transition, but faces risks if adoption slows or larger competitors dominate., Comfort Systems USAIt installs the mechanical and electrical systems inside data centers, getting paid per site regardless of landlord, so its backlog directly reflects buildout momentum.24%
- Every new data center site begins with GPU servers and racks, so these vendors capture orders at the start of each buildout cycle. · Super Micro ComputerIt assembles the GPU servers that landlords deploy, so it gets paid on every new site launch and needs to maintain market share and margins as AI orders scale., Dell TechnologiesIt ships AI servers at scale to both neoclouds and enterprises, diversifying away from single-company risk, and requires its AI server backlog to keep growing.12%
- Cash10%