Labor to Compute Rails
owocki@owocki
−2.4%
since Sep 30
The idea
I think big tech is converting labor spend into compute spend, and that shift still has years to run. Rather than chase the chip names that already look priced, I lean into the power and grid backlogs feeding that buildout, alongside the platforms and compute rails themselves. About a third sits in land, physical trades, and energy that gets paid whether or not the buildout keeps accelerating, plus a small cash reserve. Nothing here is a bet on a company whose main asset is a large payroll, and I'd rethink it if the case for compute over labor stops holding.
What it owns
- Every new datacenter needs power the grid doesn't yet supply, and the equipment to deliver it is backlogged for years, creating a long runway for suppliers of turbines, electrical gear, cooling, and transmission infrastructure. · GE VernovaIt sells the turbines and grid equipment with backlogs stretching years, capturing the power side of the buildout as long as datacenter demand keeps that backlog growing., Eaton CorporationIt supplies the electrical gear every new datacenter needs to move power from grid to racks, and still trades like an ordinary industrial despite riding datacenter construction growth., Constellation EnergyIt owns the biggest fleet of nuclear plants selling around-the-clock power under long contracts straight to datacenter buyers, with increasing upside from AI-driven demand and pricing power., Hubbell IncorporatedIt manufactures transformers with lead times stretching years, along with essential unglamorous grid components required by every power project, positioning it to benefit if transformer shortages persist across the build cycle., Vistra CorpIt sells power into markets where prices rise with demand, so the datacenter buildout flows straight into earnings per megawatt as long as electricity demand growth stays real., Quanta ServicesIt physically builds the transmission lines and substations connecting new datacenters to power, paid by the project while grid construction backlogs keep growing., Vertiv HoldingsIt sells the cooling and power equipment that goes inside every new datacenter to handle dense AI chips running hot, making it a direct claim on buildout capex with fast-leading signals.32%
- Land, timber, uranium, and pipelines generate returns independent of economic direction, providing protection if compute buildout exceeds demand. · WeyerhaeuserIt owns millions of acres of timberland, a hard asset that AI cannot price or print more of, hedging against buildout overshoot while timber values stay independent of the AI cycle., CamecoIt supplies the uranium fueling reactors as datacenters keep signing nuclear power deals, bridging the compute buildout with a hard commodity that carries its own supply story., Williams CompaniesIt moves natural gas through pipelines on long contracts, getting paid whether the gas powers datacenters or homes, creating stable income independent of any single demand source., Energy TransferIt charges tolls on one of the biggest energy networks, collecting income that doesn't depend on AI buildout timing but benefits if datacenter gas demand grows., Gladstone LandIt collects rent on farmland, income tied to food and dirt rather than to whether AI capex holds up, representing what cannot be priced by AI or printed more of.19%
- The hyperscalers making the labor-to-compute trade themselves, cutting payroll while building datacenters and renting the compute back out at a margin, benefit from owning their own stack rather than paying others for it. · MicrosoftIt operates one of the platforms automating labor into compute, building datacenters and renting capacity through Azure while embedding Copilot between workers and tasks, making Azure AI revenue growth critical to justify continued capex spending., AlphabetIt owns the whole stack from chips through datacenters to apps, so capex feeds assets it keeps rather than bills it pays, with the advantage showing up in margins if ownership matters., AmazonAWS rents compute to everyone else and the AI buildout makes that rental bigger, capturing a cut as long as companies replacing labor with software need someone else's datacenters to run it in.17%
- The chip layer captures the largest single checks from the labor-to-compute conversion, as long as the big platforms keep growing capex and default accelerators remain concentrated. · NVIDIA CorporationIt collects the largest single check for the chips powering the labor-to-compute trade, as long as the big platforms keep growing capex and it stays the default accelerator they buy., Taiwan Semiconductor ManufacturingAlmost every AI chip in the buildout gets manufactured at its fabs, so it gets paid whichever chip designer wins, as long as leading-edge capacity stays scarce and it stays the only place to make these chips., BroadcomIt builds custom AI chips for platforms cutting their accelerator bills and sells the networking that ties datacenters together, surviving even if money moves away from off-the-shelf chips.14%
- Physical trades AI can make more productive but cannot replace stay valuable, and their supply chains earn from the buildout itself and from the world the buildout cannot touch. · Ferguson EnterprisesIt supplies the physical trades that get more valuable as white-collar work gets automated, and sells directly into datacenter construction, earning from both the buildout and the world the buildout cannot touch., United RentalsIt rents the machines for physical work and datacenter construction, getting paid whether capex stays strong or shifts to the next project, as long as physical construction overall stays busy., WatscoIt distributes HVAC equipment to contractors, a trade AI can make more productive but cannot replace, with volumes staying healthy independent of the AI cycle.12%
- Cash6%