Riding the AI Capex Flood
Joe Morrissey@morrisseyjoe
−13.9%
since Oct 6
Practiced to −0.6% over 3 days before going real.
Practice chapter: simulated results; no money was invested.
The idea
I believe we're still early in a massive AI infrastructure capex boom, and the best way to ride it is to own the whole supply chain feeding that spending: chips and compute, the neoclouds, the data center buildout, and the power providers it all depends on. This basket is deliberately concentrated and aggressive, spread across many names rather than one single bet, with a small cash reserve kept aside to add on dips.
What it owns
- Every dollar of AI capex buys chips first, so the chipmakers and their foundry collect the flow before anyone else. · NVIDIAIt supplies the GPUs that labs, neoclouds, and hyperscalers are buying with fresh capital, placing it first in the spending flow., Taiwan SemiconductorNearly every AI chip produced gets manufactured in its fabs, so the capex wave flows through regardless of which chip designer wins., BroadcomIt builds custom AI chips and the networking silicon around them for hyperscalers spending aggressively on proprietary designs., Advanced Micro DevicesA boom this hungry for compute needs a second GPU source, and labs and neoclouds are funding its roadmap with real orders., Micron TechnologyEvery AI accelerator ships with high-bandwidth memory, and it supplies into a squeeze where demand from the capex wave outpaces supply., Marvell TechnologyThe clusters being built need custom interconnect and networking silicon that it supplies as hyperscaler custom chip orders grow.35%
- The GPU-rental clouds are where the fresh capital is pouring in right now, and the purest, highest-beta read on whether the window stays open. · CoreWeaveIt is the flagship neocloud provider, the purest bet on fresh capital funding GPU capacity and the first place a narrowing capital window would show., Nebius GroupIt is the challenger neocloud provider, high-beta on the same flow of capital funding GPU hosting capacity., IRENIt owns cheap power and is converting it into GPU hosting capacity, exactly the scarce input the boom is bidding up., Applied DigitalIt builds and operates the data centers the capex flood is paying for, with leverage that makes it one of the highest-beta ways to own the buildout.18%
- The racks, cooling, networking and servers that turn raised capital into running data centers. · Vertiv HoldingsEvery data center the capex wave builds needs its cooling and power equipment, so its backlog tracks the boom directly., Dell TechnologiesIt ships AI servers by the truckload to enterprises and neoclouds, a volume play on the same spending wave., Arista NetworksIt sells the networking that stitches AI clusters together for the cloud buyers driving the biggest buildout orders., CelesticaIt builds the custom racks and networking hardware hyperscalers are ordering with their record capex budgets.17%
- Compute needs electricity the grid doesn't have yet, so the generators and turbine makers are a bottleneck the spending can't route around. · VistraIt sells power into the markets where data center demand is tightening supply, so the capex boom shows up in its prices., Constellation EnergyIts nuclear fleet is signing long-term deals to supply power directly to data centers competing for scarce electricity in the boom., GE VernovaIts turbine backlog is the boom's power demand written down in orders, with years of visibility from data center driven generation buildout.14%
- The closest public proxies for the private labs themselves, funding them and selling them compute. · MicrosoftIt is the closest public way to own a stake in OpenAI and one of the biggest capex spenders funding the boom., OracleIt is renting massive compute to the labs, including a huge contracted backlog, making it a landlord to the boom's biggest spenders.9%
- Cash7%