Intelligence Revolution Data Center Stack
Clay Norris@claynorris
−10.7%
since Oct 1
The idea
I believe we're still early in a decades-long AI data center buildout, and the real money sits in the choke points, not just the obvious chips. So this basket owns the whole stack but tilts heavy toward cooling, power infrastructure, and the networking and optics tying it all together, with compute as ballast rather than the main bet. I deliberately excluded utilities and on-site generation to stay focused on what's inside the building, capped it at eleven components, and keep a cash buffer. This is a long hold; I'd rethink it if the buildout timeline itself breaks.
What it owns
- AI data centers generate heat that air conditioning alone cannot handle, making liquid cooling and chiller systems a critical supply bottleneck in the buildout. · Vertiv HoldingsIt sells cooling and power equipment positioned closest to the racks, exactly where the supply constraint tightens as operators shift toward liquid cooling., Modine ManufacturingIt makes the cooling units data centers deploy when air conditioning alone becomes insufficient, capturing the same cooling squeeze at a smaller scale., nVent ElectricIt makes liquid cooling systems and the enclosures that house racks, capturing adoption as liquid cooling spreads beyond only the largest buyers., Trane TechnologiesIt makes the large chillers that cool entire buildings, a layer that must scale regardless of whose chips run inside as data centers multiply., Johnson ControlsIt pairs chillers with the building controls that operate them, covering both cooling and environmental control as data center work grows as a business segment.25%
- Switches, fiber, connectors and optical transceivers that tie together tens of thousands of GPUs represent a layer that grows faster than chip counts themselves. · Arista NetworksIt sells the switches that cluster hundreds of thousands of GPUs into a single system, the networking layer that becomes a choke point at scale., Coherent CorpIt makes the optical transceivers moving data between racks, where AI cluster demand has outpaced supply and optics remain a bottleneck as clusters expand., Amphenol CorporationIt makes the connectors and cabling inside servers and switches, collecting a toll on the entire stack as data center buildout accelerates., Corning IncorporatedIt makes the fiber optic cable that AI data centers consume in step-change higher quantities than traditional data centers ever demanded., Credo TechnologyIt makes high-speed cables and chips linking GPUs over short distances, a pure-play on the tightest wiring segment in AI cluster design.23%
- Electrical distribution equipment, backup power systems and switchgear between the grid and the rack face years-long lead times as data center demand compounds. · Eaton CorporationIt owns the power path from grid to rack, including backup power and switchgear, sitting in one of the tightest choke points for data center buildout., Powell IndustriesIt builds the electrical distribution equipment delivering power into buildings, a smaller company capturing a substantial share of data center backlog., Hubbell IncorporatedIt makes electrical components between the grid connection and the rack, the unglamorous layer where supply remains the tightest constraint., Generac HoldingsIt provides backup power systems inside buildings, allowing data centers to meet uptime promises through equipment sales rather than utility operations.19%
- GPUs, high-bandwidth memory and servers form the foundation that the entire buildout depends on, held at moderate weight as a necessary base layer. · Nvidia CorporationIt provides the compute layer underpinning the entire AI infrastructure, held as ballast since the market has already priced the obvious component., Micron TechnologyIt sells the high-bandwidth memory every AI chip requires and cannot get enough of, a component staying sold out years in advance., Advanced Micro DevicesIt serves as the second source for AI compute, competing for real production orders that validate its chip designs against alternatives., Dell TechnologiesIt assembles servers that convert chips into data center capacity, succeeding as long as AI server orders keep growing even at thin margins.17%
- Firewalls, backup systems and management software protect billion-dollar facilities and represent a steady cost on every new data center built. · Palo Alto NetworksIt provides security systems protecting data centers, where the stakes inside have risen as AI infrastructure becomes more valuable and more targeted., Commvault SystemsIt provides backup and recovery software, the quiet layer every data center pays for as the volume of protected data grows with the buildout.5%
- AI training and inference consume enormous amounts of stored data, driving demand for capacity storage at levels not seen in the industry for a decade. · Seagate TechnologyIt makes mass-capacity drives for data storage, where AI training and the data it generates require tight supplies to sustain pricing power.3%
- Cash9%