AI Bottlenecks, Earth to Customer
EuroFinch@eurofinch
−8.3%
since Oct 9
The idea
I believe the AI buildout runs into physical limits long before demand does: power, grid equipment, copper, uranium, rare earths, data center space, and the most specialized chipmaking steps all have years-long backlogs and demand from outside AI, so I'd rather own those bottlenecks than the most cyclical layers. The portfolio leans heaviest on power and grid, then data centers and raw minerals, with smaller slices in chipmaking and a thin top layer of chip designers and software to keep the whole chain represented.
What it owns
- Data centers running AI models need new power plants and grid infrastructure, both of which have multi-year waiting lists to build and install. · GE VernovaGas turbines are among the slowest infrastructure components to add to the power system, and the company's order book extends multiple years into the future with demand from data centers and many other industries., Eaton CorporationTransformers, switchgear and electrical distribution equipment remain scarce as data centers, utilities and industrial facilities all compete for them, with orders outpacing available supply., Quanta ServicesBuilding new transmission lines and electrical substations requires skilled labor that cannot be trained and hired quickly, and this company operates the largest construction business for that work., Constellation EnergyNuclear power plants take a decade or more to build, making existing plants more valuable when data centers need reliable electricity around the clock, and major technology companies are signing long-term power purchase agreements., VistraThe company operates gas and nuclear power plants in constrained electricity markets, and government investment programs are increasing demand for the power it generates.31%
- The physical buildings, specialized cooling systems and fiber networks that convert electrical power into usable computing capacity for AI workloads require years to construct and permit. · Vertiv HoldingsAI chips generate significant heat and require specialized cooling systems and power distribution racks, which this company manufactures for data center builders ordering faster than it can deliver., EquinixThe company owns data center properties in major cities where networks interconnect, and securing new buildings with available power connections takes years to develop and authorize., Digital Realty TrustThe company leases large data center campuses to cloud providers, and finding suitable land with existing power infrastructure available remains the limiting factor in expansion., CorningThis company manufactures the optical fiber that connects AI data centers internally and also supplies telecommunications carriers, with demand growing faster than production capacity.18%
- Copper for electrical wiring, uranium for nuclear reactors and rare earth elements for motors form the ground-level inputs required for the entire AI infrastructure buildout. · Freeport-McMoRanEvery data center, power line and manufacturing facility requires copper, and opening new mines takes ten years or longer, while this company ranks among the world's largest producers., CamecoThis company mines and processes uranium that fuels the nuclear reactors powering data centers, benefiting from long-term fuel contracts and reactor extensions or restarts., MP MaterialsThis company operates the primary rare earth mine in the United States and is developing magnet production capabilities currently dominated by foreign manufacturers., Southern CopperThe company controls some of the world's largest copper reserves, which become increasingly valuable when new copper supply takes years to develop and bring online.17%
- A small number of companies control the memory production, advanced manufacturing processes and specialized packaging steps that every AI chip must pass through before use. · Taiwan Semiconductor ManufacturingNearly every advanced AI chip is manufactured by this company, including the advanced packaging stages that have become a critical constraint in chip production., ASML HoldingThis company is the sole manufacturer of the extreme ultraviolet machines required to print the most advanced computer chips, with no competitors matching its technology., Micron TechnologyAI chips require high-bandwidth memory, and this company is one of only three global manufacturers of the technology, with demand consistently exceeding supply., Amkor TechnologyThis company performs advanced packaging for computer chips, a critical step in AI chip manufacturing that remains capacity-constrained as demand accelerates.16%
- The top of the stack where AI technology reaches end customers through products and services. · NVIDIAThis company designs the chips that consume all the power, copper and data center space throughout the take, controlling the majority of the market for AI processors., BroadcomThis company designs custom AI chips for major cloud providers and manufactures the networking equipment linking data centers together, with these programs expanding., MicrosoftThe company represents the end point where AI reaches customers through cloud services and software products, with artificial intelligence boosting demand across its platform.10%
- Cash7%