AI Infrastructure, Decade Build
Deepak@deciblast
−7.8%
since Sep 30
The idea
I believe AI is a world-changing shift and the compute buildout behind it will create value for years, so instead of picking one winner I own the whole stack: chips, the equipment that makes chips, memory and networking, power and cooling, cloud giants, and the landlords who house the data centers. Quieter bottlenecks like grid gear and cooling get real weight since that demand feels just as certain but less crowded.
What it owns
- The processors doing the artificial intelligence work; infrastructure spending on AI eventually concentrates into purchases of these components. · NVIDIAIt supplies the compute processors at the center of AI infrastructure and holds its position as long as data center builders keep ordering its chips faster than competitors gain real market share., Taiwan SemiconductorAlmost all advanced AI chips, including those from competitors, are manufactured by this company, making it essential to the entire buildout regardless of which chip designer leads., BroadcomIt builds custom AI chips that large cloud companies design for themselves plus the networking silicon connecting them, and works if clouds keep wanting their own chips alongside the dominant player., Advanced Micro DevicesIt serves as the credible alternative source for AI compute processors, and the position holds as long as buyers want options and its chips remain competitive.28%
- Data centers require enormous electricity and cooling systems, and the grid equipment delivering them represents the least crowded part of the infrastructure buildout. · GE VernovaIt sells turbines and grid equipment to supply the enormous new power capacity data centers require, representing one of the quieter bottlenecks as construction outpaces available electricity., EatonIt makes the electrical distribution and switching equipment that every data center needs before a single chip operates., Vertiv HoldingsIt sells cooling and power equipment for data centers, and denser compute means more cooling per building as liquid cooling spreads., Constellation EnergyIt operates the nation's largest nuclear fleet, providing the steady carbon-free power that AI data centers are contracting for long-term., VistraIt owns power plants in the regions where data centers are expanding fastest and sells them electricity, holding value if power demand keeps prices and contracts firm.20%
- The machines that manufacture advanced chips; all chip makers depend on a small number of equipment suppliers regardless of competitive outcomes. · ASML HoldingIt is the sole manufacturer of lithography machines required to print advanced chips, making continued machine orders essential as fabs expand through the decade., Applied MaterialsIt sells the broadest range of chipmaking tools, getting paid whenever any fabrication plant expands regardless of which chip types are produced., Lam ResearchIts etch and deposition tools are essential to manufacturing both logic chips and memory, and AI demands more of both., KLA CorporationIts inspection tools find defects in manufacturing, and more advanced chips require increasingly intensive inspection per wafer.18%
- AI systems need vast amounts of fast memory and interconnecting cables, both in chronic short supply. · Arista NetworksIt supplies the networking switches that connect thousands of AI chips inside data centers, and demand grows as compute deployments expand., Micron TechnologyIt is one of three companies making the vast amounts of fast memory AI systems require, and holds value if memory demand stays ahead of supply and prices remain healthy., Marvell TechnologyIt makes the chips moving data between processors and memory, plus custom silicon for cloud companies, handling the interconnection layer of AI infrastructure.12%
- The largest buyers of AI infrastructure, held as steady ballast through a decade-long buildout. · MicrosoftIt spends heavily on AI compute and converts that infrastructure into revenue through its cloud service, making it critical to turning hardware spending into actual returns., AlphabetIt operates a major cloud service, designs its own AI chips, and integrates AI throughout its business, holding value if cloud and ad revenue grow faster than AI spending rises.9%
- These companies own the buildings where all the hardware sits and collect rent throughout the decade-long expansion. · EquinixIt owns data center buildings and collects rent from companies filling them with AI hardware, holding value as long as occupancy and rents remain strong., Digital RealtyIt owns and leases large-scale data center space that AI buildouts require, diversifying the landlord layer across multiple owners.7%
- Cash6%