Watts and Wafers
Anish@illscience
· 7
−0.8%
since Sep 15
The idea
I think the market still isn't pricing how much AI compute demand keeps compounding over the next couple of years, and I don't think this is winner-take-all. So instead of betting on one champion, I own the whole chain, tilted toward the choke points: power and wafers, plus the memory and networking everyone needs regardless of whose model wins. It's built concentrated for a two-year window rather than a decade hold, with 10% kept in cash to add on rough days. If demand growth clearly stalls, that's what would change my mind.
What it owns
- The manufacturers and equipment makers in the semiconductor production chain, where advanced chip output is constrained by factory capacity and the tools needed to build it. · Taiwan SemiconductorIt manufactures nearly every advanced artificial intelligence chip for every competitor, making it the single narrowest point in the entire production chain., ASMLIt is the only manufacturer of the extreme ultraviolet lithography machines required to print the most advanced chips, so every foundry expansion flows through it., Lam ResearchIt supplies the etching and deposition tools that every new semiconductor factory requires, making it sensitive to whether chipmakers keep expanding production capacity., Applied MaterialsIt is the broadest supplier of chipmaking equipment, providing diversified exposure to the foundry and memory maker capital spending that the take depends on., KLAIt manufactures the inspection tools that semiconductor factories use to maintain yields on advanced chips, where inspection spending accelerates as manufacturing complexity increases.21%
- The companies designing and selling the processors that power artificial intelligence systems, where demand for advanced chips continues to outpace supply. · NVIDIAIt supplies the processors that have become the standard building block for artificial intelligence systems, holding this position as long as data center spending and next-generation chip orders keep growing., BroadcomIt designs and manufactures custom artificial intelligence chips for the largest cloud providers looking to reduce costs and reduce dependence on standard processors., Advanced Micro DevicesIt provides a second source of artificial intelligence processors for cloud companies that want alternatives to the market leader, winning as long as chip sales and customer deployments keep accelerating., TeslaIt represents the bet that artificial intelligence demand extends beyond large language models into autonomous vehicles and robotics, which require enormous computing power to train.20%
- The power generation and grid infrastructure companies that supply electricity to data centers, where demand is outrunning current grid capacity. · GE VernovaIt sells the gas turbines and grid equipment needed to deliver power to data centers in a market where electricity demand is outrunning supply., VistraIt owns power plants in the exact regions where data centers are being built, and profits when electricity demand exceeds supply in those markets., Constellation EnergyIt operates the country's largest nuclear fleet, providing the clean and continuous power that artificial intelligence companies prefer, proven by long-term contracts with major technology companies., EatonIt supplies the switchgear and electrical hardware that every new data center and grid upgrade requires, providing exposure to the power infrastructure buildout.16%
- The networking, cooling, and connectivity hardware that every data center requires, regardless of which chips or architectures win in the market. · Arista NetworksIt manufactures the network switches that connect artificial intelligence chips within data centers, where network speed determines cluster performance., VertivIt supplies the liquid cooling and power distribution systems that data centers require to operate artificial intelligence chips at scale., Astera LabsIt manufactures the connectivity chips that move data between thousands of processors in large artificial intelligence systems, becoming essential as clusters scale upward., CoherentIt produces the optical transceivers that carry data between distant artificial intelligence data centers, where optical networking demand from hyperscalers keeps accelerating., Credo TechnologyIt manufactures the high-speed cables that connect racks of artificial intelligence servers within data centers, winning as hyperscalers build denser clusters.15%
- The companies making high-bandwidth memory and storage, where supply remains tight relative to the demands of training and running large artificial intelligence systems. · Micron TechnologyIt manufactures high-bandwidth memory that every artificial intelligence accelerator requires, and is one of only three companies capable of producing it at scale., Western DigitalIt manufactures storage drives for the vast amounts of data required to train and serve artificial intelligence models, benefiting as storage demand from data centers accelerates.8%
- The largest technology companies whose sustained spending on computing infrastructure anchors the entire supply chain that this take depends on. · AlphabetIt demonstrates that artificial intelligence demand is not zero-sum by building its own processors while spending heavily on external computing capacity alongside its internal infrastructure., MicrosoftIt is the largest buyer of artificial intelligence computing capacity, and its capital spending decisions directly drive demand throughout the entire supply chain.8%
- Cash10%