−7.6%
since Sep 29
The idea
I think AI compute demand keeps compounding faster than the market expects, and no single company will capture it all, so this basket spreads across the whole supply chain while leaning hardest on the choke points: power infrastructure and wafer-making equipment that every chip has to pass through, plus memory and networking that get bought no matter whose accelerator wins. It's concentrated across twenty two positions with no single outsized bet, and I keep some cash aside to add on rough days.
What it owns
- This bucket tracks the suppliers of chipmaking equipment and manufacturing capacity; advanced AI chips cannot be built without them, and their order books reflect whether capacity keeps expanding. · Taiwan SemiconductorIt manufactures nearly every advanced AI chip from NVIDIA, AMD, and the hyperscalers' custom designs, making it the single narrowest chokepoint in the supply chain., Lam ResearchIt sells the etch and deposition equipment fabs use to build logic chips and the high-bandwidth memory AI accelerators depend on, making it a broad gauge of wafer capacity growth., ASMLIt is the only company that makes the EUV lithography machines needed to print leading-edge chips, so its order book reflects whether foundries are expanding advanced production capacity., Applied MaterialsIt supplies a broad range of chipmaking equipment to foundries and memory makers, making it a diversified play on the same wafer capacity expansion without depending on one tool type., KLAIt dominates the inspection and measurement tools that catch defects in manufacturing, becoming more essential as advanced chips grow more complex and expensive to ruin.21%
- This bucket holds the companies designing and selling the accelerator chips and systems powering AI workloads across every major cloud platform and enterprise. · NVIDIAIt sells the accelerators at the center of nearly every major AI buildout, making it the most direct read on whether compute demand keeps compounding across training and inference workloads., BroadcomIt designs custom AI chips for hyperscalers and builds the networking silicon that ties accelerator clusters together, winning share even where traditional accelerator makers do not., Advanced Micro DevicesIt designs alternative accelerator chips for AI workloads; if the compute market expands fast enough, a credible second supplier wins even if it does not lead., TeslaIt is betting its future on AI for self-driving and robotics, purchasing enormous amounts of accelerator compute to advance those programs toward commercialized products.20%
- This bucket holds the power generation and distribution companies whose growth is tied directly to data centers consuming far more electricity than legacy infrastructure can supply. · GE VernovaIt sells the gas turbines and grid equipment being ordered to power new data centers, connecting its backlog and orders directly to whether data center electricity demand keeps outrunning supply., VistraIt owns power generation including nuclear capacity in the exact regions where data center demand is tightening the grid, benefiting from electricity prices driven by AI infrastructure buildouts., Constellation EnergyIt runs the largest nuclear fleet in the US and is signing long-term power contracts with hyperscalers seeking clean, always-on electricity for data centers., EatonIt makes the switchgear, transformers, and power distribution equipment every new data center needs, so its backlog grows as the infrastructure buildout accelerates.16%
- This bucket covers the connectivity, cooling, and networking suppliers whose products are essential to building AI-dense data centers; their orders rise when cluster counts and server densities climb. · Arista NetworksIt sells the network switches that connect thousands of accelerators into usable clusters, equipment that hyperscalers need regardless of which accelerator chips they choose., VertivIt sells the liquid cooling and power management systems that AI-dense data centers require, since these accelerators run far hotter than legacy infrastructure was designed for., Astera LabsIt makes the connectivity chips moving data between accelerators, memory, and processors inside AI servers, growing as cluster architectures scale and server designs evolve., CoherentIt makes the optical transceivers and lasers that carry data between racks as AI clusters scale beyond the limits of copper networking., Credo TechnologyIt sells high-speed cables and connectivity chips linking servers inside AI clusters, growing with hyperscaler orders as cluster counts and densities increase.15%
- This bucket covers the makers of high-bandwidth memory and storage systems; as AI accelerators become denser and hotter, memory supply tightens and pricing depends on demand staying ahead. · Micron TechnologyIt makes the high-bandwidth memory packaged directly onto AI accelerators, and supply constraints are tight enough that its pricing and margins depend on demand outstripping supply., Western DigitalIt makes the drives storing the data that AI models train on and generate, and its revenue reflects whether data center storage demand keeps tightening the market.8%
- This bucket holds the largest cloud and AI companies spending enormous capital on infrastructure and monetizing the results; their sustained spending and revenue growth validate whether AI demand is real. · AlphabetIt builds its own TPU chips, spends heavily on data center infrastructure, and monetizes AI through search and Google Cloud, serving as a second measure of whether AI demand is real and growing., MicrosoftIt spends tens of billions per quarter on AI infrastructure and monetizes the results through Azure and Copilot products, serving as evidence that the underlying demand is real and sustainable.8%
- Cash10%