−9.2%
since Sep 30
The idea
I think the AI buildout is really a construction problem, and construction bottlenecks at the hardest, slowest parts: power, chip-making tools, memory and compute, networking, and cooling. So this basket leans heaviest into power and grid, then foundry and fab equipment, with the rest spread across networking, memory, and datacenter cooling and construction, no single name allowed to dominate. I keep 8% in cash on purpose, since this group can drop hard on sentiment alone, and I'd rather buy those dips than sit through them fully exposed.
What it owns
- Datacenters consume enormous amounts of electricity, and the grid infrastructure to deliver it is constrained, creating a bottleneck for expansion. · GE VernovaIt manufactures the gas turbines and grid equipment that new datacenters require for power, with its order book sold out years ahead., VistraIt owns power plants that sell electricity directly to datacenters at premium rates while new generation capacity remains scarce., EatonIt produces the switchgear and power distribution equipment every datacenter needs between the grid and the server racks, with lead times stretched tight., Quanta ServicesIt builds the transmission lines and substations connecting new power generation to new datacenters, and qualified crews remain scarce., Constellation EnergyIt operates the largest nuclear fleet in the US and has signed long-term deals to supply carbon-free power to cloud providers at premium rates.27%
- Advanced AI chips require cutting-edge manufacturing at massive scale, and the companies here control the machines and foundries that make it possible. · Taiwan SemiconductorIt manufactures nearly every advanced AI chip at the only foundry with leading-edge yields at scale, making it the essential production chokepoint., ASML HoldingIt is the only manufacturer of extreme ultraviolet lithography machines required to produce advanced chips, a monopoly built over decades., Applied MaterialsIt sells the widest range of semiconductor manufacturing equipment, so it benefits from every new fab construction regardless of the chip maker., Lam ResearchIt dominates the etch and deposition tools used in vertical chip stacking, especially for the high-bandwidth memory that AI systems require.24%
- Thousands of AI chips must communicate at high speed within a cluster, requiring specialized networking equipment and optical connections that concentrate value. · Arista NetworksIt manufactures the high-speed switches that connect thousands of GPUs into functioning clusters and has been winning business from major cloud providers., BroadcomIt makes the networking chips inside most datacenter switches and designs custom AI processors for cloud providers, winning revenue either way., CoherentIt manufactures optical transceivers that connect racks together, a component that becomes more critical as clusters scale to larger sizes.17%
- AI systems depend on specialized memory and processors that only a handful of companies can supply, creating sustained demand as AI workloads scale. · Micron TechnologyIt is one of only three suppliers of high-bandwidth memory that AI chips require, and this memory remains sold out well in advance., NVIDIAIts GPUs and CUDA software remain the foundation for nearly all AI development, making it the demand engine that drives the entire ecosystem.13%
- Modern AI clusters generate extreme heat and require physical infrastructure that older datacenters cannot support, making specialized cooling and construction critical. · Vertiv HoldingsIt supplies the liquid cooling systems that modern AI racks need because newer chips generate too much heat for traditional air cooling., EMCOR GroupIt performs the electrical and mechanical construction work inside datacenters, and skilled labor remains a bottleneck that capital cannot quickly overcome.11%
- Cash8%