AI Eats Everything
gavX⚡️@thegav
−6.3%
since Oct 5
The idea
I believe AI automates both knowledge work and physical labor faster than most expect, and the value flows to whoever owns the compute, power, and robots rather than any single model maker. So the basket leans into chips, memory, networking, and data center infrastructure as the core, backed by cloud lab owners, a Musk-anchored bet on Tesla and SpaceX, and a patient robotics sleeve. Around that sits a hard-assets and strong-dollar view through gold, bitcoin, land, and select country exposure, while I skip easily cloned software plays.
What it owns
- Whoever wins the model race, every lab needs these chips, memory and networking components to build out AI infrastructure. · NVIDIA CorporationWhoever wins the model race buys these chips to train and run models, making this the safest way to own AI infrastructure without picking a winning lab. It also holds an OpenAI stake, giving indirect access to frontier research., Taiwan Semiconductor ManufacturingEvery serious AI chip, including Nvidia's, gets manufactured at TSMC, making it the physical chokepoint for AI infrastructure. This dynamic holds as long as no other foundry can produce leading-edge chips at comparable scale, though geopolitical risk to Taiwan represents the primary vulnerability., Broadcom Inc.A hedge within the chip bet: if big clouds move away from Nvidia they move to custom chips Broadcom designs, and Broadcom's networking gear ties the clusters together either way. It holds as long as hyperscaler AI spending keeps growing., Micron Technology, Inc.AI chips require extreme amounts of high-bandwidth memory, and Micron is one of only three companies globally that manufactures it. The take depends on memory remaining a constrained resource; a significant price decline would signal that constraint is easing., Arista Networks, Inc.Giant AI clusters are only as fast as the network linking the chips, and Arista sells that plumbing to the biggest data center builders. It's part of the infrastructure-plays-are-safer core; it needs cluster buildouts to keep accelerating.25%
- Two bets on Elon Musk's companies: one public, one through a fund holding private shares. The take works if Tesla ships self-driving and Optimus robots at scale, and if SpaceX keeps dominating launch and Starlink. · Tesla, Inc.The publicly traded way to own one half of the Musk stack. It works if Tesla turns AI work into real products: self-driving that ships at scale and Optimus robots doing physical work, not just demos., SpaceXAccessed through a fund holding private SpaceX shares, this is the other half of the whole-stack Musk conviction. It works if SpaceX keeps compounding in launch and Starlink while staying the dominant path off the planet.17%
- Gold, bitcoin, a stronger dollar, and land in countries with sound policy express a view that hard assets and scarce property outpace cash. The take holds if real things keep beating paper money and if capital keeps flowing to the right geographies. · SPDR Gold SharesGold held as the ballast in the basket to preserve value while AI bets do the compounding, expressing a view that hard assets will outpace cash. It works as long as real assets keep beating paper money., Global X MSCI Argentina ETFA fund of Argentine stocks expressing a bet on a country with policy headed the right direction. The premise breaks if Argentina's reform path reverses., BitcoinA small position in the digital version of the hard-assets-beat-cash view. It works if bitcoin keeps behaving like scarce property rather than a leveraged technology trade., Invesco DB US Dollar Index Bullish FundThis fund rises when the dollar strengthens against the pound and euro, expressing bearishness on those currencies. It holds as long as the US keeps attracting the AI talent and capital that Europe doesn't., The St. Joe CompanySt. Joe owns huge tracts of land in Florida, combining hard assets with right-place policy in one position. It works if people and money keep moving to Florida and land values follow.16%
- Indirect ownership of frontier AI labs and their models through public companies. The take holds if model builders and the infrastructure they control keep capturing the highest-value pieces of AI's economic upside. · Alphabet Inc.A direct public stake in a frontier AI lab through a large Alphabet ownership position in Anthropic, plus Gemini, DeepMind and its own TPU chips. The premise is that owning models and the compute underneath keeps Google among the highest-value pieces of the pie., Amazon.com, Inc.Amazon is Anthropic's biggest backer and AWS runs a huge share of AI workloads for paying customers. The take holds as long as AWS keeps growing on AI demand and the Anthropic stake keeps compounding in value., Meta Platforms, Inc.Meta is pushing open-source models hardest with Llama while profiting through its ad machine, expressing a view that open models will commoditize a layer that doesn't need to stay expensive. It belongs as long as Meta's open models stay near the frontier.15%
- Electricity, turbines, cooling and electrical gear are the scarcest physical layer under AI buildout. The take works if data center power demand keeps outrunning supply for years, showing up in order books and construction. · GE Vernova Inc.GE Vernova sells the gas turbines and grid equipment that get ordered when data centers need power faster than nuclear plants can be built. It works if the power shortage this take predicts shows up in its order book., Constellation Energy CorporationIf AI eats everything it eats electricity first, and Constellation owns the biggest fleet of nuclear plants selling power directly to data center buyers. The take needs AI power demand to keep outrunning supply for years., Eaton CorporationEaton makes the electrical gear and power management that every new data center and grid upgrade requires, offering a quieter way to own the electricity bottleneck. It needs data center construction to stay hot., Vertiv Holdings CoDenser AI chips run hotter and Vertiv sells the cooling and power systems that keep them from melting. It wins on infrastructure spend no matter which lab takes the pie; the risk is a slowdown in data center buildouts.14%
- Physical work automated by AI over years, not quarters, through two positions in different segments. The take needs robot adoption in factories and warehouses to keep broadening across the multi-year horizon. · Teradyne, Inc.Teradyne owns Universal Robots arms plus a chip-testing business riding the same AI wave, expressing the embodiment bet that physical work gets automated. The long-term pillar needs robot adoption in factories to keep broadening., Symbotic Inc.Symbotic automates entire warehouse operations, representing physical embodied work happening at commercial scale today with Walmart as a major customer. Success depends on continued deployment wins and new customer adoption to validate the broader robotics opportunity.5%
- Cash8%