AI Barbell, Skip the Middle
FelixQE@felixqe
−4.1%
since Sep 30
The idea
I believe AI spending pays off unevenly: the picks and shovels sellers get paid no matter whose bet wins, and a few companies with real distribution are already turning AI into earnings. The messy middle, spending heavily with no pricing power, is what I skip entirely. So this basket splits between compute and power infrastructure on one end and proven monetizers on the other, holding cash to add if the whole theme sells off together. I'm running this over a one to two year window, betting tightening budgets force the payoff question to get answered.
What it owns
- Large technology buyers of AI that already convert it into reported earnings rather than just spending on it. · Microsoft CorporationIt converts AI it buys into revenue through Copilot and Azure AI sold on its existing distribution, making it the clearest case of early monetization among major tech buyers., Meta PlatformsIt spends heavily on AI but already collects: better ad targeting from AI drives higher revenue per user right now, so AI spending gets visibly outrun by the earnings it generates., Alphabet Inc.It owns both frontier AI models and billions of users to sell them through Search, Cloud and Android, the distribution advantage that separates winners from the middle tier., Amazon.comIt gets paid twice: AWS rents compute to the AI crowd and its retail operation uses AI with real pricing power, so it wins on both infrastructure sales and internal AI monetization.33%
- Chip makers, foundries and networking suppliers who collect revenue on every dollar spent in AI infrastructure regardless of which systems win. · NVIDIA CorporationIt gets paid for AI infrastructure buildout regardless of which buyer's bets succeed, provided training and inference stay concentrated on its chips and data center orders keep growing., Taiwan Semiconductor ManufacturingIt manufactures nearly every advanced AI chip including competitors' designs, so it collects on the buildout regardless of which chip designer wins the market., Broadcom Inc.It builds the custom AI chips that hyperscalers use to reduce NVIDIA dependence, so it wins even if the market fragments away from off-the-shelf processors., Micron TechnologyIt makes high-bandwidth memory that AI chips require to function, one of only three suppliers globally, so it gets paid on the buildout as long as demand outpaces supply., Arista NetworksIt supplies the fast networking gear that ties thousands of AI chips together inside data centers, winning contracts as cloud giants keep expanding AI clusters.32%
- Electricity, cooling and grid equipment sellers who get paid on data center buildout the same way infrastructure providers always do. · Vertiv HoldingsIt sells the liquid cooling and power equipment that AI chips require, gear data centers cannot skip even when they become selective about which systems to fund., GE VernovaIt sells turbines and grid equipment to supply the electricity that data centers need but the grid was not built to provide, a growing requirement as AI power demand expands., Eaton CorporationIt makes the electrical equipment that moves power from grid to racks, a deferred purchase no data center can avoid regardless of broader AI spending trends., Constellation EnergyIt sells always-on nuclear power that hyperscalers sign long-term contracts for at premium prices to secure reliable electricity for AI data center buildouts.18%
- Software firms collecting real revenue for AI that solves problems inside customer operations today, not proof-of-concept versions. · Palantir TechnologiesIt is one of the few software companies where customers demonstrably pay for deployed AI in operations rather than pilots, putting it among genuine converters of AI into revenue., ServiceNowIt charges real money for AI features inside workflows companies already depend on, pricing power the losing middle lacks because customers pay up rather than treat AI as free.7%
- Cash10%