AI Abundance, Scarcity Wins
Andrew Chang@achang
−5.1%
since Sep 30
The idea
I believe AI agents will handle large chunks of knowledge work within a few years, making intelligence cheap while the rest of the economy adjusts slowly. So this basket owns the bottlenecks that stay scarce anyway: chips and chipmaking tools, the power and grid buildout data centers need, the digital control planes like Microsoft and Google that give agents identity and distribution, plus identity security, payment rails and industrial automation. It's fully invested with no cash, spread across many names rather than concentrated bets, meant to be held for years as this plays out.
What it owns
- Advanced chips and the handful of companies able to make and equip fabs are the physical floor under abundant intelligence. · NVIDIA CorporationIt is the compute that makes intelligence abundant in the first place, with agent workloads keeping demand for its chips running ahead of supply., Taiwan Semiconductor ManufacturingNearly every advanced AI chip, whoever designs it, gets made here, making it one of the hardest physical bottlenecks in the economy., ASML HoldingIt is the only company on earth that makes the EUV machines advanced chips require, about as pure a scarce bottleneck as they come., BroadcomIt builds the custom AI chips and networking hyperscalers use to reduce dependence on any single supplier, winning from compute demand however design battles shake out., Applied MaterialsIt sells the broad toolset every new fab needs, so it collects from the whole semiconductor buildout rather than any one chip architecture., KLA CorporationIts inspection and measurement tools make advanced chips manufacturable at acceptable yields, a job that gets harder as chips grow more complex.29%
- The platforms that give agents identity, permissions, data, workflows and distribution collect rent on every unit of intelligence put to work. · Microsoft CorporationIt already owns the identity, permissions, data and workflow layer where knowledge work happens, so agents doing that work have to run through its stack., AlphabetIt pairs frontier models with the data and distribution agents need to be useful, from search to Android to Workspace., ServiceNowIt is where enterprise workflows are defined and routed, making it a natural control point when agents start executing those workflows., SalesforceIt holds customer data and sales workflows that AI agents will need to act on, positioning itself as the central layer governing agent behavior in customer-facing work.22%
- Every new data center waits on electricity and grid equipment, the slowest-moving physical bottleneck in the story. · Vertiv HoldingsIt sells the power and cooling systems AI data centers cannot run without, and denser chips make its job more valuable with each generation., GE VernovaIt makes the gas turbines and grid equipment the data center power buildout is waiting on, with order backlogs stretching years out., Eaton CorporationIt makes the electrical guts of data centers, the switchgear and power distribution every new AI facility needs before a single chip turns on., Constellation EnergyIt owns the largest fleet of nuclear plants in the US, the scarce clean around-the-clock power hyperscalers are signing long deals to secure., Quanta ServicesIt is the skilled workforce that actually builds and upgrades the grid, exactly the kind of slow-to-scale bottleneck the economy hits when demand surges.22%
- When millions of agents act on people's behalf, proving which one is allowed to do what becomes a scarce control point. · OktaIt answers who each agent is and what it is allowed to do, building the identity rails for agent and machine authorization at scale., CloudflareIt sits in front of a huge share of internet traffic and positions itself as the gateway that decides which agents and bots get through., CrowdStrike HoldingsA world full of autonomous agents means far more machine activity to secure, and it is the platform enterprises reach for to watch it.12%
- Turning abundant intelligence into real economic activity means transactions, and the rails take a toll on each one. · VisaIt collects a toll on transactions whether people or agents initiate them, so agent commerce expands its revenue without changing its model., MastercardIt is the other half of the payment rails, building the credentials that let merchants trust agent-initiated payments on card networks.9%
- The control layer for the physical economy, where abundant intelligence eventually meets factories and machines. · Rockwell AutomationIt is the control layer where software meets machines on the factory floor, the industrial half of the bottleneck story., TeradyneIts test equipment is required to ship every advanced chip, and its robotics arm is a call on automation reaching the physical economy.6%