Six Bets Against Consensus
Conor Watkins@conor
−2.2%
since Oct 7
The idea
I'm betting against six consensus views at once rather than picking one. On AI, I own onsite power, licensed data providers, and security tools wrapped around agents, while also hedging with software firms that benefit if compute gets cheap and overbuilt instead of scarce. On housing, I own single-family rental landlords and insurers positioned for a harder coastal market. Each belief gets its own slice, capped so no single one can sink the basket, with cash held back to add if one group gets hit hard. I'd rethink the power bet specifically if grid interconnection reform actually speeds up.
What it owns
- Data centers that face years-long grid connection queues buy their own generators, fuel cells and cooling systems instead, creating sustained demand for onsite power equipment. · Bloom EnergyIt manufactures fuel cells that data centers install on their own property to generate power while waiting years for grid connections, making onsite generation the practical path forward for large computing facilities., CaterpillarIt manufactures gas generators and turbines that data centers buy for onsite power when grid connections take too long, adding a new customer base to its traditional industrial markets., Vertiv HoldingsIt supplies the power and cooling gear that hyperscalers install inside data centers, and demand stays strong as long as AI capacity builds continue., Modine ManufacturingIt manufactures cooling systems for data centers, a focused way to own the thermal management side of onsite power buildout as hyperscalers avoid grid queues.20%
- AI labs pay licensing fees for trusted human-created archives because synthetic training data produces weaker results, and these companies own the largest, hardest-to-replicate collections. · RedditIt owns years of verified human question-and-answer archives that AI labs license for training, and that licensing income grows as models improve and training demand accelerates., RELX plcIt owns Elsevier's scientific journals and LexisNexis legal databases, which AI labs license for training because those expert-written archives are difficult to replicate or synthesize., Thomson ReutersIt owns Westlaw and other expert legal content that lawyers and AI tools both depend on, commanding higher licensing fees because that verified content stays difficult to replace., John Wiley & SonsIt publishes academic journals and licenses that content to AI developers for training, with a direct line of sight to growing licensing revenue as models keep improving.17%
- Rising storm risk lets reinsurers and specialty carriers charge higher premiums at renewal, while builders and manufacturers of storm-resistant homes and components gain an edge. · RenaissanceReIt insures the insurers against catastrophic storms and keeps pricing power as long as hurricane risk stays high and renewal rates hold firm., Everest GroupIt reinsures catastrophe risk and writes specialty insurance policies, capturing margin expansion as coastal property becomes harder to place and prices rise at renewal., Kinsale CapitalIt writes hard-to-place specialty policies for coastal property that standard insurers have abandoned, a market that grows when insurance becomes the binding constraint on coastal development., Simpson ManufacturingIt makes structural connectors and ties that hold houses together in high winds, and demand rises when insurers offer premium discounts for storm-resistant construction.17%
- When the cost of running AI models falls, software companies whose biggest expense is AI inference see their profit margins widen, offsetting losses elsewhere in the portfolio. · ServiceNowIt sells AI agents to enterprises and operates them at scale, so falling inference costs expand its margins while usage fees stay flat or rise., IntuitIt runs AI assistants for tens of millions of users across TurboTax, QuickBooks and Credit Karma, so falling inference costs hit its bottom line directly as each interaction becomes cheaper., DuolingoIt builds AI into language lessons and conversational practice, making inference cost a major expense per learner, so margins expand when the price of each AI interaction falls.14%
- Companies deploying AI agents need identity controls, security monitoring and audit trails before they go live, creating mandatory purchases from vendors that provide those capabilities. · ZscalerIt controls what users and software can reach inside corporate networks, and AI agents need those same access controls before companies deploy them at scale., CrowdStrikeIt secures the devices and workloads that run AI agents, so security becomes a required line item before each new agent deployment., OktaIt manages logins and permissions across systems, and every AI agent needs an identity and audit trail before deployment, making access control a required purchase.14%
- If home ownership stays out of reach for a generation, single-family landlords and apartment operators collect rising rents from tenants who cannot buy. · Invitation HomesIt operates the largest fleet of single-family rental homes in the country and collects rising rents from tenants who remain priced out of home ownership., American Homes 4 RentIt builds new houses specifically designed for rental and leases them to tenants, capturing the build-to-rent demand from a generation that cannot afford to buy.9%
- Cash10%