Scarcity After Intelligence
Andrew Chang@achang
+8.6%
since Oct 3
The idea
My bet is that as AI makes intelligence cheap and abundant, value shifts to whatever stays scarce: proprietary data that keeps regenerating itself and customer relationships nothing can route around. So the basket splits across distribution owners like platforms and marketplaces, and data owners in ratings, vertical records, and risk and identity, each name sized modestly rather than concentrated in one bet. I deliberately skip data businesses that are just scrapable content, since that's commoditized by the same forces I'm riding.
What it owns
- Control over customer relationships and transaction endpoints remains scarce as intelligence gets cheaper, since end customers and their attention still have to go somewhere. · Meta PlatformsIt owns billions of daily customer relationships, and the scarce thing becomes the feed that decides what gets seen as AI makes content cheaper., AmazonIt is where purchases complete with inventory and logistics, and AI shopping agents still need to buy from somewhere with trust and distribution., AppleIt owns the device in most pockets, a customer relationship no model routes around, so whatever AI wins must reach people through its hardware and store., Booking HoldingsIt owns the travel demand relationship, and hotels still need its customer pipeline even if AI plans the trip., VisaIt is the rail every transaction crosses regardless of which AI arranged the purchase, the purest distribution no model routes around., ShopifyIt owns the direct relationship between millions of merchants and their customers, and AI tools make merchants more capable without replacing the platform they sell through., Uber TechnologiesIt owns the physical network of drivers, riders and local delivery that stays scarce, since atoms do not get cheaper the way tokens do.38%
- Legal, clinical, property and regulatory records are paywalled and continuously regenerated by their industries, making them more valuable to query as models proliferate. · RELXIt owns legal records and scientific publishing that are paywalled and licensed, not scrapable, the exact line between data that compounds and data commoditized., IQVIAIt owns clinical trial and prescription data fenced by patient privacy law and regenerated by the drug pipeline itself, making it more valuable as AI scales., Thomson ReutersIt sits inside legal and tax workflows with proprietary records that professionals need, and every AI legal tool requires its licensed content as the trusted source., Veeva SystemsIt is where pharma companies keep regulated records, a system of record the industry cannot easily leave and any AI layer must plug into., CoStar GroupIt built a commercial property database through decades of on-the-ground collection, so it is proprietary and continuously refreshed in ways no scraper replicates.22%
- Debt and capital flows pass through rating and index franchises whose authority is regulatory and reputational, not technical, so cheaper AI cannot displace them. · S&P GlobalIt owns the ratings and index franchises that debt and fund flows pass through as a regulatory and reputational toll that intelligence alone cannot recreate., Moody'sIt is half the ratings duopoly plus decades of proprietary credit data that becomes more valuable as models need licensed, legally fenced inputs., MSCIIt owns the benchmarks that trillions in funds are contractually tied to, a distribution lock on capital flows that cheaper intelligence does not dissolve., Intercontinental ExchangeIt owns exchanges and mortgage data pipes where transactions must happen, legally fenced data regenerated every trading day that cannot be scraped.22%
- Credit files and claims histories are legally fenced and rebuilt every day, so models need licensed access and cannot scrape their way to equivalents. · Verisk AnalyticsIt holds the insurance industry's claims history, contributed by insurers themselves, so the data regenerates and nobody can rebuild it elsewhere., Fair IsaacIt provides the score standard that US lending runs on, and no model can generate a credit score lenders and regulators accept without FICO's franchise., EquifaxIt owns credit files and income verification databases, the legally fenced and continuously regenerated data that becomes more valuable as intelligence gets cheap., TransUnionIt owns consumer credit files alongside the other bureau, same logic that the file is the scarce input and better AI makes it worth more per query.19%