Cliff Sosin argues cheap intelligence will not destroy the complements that matter most, naming data, attention, physical proximity, scale and regulatory capture, with Meta as his main example.
Sosin says a frontier lab with better models that nobody can copy would have a great business, but expects good-enough intelligence to commoditise quickly.
Sosin avoids the AI build-out because he cannot forecast supply or demand within several orders of magnitude, and says he would advise Carvana's CEO to give AI agents read-only access to every system.
Cheap intelligence makes data, attention and scale the scarce assets
If intelligence becomes cheap and ubiquitous, the value sits in what it cannot replace. Sosin argues intelligence is bounded in complex systems, so data, attention, scale and physical proximity gain worth while commodity intelligence faces price competition. That logic shapes what he owns and why he avoids the AI build-out.
Takeaways
Intelligence is bounded, especially in complex systems. Sosin calls intelligence more like a ratio than an unbounded scalar. In dynamic systems such as weather, small errors compound, as the butterfly effect began with a rounding error in a forecast model. What stays unknowable is learned only through experiments and time.
Cheap intelligence makes its complements the scarce assets. If intelligence becomes cheap and ubiquitous, value shifts to data, attention, physical proximity, scale and even regulatory capture. Meta has data and attention plus roughly $250bn of revenue and similar spending, which Sosin thinks cheap intelligence cannot destroy.
Good-enough intelligence will commoditise quickly. Sosin expects good-enough intelligence to commoditise fast, and elite demand may be diamond-shaped with few uses for a 300 IQ. Markets are U-shaped. Cogent routes 30 or 40% of IP traffic yet is worth only a few billion dollars.
Sosin stays out of the AI build-out. He cannot forecast AI supply or demand within several orders of magnitude, and the value chain makes it unclear who earns the rents. Strong price signals and talent drive such build-outs, and booms run until they are overdone.
Agents push prices toward the lowest cost. Physical proximity and scale may help, but in garbage collection agents will move customers to the cheapest provider, even for savings of 50 cents a month. Prices will move far more often. Sosin calls price a signal wrapped in an incentive.
Capital markets are underappreciated and America leads. Sosin says capital markets redeployed about 3% of GDP into AI, and anyone can buy $100,000 of stock and own well run companies at almost no cost. His pet theory credits American risk-taking, culture and institutions.
Humans stay useful if they learn fundamentals. Sosin compares investing to freestyle chess, where human plus computer beats either alone, though he expects his own role may be competed out over time. Children should learn basics without models, and teachers should set harder tasks that use them.
Where Sosin Puts His Money
Patrick says about 80% of Sosin's portfolio is Carvana. Sosin is comfortable with the concentration because its infrastructure and brand are so differentiated.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
He would advise Carvana's CEO to give AI agents read-only access to every system.
Patrick says about 5% sits in Meta.
Grindr has lots of attention and data but a small team and budget, so more intelligence could help it. Sosin says he does not own it.