When smarts get cheap, own what smarts cannot copy

Investor Cliff Sosin had been thinking about how to judge companies once artificial intelligence keeps improving. The useful move, he said, was to assume everyone already has really great intelligence and then ask what is left.

His answer, in his words, was this: "What you want to own are the complements to intelligence."

A complement here means something used together with intelligence that becomes more valuable as intelligence gets cheaper. Sosin puts data, attention, love and affection, physical proximity, scale, and, more cynically, regulatory capture in that group.

That framing comes from his view of software. Many software tools, he said, are "crystallized and hardened pieces, like projections of intelligence." If smarts become cheap and ubiquitous, he said, the value shifts to what intelligence cannot replace.

His detailed example is Meta. The core business, in his description, is no longer social media but personalized TV, the tailored feed of Instagram videos. Seen from far away, Meta brings in about $250 billion a year and spends about $250 billion a year, counting both costs and capital spending.

He offered that as a deliberately simplified picture, saying in a highly reductionist sense all Meta buys is intelligence. A data center is an intelligence producing tool, bought inference is just intelligence, and paying smart people to write algorithms and code is intelligence. Reviewing videos he called intelligence adjacent.

On costs, he said the price of intelligence might fall "by a factor of 100 or something over the next five years." Patrick O'Shaughnessy put the cost-side shorthand as "First principles, their costs kind of go away." Sosin treated that as only the cost side of the story, adding that Meta might not take its costs to zero and might instead spend more than zero. Better intelligence could also mean "they do a far better job of picking the videos to show people next."

The rest is revenue. Better, cheaper intelligence lets creators make better content, lets Meta target that content more enjoyably, and lets advertisers design better targeted ads with smoother paths to purchase. What is not destroyed, in this account, is enormous data about viewing and enormous attention, which compound each other. The common pushback is that outsiders do not understand how hard and complicated the work is. His reply was, okay, sure, but that is all just intelligence, meaning the cost of that hard work is what falls.