For three years the whole AI story was language: bigger chatbots, longer context, another model that writes a slightly better email. That era is not over, but the most interesting money and the biggest names in the field have started chasing something else entirely: AI that understands the physical world. Not the next word in a sentence, but how a room is laid out, how a machine part behaves under load, how the universe actually works.
This is the frontier we are calling AI Innovation, and right now it is dominated by a handful of extraordinary teams building “world models” and physics AI. This section profiles the companies and the founders betting that the next decade of AI belongs to machines that can reason about reality, not just text about it. We start optimistic, because these are genuinely exciting bets, and we always keep a friendly sceptic in the room, because a billion dollars of funding is not the same as a working product.
The deep dives
World Labs: Fei-Fei Li’s $1.23bn Bet on Spatial Intelligence
The “godmother of AI” wants machines to understand 3D space. Her first product, Marble, builds whole persistent worlds from a photo.
Project Prometheus: Jeff Bezos and the $18bn “Artificial General Engineer”
Bezos went back to co-CEO to build AI that invents and manufactures physical objects. It raised $18bn before shipping anything.
AMI Labs: Yann LeCun Bet His Reputation That LLMs Are a Dead End
The Turing Award winner left Meta to build the world models he says language models can never become, and open-sourced the plan.
Accelerated Understanding: The Team That Told Bezos No and Bet on Physics Instead
Anima Anandkumar ditched the transformer entirely for “neural operators” that simulate how physical systems evolve.