LeCun and Fei-Fei Li Each Raise ~$1B for 'World Models'

Two of the most cited voices in modern AI, Yann LeCun and Fei-Fei Li, are trying to redirect the field's next spending cycle away from ever-larger language models and toward what they call world models, according to BBC reporting. Both have now backed the argument with money that matches their conviction, each raising roughly $1 billion for separate startups on the same thesis.

LeCun, Meta's former chief AI scientist, has been the loudest technical skeptic of the scaling-is-all-you-need story. His pitch, in his own words, is that today's systems 'can manipulate language, and they fool us into thinking they are smart because they manipulate language. But in fact, they are completely helpless when it comes to the physical world.' His new venture AMI Labs, founded in March, is trying to build models that learn how physical reality behaves rather than predicting the next word, an approach Fortune has also detailed in reporting on the same fundraising wave.

Li, the Stanford professor known as the 'AI godmother' who founded World Labs in 2024, is coming at the same problem from spatial intelligence. Her framing is that 'we need to go from large language models to large world models', systems that can reason about 3D space and how objects exist and interact within it. That two of the field's original architects are staking roughly a billion dollars each on this route is the part worth paying attention to.

Why it matters if you are not funding AI startups: the assumption underneath most current AI investment is that more data plus more parameters plus a language backbone will keep pushing capability forward. LeCun and Li are saying, with real capital, that this route tops out before it gets to general intelligence and that another substrate, one grounded in video, physics and embodiment, is the one that will produce agents that can act rather than just talk.

The honest caveat is that 'world models' is still more a research program than a shipped product. Take the specifics as reported, not settled. What the reporting does not give you is a timeline, a public benchmark, or a claim that either company has a working system today. If the bet lands, the beneficiaries are the builders working on robotics, simulation and any product that has to touch the physical world, who would get a foundation better suited to their problem than a chatbot API.