Global models: two well-funded laboratories that say nothing, not even to their suppliers

AMI Labs and World Labs are refusing to provide details of their products or timetable. Their data providers, who are supposed to prepare the raw material, say they are working in the dark.

Companies that rely on models of the world — systems that are supposed to learn a representation of the physical environment rather than predict text — are practicing a silence that is starting to be a problem for their own partners. On September 18, 2026, TechCrunch devoted an investigation to AMI Labs, founded by Yann LeCun, and World Labs, created by Fei-Fei Li, of which Marble remains the most successful product.

Michael Rabbat, co-founder of AMI Labs and head of global models, sums up the official position: the company says it is still in the research and construction phase, and therefore does not comment publicly on product projects or timelines. The company is less than a year old. On the supply chain side, the discomfort is more direct: Alex de Vigan, managing director of data provider Physicl, explains that he could build more useful datasets if he knew what these labs are actually working on.

TechCrunch puts forward a science-fiction explanation, that of the dark forest: the world's models find outlets in robotics, video games, CGI and autonomous driving, so revealing a precise direction is tantamount to telling OpenAI, Anthropic and others where to focus their efforts. The result is a sector that is heavily funded, much commented on, and no one knows when it will produce something commercial.