This week, at the All In conference (not related to the podcast), I moderated a panel on world models and probed one of AI’s more opaque corners. The two most-discussed organizations in the space are Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs. Both have attracted attention and funding, yet neither appears to be prioritizing rapid commercialization.
What world models do
World models aim to automate spatial intelligence by creating internal representations of environments that support navigation, physical interaction and simulated exploration. That technical approach can be applied to a wide array of use cases — robotics, interactive video, explorable game environments, more advanced self-driving systems or CGI — many of which could be commercially valuable.
Reserve instead of disclosure
On the panel Michael Rabbat, a co-founder of AMI Labs and the company’s VP of World Models, participated. When pressed about specific projects he repeatedly declined to provide details: “We’ll talk about it when we’re ready to talk about it.” In follow-up email he added, “We’re still in a research and building phase, so we’re not talking publicly about any product plans or timeline.”
AMI is less than a year old, which makes a cautious public stance reasonable. Still, the same kind of reticence appears across the world-modeling field. World Labs’ Marble platform is probably the most developed product publicly shown so far: demos include media creation, building explorable environments for games, and CGI-style effects. Robotics is discussed as a use case, but the platform’s presentations largely highlight capabilities rather than final products.
Suppliers are kept in the dark, too
Secrecy extends to suppliers. I spoke on the sidelines of the conference with Alex de Vigan, CEO of Physicl — a data supplier for various world-model projects. De Vigan said Physicl’s data has been helpful for the work his customers are doing, but the supplier still does not know exactly what those customers are building: “I wish they would tell us more. We could build more useful data if we knew what they were working on.”
Strategic reasons for silence
Part of the secrecy stems from the sheer versatility of world models. The same modeling approach that helps an autonomous vehicle navigate traffic could also help a humanoid robot handle physical tasks or transform a few minutes of footage into an explorable 3D space. AMI has explored manufacturing, biomedicine, robotics and healthcare AI via its Nabia partnership — but it’s unclear which of those directions, if any, will become the central commercial focus.
There is little disagreement that world-model technology can underpin multiple viable businesses. Given how easy it has been to attract funding for the area, labs face limited pressure to narrow their focus early. There is an incentive not to: if a lab were to announce tomorrow that it had built a humanoid actuator or next-generation Hollywood rendering system, other labs and larger AI firms — including potential competitors like OpenAI and Anthropic — would quickly take notice.
That is the flip side of abundant funding: the same capital that allows a team to build quietly also funds potential competitors. To postpone direct competition as long as possible, labs keep their exact product plans under wraps. Some observers compare this posture to the ‘dark forest’ scenario from Cixin Liu: if you don’t know who else is in the woods, it’s safer not to reveal yourself.
Implications
For partners and suppliers, strategic opacity creates uncertainty and complicates efforts to tailor data or services to specific end uses. For labs, the trade-off is worth it: maintaining optionality and reducing early competitive pressure. The world-model field therefore remains technically promising and commercially wide-ranging, but the timing and direction of market-ready products will likely stay unclear until labs are ready to make public commitments.
