This week at the All In conference, a panel on world models was more about corporate secrecy than it was about the technology itself. The two leading companies in the field — Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs — have been gathering attention and funds, yet their hunger for profit remains small. When the moderator pushed them to explain what they actually construct, the responses stayed deliberately hazy. Very hazy indeed.
The central issue at hand is straightforward: what part of the real economy will world models occupy? The technology offers automation of spatial intelligence, with uses spanning robotics, interactive video, and more advanced self-driving systems. Yet the companies behind these tools are not revealing where the first commercial successes will occur.
The AMI Labs Test
AMI Labs is less than a year old, which gives it a natural excuse to stay quiet. But the company’s VP of World Models, Michael Rabbat, made the silence official. During the panel, he dodged a direct question about what the company is building. “We’ll talk about it when we’re ready to talk about it,” he said.
Rabbat responded via email with additional clarification, writing “We’re still in a research and building phase, so we’re not talking publicly about any product plans or timeline,”.
What World Labs Has Shown So Far
The Marble platform from World Labs is further along than other offerings, covering plain media creation, explorable environments for video games, and CGI effects. Robotics applications exist as well, though the platform appears more built to show what it can do than to deliver a finished product.
These demonstrations are indeed striking. Yet they remain demonstrations rather than roadmaps. Furthermore, the secrecy surrounding what World Labs is actually constructing extends well beyond the company’s own boundaries.
The Supplier Who Does Not Know
During the same conference’s off-hours, I had a conversation with Alex de Vigan, the CEO of Physicl, which supplies data to the world-modeling industry. His firm provides data for whatever AMI and World Labs happen to be building at any given time. Yet he confesses that he has no clue what they are making.
“I wish they would tell us more. We could build more useful data if we knew what they were working on,” de Vigan told me.
A supplier in such a position holds a remarkable standing: a firm pays for data that fuels its own offerings, yet it has no knowledge of which of its products those data are sustaining.
Why the Secrecy Makes Sense
The reason for the secrecy has to do with how these systems work. The same kind of modeling that allows a self-driving car to move through traffic can also serve a humanoid robot carrying boxes, or turn a brief period of video footage into a place people can explore.
Through its Nabia partnership, AMI has already dipped its toe into several fields, including manufacturing, biomedicine, robotics, and AI software for doctors. It is unlikely to chase all of these, but perhaps one or two of them are standing out.
If AMI were to announce tomorrow that it had built a humanoid OpenClaw or a next-generation Hollywood rendering system, it would create competition. A great many other labs would suddenly take an interest in the field. The lab would then face potential competition from the other world model companies, the neolabs, and even OpenAI and Anthropic.
In certain respects, it is the opposite side of that simple fundraising. Your rivals can raise funds as well — the very money that enables you to construct quietly is also supporting numerous potential adversaries once the route to market comes into view. Yet even though that opposition is unavoidable, it is best to postpone it for as long as feasible, which means staying silent on precisely what you are constructing.
A Dark Forest Scenario
Anyone familiar with Cixin Liu’s work will spot this as a dark forest situation: when you cannot tell who else is hiding among the trees, it pays to avoid drawing notice to yourself.
The comparison holds up. In the dark forest, exposing your location marks you for danger. In world-modeling, disclosing your aim identifies you as a rival.
The firms are engaged in a contest where no one wants to reveal their own plans while trying to learn all they can about what rivals are working on. This balance holds up so long as raising money remains simple.
What Comes Next
One announcement will break the silence, and others will come after it. The moment will not be fixed by chance; the timing is being decided at this very instant, a choice made now.
These firms are counting on the information edge outlasting the funding edge, so they are holding their hands close rather than showing their cards too soon.
Watching the demos is currently the strongest method for keeping up with the world-modeling space, since they reveal what can already be done while the secrecy surrounding the work points toward what remains to come.
Key Facts Box
– AMI Labs is less than a year old
– AMI’s VP of World Models is Michael Rabbat
– World Labs’ most developed offering is Marble
– Conference: All In
– AMI’s Nabia partnership covers manufacturing, biomedicine, robotics, and AI software for doctors
AMI Labs, which is less than a year old, is run by Yann LeCun and employs a VP of World Models named Michael Rabbat. World Labs is led by Fei-Fei Li, whose most developed offering is called Marble. Alex de Vigan serves as CEO of Physicl, a data provider for world-modeling companies. The event bringing these groups together is known as All In.
- AMI Labs and World Labs are the two biggest players in world models.
- Neither company has publicly committed to a product timeline.
- Suppliers like Physicl provide data without knowing the final application.
- Announcing a specific product attracts competitors from OpenAI and Anthropic.
- The secrecy is a strategic choice, not a marketing strategy.
See the video the story is built around at TechCrunch.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.