Dark Forest of World Models: Why AMI Labs and World Labs Won't Say What They're Building

One of the strangest things about the current AI boom is how much of it now happens in silence. This week, TechCrunch AI editor Russell Brandom published a piece titled simply “World model companies are keeping a lot of secrets,” drawn from a panel he moderated on world models at the All In conference in Los Angeles. What he found was a corner of the AI industry that has raised billions of dollars on the strength of a research vision — and has almost nothing to say about what it intends to sell.

The two companies at the center of the space are AMI Labs and World Labs. The first is the venture Yann LeCun launched after leaving Meta, formally known as Advanced Machine Intelligence, built around his long-running argument that large language models are a dead end for real understanding and that the future belongs to systems that learn predictive models of the physical world. The second is Fei-Fei Li’s spatial intelligence company, whose Marble product remains the closest thing the field has to a shipped commercial offering. Between them, the two companies have absorbed well over two billion dollars in funding. And yet, as Brandom puts it, both “rank pretty low on the trying-to-make-money scale.”

What he actually asked

The closest thing Brandom found to an authority on the record was Michael Rabbat, a co-founder of AMI Labs and the company’s VP of World Models, who joined him on the panel. Pressed on what exactly the company was building, Rabbat was cagey: “We’ll talk about it when we’re ready to talk about it.” Over email, he went slightly further: “We’re still in a research and building phase, so we’re not talking publicly about any product plans or timeline.”

To be fair, AMI Labs is less than a year old as a company. LeCun’s departure from Meta and the venture’s $1.03 billion seed round — at a $3.5 billion pre-money valuation, the largest seed round in European history — only became public in March 2026. A research lab in month nine keeping its powder dry is not a scandal. But Brandom’s point is that the caginess extends across the entire space, not just to one young company.

World Labs’ Marble, launched in November 2025 as the company’s first commercial product, is probably the most fully developed offering in the field. Its demos range from straightforward media creation to building explorable environments for video games and CGI effects, with robotics use cases in the mix as well. But Brandom’s read is that the platform “seems more designed to demonstrate capabilities” than to serve as a focused commercial product. It is a proof of concept with a pricing page.

The suppliers don’t know either

The most striking detail in the piece is that the secrecy extends downstream to the companies’ own suppliers. On the sidelines of the conference, Brandom spoke to Alex de Vigan, CEO of Physicl, a startup that supplies simulation-ready 3D data to the world model business — a data layer for physical AI, built by a team with roots in the e-commerce imagery company Nfinite. De Vigan says he knows Physicl’s data has been useful for whatever the labs are building. He just doesn’t know what that is.

“I wish they would tell us more. We could build more useful data if we knew what they were working on,” de Vigan told him.

That is an unusual posture for a supplier relationship. Data vendors usually compete on knowing their customers’ pipelines better than the customers do. Here, the vendors are flying blind — shipping generic sensor data and digitized objects into a demand signal they can’t see, which means the entire supply chain for the world model economy is optimizing blind.

Why the fog is rational

Part of the mystery, Brandom notes, comes from how versatile world models are as an idea. The simplest version is a navigable map of the world, similar to the AI models that power self-driving cars. But the same modeling approach that helps a Waymo weave through traffic could also help a humanoid robot carry boxes, or turn a few minutes of video footage into an explorable environment. AMI Labs has already publicly dipped its toe in manufacturing, biomedicine, robotics, and AI software for doctors — the last through its exclusive partnership with the French health-tech startup Nabla, which has first access to AMI’s emerging world model technologies and aims to build FDA-certifiable agentic AI for healthcare on top of them. Surely the company won’t pursue all of those directions. But maybe one or two of them are quietly standing out, and nobody outside the building knows which.

And there is a hard-nosed strategic reason not to say. As long as fundraising is easy, there is no particular pressure to focus on one market. Worse, focusing publicly is dangerous: if AMI announced tomorrow that it had built a humanoid robot stack or a next-generation Hollywood rendering system, a lot of other labs would suddenly be very interested in the space. The company would face potential competition from the other world model companies, from the “neolabs,” and even from OpenAI and Anthropic, both of which have the capital to enter any adjacent market the moment it looks winnable.

In some ways, Brandom writes, it’s the flip side of all that easy fundraising: your competitors can fundraise too, and the same money that lets you build under the radar is also funding lots of potential rivals once the path to market becomes clear. Even if that competition is inevitable, it’s best to delay it for as long as possible — which means keeping quiet about exactly what you’re building.

He closes with a metaphor that will land with any science fiction reader: a dark forest scenario, borrowed from Cixin Liu. If you don’t know who else is in the woods, it’s best not to attract attention.

The context the piece assumes

A few numbers make the secrecy more explicable. AMI Labs’ $1.03 billion seed came together in roughly six months after LeCun’s exit from Meta, with the FT reporting the round values the company at $3.5 billion pre-money — extraordinary for a lab with no product and no stated timeline. World Labs, founded on Li’s spatial intelligence thesis, raised a $1 billion round in February 2026 led by Autodesk’s $200 million strategic investment, on top of its earlier rounds. Marble, launched November 2025, and Atlas, its multimodal world model, give it the most public-facing portfolio in the field.

Nobody doubts there are viable businesses to be built on world model technology — robotics, autonomous navigation, interactive video, simulation for CGI and industrial design are all plausible endpoints. And as long as capital keeps arriving without revenue milestones attached, the rational move for every player is ambiguity: stay unpinned to any one market until the physics of competition force a reveal.

The quiet implication of Brandom’s reporting is less comfortable for the rest of the industry. If the two best-funded labs in a field can raise two billion dollars while declining to state what they’re building — and while their own data suppliers guess at the use case — then the world model sector is currently priced on vision and pedigree rather than on products. That can hold for a long time in a hot market. But the dark forest only stays dark until one player decides the best move is to turn on the lights.