This week I moderated a panel discussion about world models at the All In conference (not related to the podcast). This gave us the opportunity to delve into one of the most mysterious corners of the world of AI. The leading companies in this field are Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs. Both have amassed a lot of buzz and funding, but rank pretty low on the money-making scale.
Because the core of the World Model is about automating spatial intelligence, this field could go in many exciting and lucrative directions, from robotics to interactive video to more complex self-driving systems.
But when we started thinking about how this technology would actually be commercialized, things got foggy. The closest I could come to an authority was Michael Rabbat, co-founder of AMI Labs and the company’s VP of World Models, who joined us on the panel. But when I asked him what exactly the company was working on, he was adamant. “When we’re ready to talk about it, we’ll talk about it,” he clarified in an email. “We’re still in the research and construction phase, so we don’t talk publicly about product plans or timelines.”
To be fair, AMI is less than a year old, so it’s worth keeping quiet about. But this kind of fickleness extends to the entire world modeling space. World Labs’ Marble is perhaps the most fully developed product in this space, with demos ranging from simple media creation to building explorable environments for video games and CGI effects. There are also robot use cases, but the entire platform seems designed to demonstrate functionality.
The secret extends to the suppliers of these companies. On the sidelines of the same conference, I spoke to Alex de Vigan, CEO of Physicl, a data supplier to fast-growing global model businesses. He says he knows Physicl’s data is useful for everything they’re building, but he doesn’t yet know exactly what that is. “I wish they would have told us more. If we knew what they were working on, we could have built more useful data,” de Vigan told me.
Part of the puzzle comes from how versatile world models are as ideas. The simplest version is a navigable map of the world, similar to the AI models that drive self-driving cars. But the same modeling approach that helps Waymo weave through traffic could also help humanoid robots carry boxes or turn minutes of video footage into an explorable environment. AMI is already dipping its toe into manufacturing, biomedical, robotics, and even AI software for physicians through its partnership with Nabia. Of course, you won’t pursue them all, but maybe one or two stand out to you?
No one doubts that there are many viable businesses that can be built on the technology of the world model. As long as funding is easy, there’s no extra pressure to focus on one thing. In fact, there are good reasons not to. If AMI announced tomorrow that it had built a humanoid OpenClaw or next-generation Hollywood rendering system, many other labs would suddenly have a huge interest in this area. Soon, the institute will face potential competition from other global model companies, Neolab, and even OpenAI and Anthropic.
In a way, this is the flip side of easy funding. Competitors can also raise funds. And once the path to market is clear, the same funding that allows you to build under the radar will also fund many potential rivals. But even if that competition is unavoidable, it’s best to delay it for as long as possible. That means staying silent about what you’re building.
Cixin Liu fans will recognize this as a dark forest scenario. If you don’t know who else is in the forest, it’s best not to attract attention.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
