The hottest corner of AI right now has a strange problem: nobody will say what they’re selling. TechCrunch AI moderated a world models panel at the All In conference this week and came away with more questions than answers. The two flagship players, Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs, have plenty of buzz and funding but, as TechCrunch AI puts it, “rank pretty low on the trying-to-make-money scale.”
That’s not an accident. It’s a strategy.
What world models actually are
Quick primer. A world model is an AI system that understands physical space: how objects sit in a room, how they move, what happens when one thing pushes another. Think spatial intelligence rather than language intelligence. The same core capability could power a self-driving car, a warehouse robot, or a tool that turns a few minutes of phone video into an explorable 3D scene.
That versatility is exactly why the field feels foggy.
The panel that went nowhere
Michael Rabbat, AMI Labs co-founder and VP of World Models, joined the panel. When TechCrunch AI pressed him on what the company is building, he said: “We’ll talk about it when we’re ready to talk about it.” Over email he added that AMI is “still in a research and building phase” with no public product plans or timeline.
AMI is less than a year old, so some quiet is fair. But the pattern repeats across the whole space:
- World Labs’ Marble is the most developed product out there, with demos covering game environments, CGI effects, and some robotics. TechCrunch AI notes the platform seems built to show off capabilities rather than solve one customer’s problem.
- AMI has already dabbled in manufacturing, biomedicine, robotics, and clinical AI software through a partnership with Nabia. That’s a lot of open doors for a company with no stated product.
- Even suppliers are in the dark. Alex de Vigan, CEO of data provider Physicl, told TechCrunch AI: “I wish they would tell us more. We could build more useful data if we knew what they were working on.”
When your own data vendor doesn’t know what you’re doing, that’s secrecy by design.
Why the silence makes sense
The logic is simple, and I think TechCrunch AI nails it. As long as fundraising stays easy, there’s no pressure to pick a lane. And there’s a real cost to picking one publicly.
Imagine AMI announced tomorrow that it had cracked humanoid robot control or a next-gen Hollywood rendering pipeline. Every other world model lab, the neolabs, plus OpenAI and Anthropic would pile in within weeks. The same capital that lets you build quietly is also sitting in your rivals’ bank accounts, waiting for someone to prove a market exists.
TechCrunch AI compares it to the “dark forest” idea from Cixin Liu’s novels: if you don’t know who else is in the woods, don’t make noise. Competition is inevitable. Delaying it is the whole game.
What stands out to me is how different this is from the LLM race. OpenAI, Anthropic, and Google ship in public and fight on benchmarks. World model labs are doing the opposite: raising like a frontier lab, shipping like a stealth startup.
What the next 1-3 years probably look like
My read on where this goes:
- Funding tightens, focus follows. The secrecy holds as long as investors accept “research phase” as an answer. Once one lab needs revenue to justify its next round, it picks a vertical and the fog lifts. I’d expect that within 18 to 24 months.
- Robotics is the likely first real market. Humanoid and warehouse robots need spatial intelligence more urgently than Hollywood does, and those buyers have budgets. Media and gaming will be the demo reel. Robotics will be the invoice.
- The big labs move the moment someone proves demand. Google, OpenAI, and Anthropic all have video and multimodal models one step away from a world model. The first public commercial win triggers the dogpile.
- Data suppliers get squeezed. Companies like Physicl are building blind today. Once use cases go public, expect labs to either demand specialized data or bring collection in-house.
What to do with this
If you’re building on AI:
- Don’t wait for the labs to publish a roadmap. If your product touches robotics, simulation, or 3D content, start testing Marble and any open world model weights now, before the APIs get productized and priced.
- Watch hiring, not press releases. Job listings at AMI and World Labs will reveal the chosen vertical long before an announcement does.
- If you sell data or tooling, build for the two or three most likely applications rather than a generic “world model” customer. Robotics and simulation data are the safer bets.
The labs are quiet because quiet works. That won’t last forever, and the first one to break silence will tell everyone where the money is. TechCrunch AI has the full panel writeup for readers who want more.