AI · Sep 17, 2026
Google DeepMind launches institute to widen the AGI debateWorld model companies are keeping a lot of secrets
Sep 18, 2026, 1:18 PM · TechCrunch

AMI Labs and World Labs are raising and buzzing while staying silent on products—Russell Brandom calls it a dark-forest strategy while fundraising stays easy.
Why it matters
Russell Brandom’s TechCrunch dispatch from an All In conference panel on world models finds the category rich in funding and thin on commercialization talk. The marquee names are Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs—buzz-heavy, still low on the “trying to make money” scale.
World models aim at spatial intelligence with paths into robotics, interactive video, and more complex self-driving. AMI co-founder and VP of World Models Michael Rabbat stayed cagey on product plans: “We’ll talk about it when we’re ready,” later emailing that the company is still in research and building with no public product timeline. AMI is less than a year old.
World Labs’ Marble is the most developed product cited, with demos spanning media creation, explorable game environments, and CGI—plus robotics angles that still feel capability-demo-first. Even suppliers are dark: Physicl CEO Alex de Vigan says his spatial data helps customers but he wishes they would say what they are building so he could ship more useful datasets. AMI has dipped into manufacturing, biomedicine, robotics, and doctor-facing software via a Nabia partnership without naming a single beachhead.
From the desk
We’re fine with research silence in year one—and we’re not fine treating secrecy as a substitute for a product thesis forever.
Spatial intelligence is real and useful: robots, maps, simulators. Easy fundraising removes the forced focus that usually picks a customer. Brandom’s dark-forest read is sharp: announce a humanoid stack or Hollywood renderer and every neolab plus OpenAI and Anthropic suddenly cares. Delay clarifies nothing for partners like Physicl who could make better data if they knew the job.
Useful AI in this lane will look like boring vertical wins—warehouse motion, editable 3D from video—not perpetual demo reels. I’m watching for AMI’s first named product and whether Marble moves from showcase to paid production workflows before the fundraising weather changes.
Context
TechCrunch published on September 18, 2026, after Brandom moderated the All In conference panel (unrelated to the podcast of the same name).
Who feels it
- Robotics and AV teams
- Treat world-model demos as research until latency, failure modes, and licensing are public.
- Data suppliers
- Opaque customer roadmaps make it harder to prioritize collection pipelines.
- Investors
- Easy capital delays focus—and funds the rivals who will pile in once a path is clear.
What to watch
- AMI’s first public product or timeline statement.
- Whether Marble lands paying production customers beyond demos.
- OpenAI/Anthropic moves into world-model product lines once a beachhead is obvious.
Also covering this
AI · Sep 20, 2026
World model companies are keeping a lot of secrets