Startups building “world models” — AI systems designed to understand spatial environments — are shrouded in secrecy. Despite major funding and high-profile leadership, firms like AMI Labs (led by Yann LeCun) and World Labs (with Fei-Fei Li) offer vague public signals about what’s coming next. Their research may span robotics, interactive video, gaming, self-driving vehicles, and even biomedicine, but details on actual products or timelines are scarce. (These companies are known more for buzz than revenue right now.)
What are world models — and what are they building?
At heart, world models aim to automate spatial intelligence: creating AI that can map, simulate, and interact with real or simulated environments. Whether it’s a self-driving system navigating city streets, a robot moving objects in a factory, or software turning video footage into navigable 3D spaces — many use cases are possible. AMI Labs already has exploratory projects in robotics, manufacturing, and even healthcare through its Nabia partnership. World Labs showcases its product Marble for media creation and CGI or game-engine environments. But neither firm has disclosed firm launch dates or which verticals will end up being primary bets.
Why all the mothballs?
The silence isn’t accidental. Both startups are relatively new and still refining their tech. Moreover, there’s strategic value in hiding what you’re working on. AMI’s executives say they’ll speak more when they’re ready. Meanwhile, suppliers like Physicl — which provide data inputs to world model firms — are kept in the dark about downstream applications. One supplier noted they could improve data if they knew what their clients were building.
There’s also competitive logic at play: staying vague helps avoid giving rivals a roadmap. As firms raise capital easily in the current investment climate, there’s less pressure to move fast, but more risk that competitors (including neolabs or established AI labs like OpenAI or Anthropic) will poach any revealed idea. It’s a kind of “Dark Forest” mindset: better to stay hidden until you’re ahead.
Finally, being generalist helps — especially when you’re not yet sure which vertical will pay off. By signaling involvement in many areas, these labs keep options open without committing publicly to one trajectory.
Still, that opacity comes with tradeoffs. Without clear roadmap, recruiting talent, forming partnerships, and achieving regulatory clarity all get harder.
What to watch: who actually ships. In coming months, we’ll be looking for world model companies to name their first prominent products, stake claims in specific industries, or show how their research translates to revenue. Until then, most of this space is defined as much by secrets as solutions.
The blur between promise and proof matters. While world models are one of the most compelling emerging technologies, the gap between theoretical capability and commercial reality remains wide. Those who can tame that silence and deliver will reshape fields across robotics, video, healthcare, and more. Watch for the first world model breakthroughs to move from research papers and demos into real-world impact.
