Via theverge.com
Fei-Fei Li's spatial intelligence startup is betting that simulated 3D worlds can solve one of robotics' most expensive problems.
Teaching a robot to navigate your living room without breaking your coffee table is surprisingly expensive. The traditional approach requires mountains of real-world data, physical trial runs, and the kind of patience that venture capital funding timelines don’t typically afford. World Labs thinks it has a better way.
The spatial intelligence company, co-founded by AI pioneer Fei-Fei Li in 2024, acquired SceniX on July 21, bringing high-fidelity simulation technology in-house to supercharge its approach to robot training. The deal folds SceniX’s expertise into World Labs’ existing Marble platform, which generates photorealistic, dynamic 3D environments from text and image inputs.
Why simulated worlds matter more than you think #
World Labs is attacking this bottleneck by creating digital environments where robots can fail spectacularly, learn from it, and try again, all without bending a single piece of metal. The Marble platform doesn’t just render pretty pictures. It builds interactive 3D scenes that simulate physics, lighting, and object behavior with enough fidelity that skills learned in simulation can transfer to physical robots.
The SceniX acquisition deepens this capability considerably. SceniX brought experienced co-founders and specialized simulation expertise that complement Marble’s generative approach. The combined technology stack enables what the industry calls “real-to-sim-to-real” training, a pipeline where real-world environments get digitized, robots train inside the digital twin, and then the learned behaviors get deployed back into physical space.
The funding story behind the platform #
World Labs hasn’t been shy about raising capital. The company pulled in $230 million in funding during 2024, its founding year, pushing its valuation past $1 billion almost immediately.
Then in February 2026, Autodesk made a $200 million strategic investment, signaling that the design software giant sees World Labs’ simulation tech as complementary to its own 3D modeling ecosystem. That brings total known funding to at least $430 million.
Fei-Fei Li’s involvement adds credibility that money alone can’t buy. As the former director of Stanford’s AI Lab and one of the researchers behind ImageNet, the dataset that essentially kickstarted the deep learning revolution, Li brings both technical authority and a Rolodex that spans academia, government, and Silicon Valley’s top-tier investors.
What this means for the AI investment landscape #
The robotics training bottleneck isn’t just a World Labs problem. It’s an industry-wide constraint that affects everyone from Amazon’s warehouse automation division to the dozens of startups building humanoid robots.
Competitors will likely respond. Companies like NVIDIA, which already offers its Omniverse platform for industrial simulation, and Google DeepMind, which has invested heavily in sim-to-real transfer research, are natural rivals in this space.
The real risk for World Labs, and for investors betting on simulation-first robotics, is the sim-to-real gap. Digital environments, no matter how photorealistic, still differ from the physical world in subtle but consequential ways. A robot that performs flawlessly in simulation might stumble when confronted with a slightly different floor texture or unexpected lighting condition. Closing that gap is the central technical challenge, and the SceniX acquisition suggests World Labs knows it can’t do it alone.
With $430 million in known funding and a growing technology stack, World Labs is positioning itself as the default simulation layer for the robotics industry.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our