The startup, co-founded by Stanford's Fei-Fei Li, is racing to build AI systems that understand three-dimensional space—a frontier that could reshape everything from video games to robotics.
When Fei-Fei Li emerged from her Stanford lab last year to co-found World Labs, the AI community took notice. Now, barely 18 months later, the spatial intelligence startup has closed a $1 billion funding round—one of 2026's largest AI deals—with an unusual anchor: Autodesk, the design software titan that's been quietly watching the generative AI boom from the sidelines.
Autodesk's $200 million stake, disclosed February 18, signals more than financial backing. The 42-year-old company, whose AutoCAD and Maya tools have long dominated architecture and animation studios, is positioning itself as a strategic adviser to World Labs. The collaboration, according to both companies, will focus initially on entertainment workflows—though the partnership's scope suggests broader ambitions.
"If AI is to be truly useful, it must understand worlds, not just words," Li said in Autodesk's announcement. It's a framing that distinguishes World Labs from the language-model arms race that has consumed much of the industry's attention and capital over the past three years.
The Bet on "World Models"
World Labs builds what it calls Large World Models, or LWMs—foundational AI systems designed to perceive, generate, and interact with persistent three-dimensional environments. Think of it as GPT-4 meets a physics engine, capable of understanding not just text but geometry, lighting, spatial relationships, and how objects behave across time.
The concept isn't entirely new. Google DeepMind has been working on its Genie program, and Luma AI—which has reportedly raised roughly $900 million at a valuation north of $4 billion—is chasing similar territory. But World Labs is making an audacious technical bet: that true spatial intelligence requires linking perception, reasoning, generation, and interaction within 3D space. It's the kind of grand, foundational claim that either ages brilliantly or becomes a cautionary tale.
The team Li assembled reads like an AI research dream roster. Co-founders include Justin Johnson, formerly an assistant professor at the University of Michigan; Christoph Lassner, a neural rendering specialist; and Ben Mildenhall, who helped create NeRF (Neural Radiance Fields) during his time at Google Research. According to LinkedIn, the San Francisco-based company employs somewhere between 11 and 50 people—a remarkably lean operation for a firm now valued in the billions.
From Stealth to Market in Record Time
World Labs has moved with unusual speed. The company surfaced from stealth in September 2024 with $230 million already in the bank, backed by Andreessen Horowitz, NEA, and Radical Ventures. That funding had come in waves: roughly $200 million while still operating in stealth mode in April 2024, then another $100 million or so in a July–August round led by NEA that pushed the company past unicorn status.
By November 2025, World Labs launched Marble, a multimodal world model that converts text, images, video, or rough 3D sketches into fully editable three-dimensional environments. Users can export the results as Gaussian splats, triangle meshes, or video—formats that plug directly into existing creative pipelines. Two months later came the World API, a public interface that lets developers programmatically generate explorable 3D worlds and integrate them into tools like NVIDIA Isaac Sim, MuJoCo, and RoboSuite.
The company uses a credit-based pricing model: $1 buys 1,250 credits, with generation costs varying by input complexity. Early partners include Escape.ai in media, Preview for film pre-visualization, Fenestra in architecture, and xFigura for professional design. It's a deliberately curated roster, heavy on creative and technical users who can tolerate rough edges in exchange for cutting-edge capabilities.
Strategic Investors Circle
The latest round includes AMD, NVIDIA, Emerson Collective, Fidelity Management & Research Company, and Sea, structured as convertible preferred stock according to S&P Capital IQ. World Labs hasn't disclosed a post-money valuation, but Reuters reported that Bloomberg had noted January discussions around a roughly $5 billion figure—more than double the company's previous valuation just months earlier.
Perhaps more telling than the valuation is the investor composition. NVIDIA's presence makes sense; spatial AI demands serious compute, and World Labs selected Google Cloud as its primary infrastructure provider back in October 2024. AMD's involvement suggests hedging strategies on both sides. Cisco Investments, which joined in November 2025, hints at industrial and enterprise applications beyond the creative markets World Labs has emphasized publicly.
And then there's Autodesk. The company's participation represents a significant strategic shift. For years, Autodesk has watched scrappy startups and tech giants alike incorporate generative AI into design workflows. Now it's making a major bet that the next phase—true spatial intelligence—requires partnering with, rather than building in-house.
The collaboration will extend to "research and model" level work, whatever that means in practice. Initially, both companies say they'll focus on entertainment. But Autodesk's customer base spans architecture, engineering, construction, manufacturing, and product design. If World Labs' technology proves robust, the applications could extend far beyond virtual film sets.
The Embodied AI Endgame
Target markets World Labs has publicly named include storytelling and creative tools, design and simulation, robotics and embodied AI, and scientific discovery. That's an ambitious, perhaps overly broad, list. But the company's integrations with robotics simulators—NVIDIA Isaac Sim, MuJoCo, RoboSuite—suggest the founders are thinking beyond entertainment.
Embodied AI, the idea that intelligent systems need spatial understanding to operate in physical environments, has become Silicon Valley's next big thing. Tesla's Optimus, Figure AI's humanoid robots, and countless warehouse automation startups all depend on machines that can navigate and manipulate three-dimensional space. If World Labs can build foundation models that genuinely understand geometry, physics, and spatial relationships, the market opportunity extends well beyond Maya renders and Unity game engines.
Then again, this is the classic AI startup challenge: demonstrate value in tractable near-term applications while pursuing moonshot foundational research. World Labs is barely 18 months old as a public entity. Its technology, while impressive in demos, has yet to prove itself at production scale.
What Comes Next
With $1 billion in fresh capital—bringing total known funding to roughly $1.43 billion—World Labs now faces the grinding work of scaling. The company needs to expand its team significantly (11 to 50 employees won't cut it for long), refine its models, build enterprise-grade infrastructure, and navigate the complex politics of partnering with incumbents like Autodesk while competing with Google and well-funded peers.
The Autodesk relationship will be an early test. Strategic investments between startups and legacy giants often produce more press releases than actual product integration. If World Labs and Autodesk can ship something meaningful—a plugin that fundamentally changes how designers work, for instance—it would validate the spatial intelligence thesis in a tangible way.
Competition, meanwhile, is intensifying. Google DeepMind has vast resources and a head start in generative video and spatial understanding. Luma AI is well-funded and moving aggressively. Smaller startups are nipping at adjacent markets. And then there's the possibility that OpenAI, Anthropic, or other foundation model leaders simply extend their architectures into 3D—a scenario that would compress margins and accelerate the race.
For now, World Labs has runway, pedigree, and partnerships. Whether spatial intelligence becomes the next frontier in AI—or a promising research direction that takes longer to commercialize than the hype suggests—remains an open question. Li, who helped pioneer computer vision research over two decades at Stanford, is betting her reputation that understanding worlds, not just words, is where artificial intelligence needs to go next.
The venture investors who just wrote nine-figure checks are betting she's right. By this time next year, we'll have a clearer picture of whether that confidence is justified.
