WORLD LABS

Fei-Fei Li is not a random founder. She is the Stanford professor who created ImageNet, the giant labeled image dataset that helped set off the deep-learning boom in 2012.

People call her the godmother of AI, and she ran AI at Google Cloud for a stretch. In 2024 she started World Labs around a single idea she calls spatial intelligence.

Her argument is that today's AI can describe a picture but does not truly understand three-dimensional space the way humans and animals do. She recruited top researchers, including Ben Mildenhall, a co-creator of the NeRF technique for building 3D scenes, and Justin Johnson.

The company came out of stealth in September 2024 already worth over a billion dollars.

World Labs builds AI models that understand and generate 3D worlds. The plan is to sell that capability to the people who need virtual environments.

Think game studios, film and visual effects, designers, robotics companies, and anyone building for virtual or augmented reality. Its first product, Marble, lets a user type a prompt or upload an image and get back a full 3D scene.

In plain terms, World Labs wants to be the engine that turns ideas and photos into explorable worlds, and charge the creators who use it.

Marble is the first product. It generates full, explorable 3D worlds from a text prompt or a single image.

Under the hood, World Labs is building what it calls large world models, AI systems meant to understand and create three-dimensional space rather than just flat pixels or text. The company frames all of it under one goal, spatial intelligence.

Everything it ships is aimed at making machines understand the physical world in 3D.

The strategy so far is reputation and research firepower, not a splashy consumer launch. Fei-Fei Li's name alone pulled in $230 million and a $1 billion valuation before there was a product to try.

World Labs used that to hire elite 3D and computer-vision researchers, the people who literally invented some of the field's key techniques. Then it shipped Marble to prove the technology was real, not just a pitch.

The bet is that spatial intelligence becomes a foundational layer for AI, the same way image recognition did, and that being early and credible wins the market.

Spatial intelligence is genuinely hard, and World Labs is selling a future that does not fully exist yet. Building AI that understands 3D space is far less mature than text or image models, so the technical risk is real.

The company also raised enormous money at enormous valuations, reportedly heading toward $5 billion, which sets sky-high expectations. It has to turn a big research idea into products that game studios, roboticists, and designers will actually pay for.

Being early to a real trend is great. Being early to one that arrives slower than your funding assumes is dangerous.

The money came fast and from the top of Silicon Valley. World Labs raised $230 million out of stealth in 2024 at a valuation over $1 billion.

Andreessen Horowitz, NEA, Radical Ventures, and Nvidia's venture arm all backed it. NEA led a $100 million slice of the early funding.

By early 2026 the company was reportedly in talks to raise more at around a $5 billion valuation. Total funding has climbed past $1.2 billion.

When this many blue-chip funds pile into a company this young, they are buying the founder and the idea as much as the product.