AMD 以 82 亿美元收购空间智能初创公司 World Labs，李飞飞出任首席科学家

Chipmaker AMD is acquiring World Labs, a startup focused on spatial intelligence. AI pioneer Fei-Fei Li will join AMD's leadership team as Executive Vice President and Chief Scientist.

The all-stock deal is valued at about $8.2 billion, according to AMD. The transaction is expected to close by the end of 2026, pending regulatory approval. Founder Li will report directly to CEO Lisa Su and lead frontier research at AMD. In a blog post, she wrote that scaling research demands tighter integration with hardware. "Without a focused hardware effort, AI is hobbled in efficiency," Li said.

World Labs was founded in early 2024 by Li along with researchers Ben Mildenhall and Justin Johnson in San Francisco and has about 70 employees. The startup builds so-called world models, AI software that can generate, reconstruct, and simulate three-dimensional environments from text, image, and video inputs. In early September, World Labs unveiled Atlas, its first world model. Atlas generates, reconstructs, and simulates complete spatial scenes from just a few images.

The idea behind World Labs is that language alone isn't enough for capable AI models. "The universe isn't made up of words; it's made of real things," Li wrote.

In a detailed essay, she recently laid out why today's multimodal language models fail at spatial tasks. They perform barely better than chance when estimating distances, orientations, and sizes. They can't mentally rotate objects or predict basic physics. Many of the most important problems AI is supposed to solve, from science to entertainment to robotics, need models that can reason about the structure and behavior of the physical world, Li argued.

Some see world models as the next big bet after pure language models. Google DeepMind co-founder Demis Hassabis, for example, considers them a critical step toward artificial general intelligence. An international research team led by Peking University has proposed OpenWorldLib, a standardized definition that requires three abilities. World models must perceive their environment, interact with it, and remember past interactions.

AMD wants model expertise to close the gap with Nvidia

For AMD, the acquisition is about gaining a deeper understanding of how AI workloads are built and using that knowledge to shape future hardware roadmaps. "The more you understand end to end, the better system you are going to build," AMD CEO Lisa Su said in an interview with Bloomberg Television. That's why the company bought World Labs, she said.

The deal reinforces AMD's position as Nvidia's most serious challenger in the AI chip market, even though the two companies aren't remotely in the same league financially. AMD has a market cap of about $1 trillion. Nvidia sits at roughly $5 trillion.

Nvidia has been bundling AI chips with models and services for years, releasing open-weight models and building out its own services to fuel an ecosystem that ultimately runs on its hardware. Its recent $13 billion acquisition of AI platform Hugging Face doubled down on that strategy.

Before the acquisition, World Labs had closed a $230 million Series A in 2024 with investors including Andreessen Horowitz, Nvidia, and AMD. In early 2025, the company raised another $1 billion from backers like Autodesk, with Bloomberg pegging the valuation at about $5 billion. The $8.2 billion acquisition price means the startup's value multiplied several times over within just a few months.

The sale may also be an admission that the money World Labs had raised, as impressive as it sounds, was nowhere near enough to keep pace with the leading AI labs and tech giants that are spending at a breakneck clip right now. AMD's announcement reads less like the company was buying World Labs for its still-conceptual products and more like it was paying for the team's expertise.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.