AMD Agrees to Buy Fei-Fei Li’s World Labs for $8.2 Billion to Steer Its Chip Roadmap

AMD said Monday that it has agreed to acquire World Labs, the world-model research lab Fei-Fei Li co-founded in 2024, in an all-stock transaction valued at about $8.2 billion. The companies expect the deal to close by the end of 2026, subject to regulatory approvals and other customary closing conditions. Li will join AMD as executive vice president and chief scientist, reporting to CEO Lisa Su, and co-founders Justin Johnson and Ben Mildenhall will keep leading the World Labs team, according to AMD’s announcement and World Labs’ own post. CNBC reports that it is AMD’s second-largest acquisition on record, behind the roughly $50 billion Xilinx deal in 2022.

What AMD Says It Is Buying

The rationale in AMD’s release is about workloads, not a product line. “As AI expands into reasoning, robotics, simulation and physical AI, the demands on compute infrastructure become more diverse,” the release says, and World Labs’ expertise “will give AMD deeper insight into how workloads are evolving and help shape its future technology roadmaps.” Su framed it the same way: “Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving.”

Analysts read the deal along similar lines. Tech analyst Patrick Moorhead called it “a talent and model-insight buy, not a revenue play,” adding that “AMD gets the people building the models its future chips will have to run,” in comments published by Yahoo Finance. Citigroup analyst Atif Malik, quoted in the same piece, listed three likely motives: acquiring more leading AI engineers, positioning AMD better in physical AI, and building better end-to-end systems. CNBC adds that AMD plans to keep World Labs separate from its chipmaking business until the deal closes.

What World Labs Builds

AMD describes World Labs as a developer of spatial-intelligence models that “generate, reconstruct and simulate interactive 3D environments from text, image and video inputs,” along with technology for robotic learning and simulation. Its first product, Marble, creates persistent 3D worlds from images, video or text. In her own post about the deal, Li says a recently released model, Atlas, predicts the next camera view from a set of 2D images much as a language model predicts the next token, and that the company’s acquisition of SceniX is building it toward robotics simulation.

The robotics angle matters to a chipmaker. TechCrunch notes that synthetic data from world models is seen as key to training general-purpose robots, and Ars Technica reports that this is a field where Nvidia has a strong lead. AMD has released its own open-source world model, Micro-World, Ars says, but it has no stack that competes with Nvidia’s Cosmos models and applications.

A Partner and Investor Becomes the Owner

The two companies already knew each other well. World Labs says it began a “deep technical partnership” with AMD last year, starting with model training and inference optimization on AMD GPUs, and Li writes that Su “was an early investor and believer in our mission.” When World Labs announced a $1 billion round in February, it named both AMD and Nvidia among its investors, alongside Autodesk, Emerson Collective, Fidelity and Sea. TechCrunch also notes that Li was a guest at AMD’s CES presentation earlier this year, and CNBC describes a demo she and Su presented in which Marble turned a few images into a 3D scene.

The price is harder to benchmark. World Labs emerged from stealth in 2024 with $230 million at a $1 billion valuation, TechCrunch reported in February, and it declined to say whether the new round raised that figure. Reports a month earlier had suggested it was aiming to raise at a $5 billion valuation. If the company did raise at that level, the $8.2 billion deal value is roughly 64 percent higher, about seven months later. AMD’s release gives the dollar figure and the all-stock structure, but not the share count, the exchange ratio, or how World Labs’ existing investors are treated.

How It Fits With AMD’s Other AI Deals

The purchase follows other recent AMD moves across the AI stack. On July 22, AMD and Anthropic announced a partnership to deploy up to 2 gigawatts of Instinct MI450 Series GPUs, with the first gigawatt starting in the first half of 2027, and AMD committed to a strategic equity investment of up to $5 billion in Anthropic in the future. On August 6, AMD announced a definitive agreement to acquire Taalas, a Toronto company building specialized AI inference silicon. Taken together with World Labs, those moves span a major customer, a chip-design bet and model research.

The Anthropic commitment follows the familiar pattern in which a chip supplier takes a stake in a customer, the reverse of the structure in which Akamai is giving Anthropic equity, which we covered in Anthropic’s $11.6 billion Akamai deal.

Where This Leaves Nvidia

Nvidia already offers open-weight world models under the Cosmos name, TechCrunch notes, while AMD has so far offered only text and video models to the public. Fortune adds that Nvidia has built its own world models and partnered with startups such as Wayve, and Nvidia was itself named among World Labs’ investors in February. The rivalry is playing out partly through deals: CNBC puts Nvidia’s combined spending on Groq and Hugging Face at $33 billion, and we covered the first reports of the Hugging Face deal in August. For technical background on Nvidia’s world-model work, see our look at Cosmos-H-Dreams and real-time surgical simulation.

What Is Still Unknown

Neither announcement says what happens to World Labs’ existing products. AMD’s release says the team will keep focusing on AI model research after the deal closes, but it does not mention Marble or any other product. Both sides stress an open ecosystem, and World Labs says it is committed to building out “widely accessible open models,” but none of the announcements defines open, names a license, or says whether future World Labs models will stay hardware-neutral or be tuned first for AMD GPUs. Which regulators must sign off is not specified either. The Next Web notes that Li’s own framing is about hardware: “Without having a focused hardware effort, AI is hobbled in efficiency. And scale.”

Fortune reported that AMD shares were roughly flat in after-hours trading Monday, when the deal was announced. The real test is whether World Labs’ research shows up in AMD’s silicon and software roadmaps, and that will take years, not quarters, to answer.