AMD agrees to buy World Labs for $8.2 billion in stock as Fei-Fei Li joins as chief scientist

Advanced Micro Devices announced on September 28, 2026 that it has entered into a definitive agreement to acquire World Labs, the San Francisco startup co-founded and led by Dr. Fei-Fei Li, in an all-stock transaction valued at approximately $8.2 billion. The deal brings a team of AI model researchers into the Santa Clara chipmaker and installs Li as executive vice president and chief scientist, reporting to AMD chair and chief executive Dr. Lisa Su.

World Labs is barely two years old. Li founded the company in 2024 to build what the industry calls world models: systems that do not merely generate text or images but reconstruct, simulate and reason about three dimensional physical space. Based in San Francisco, World Labs develops 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. In November 2025 it shipped its first commercial product, Marble, which turns a text prompt, a photo or a video into an editable 3D environment, with paid plans running up to $95 a month.

The purchase lands in the middle of the chip industry race toward physical AI, the branch of machine learning aimed at autonomous machines such as robots and self-driving cars. Nvidia, AMD long-standing rival, holds the dominant position in AI training and inference hardware and has pushed its own physical AI effort through Cosmos, its world-model platform for robotics and simulation. For AMD, the deal is its second-largest acquisition ever, behind the roughly $50 billion Xilinx purchase that closed in 2022.

Li is one of the most recognizable names in the field. A Stanford computer science professor, she built ImageNet, the visual database and the AI competitions it inspired, work that helped make modern deep learning possible, and she is sometimes called the godmother of AI. The deal brings together two of the most prominent women in the industry, AMD Su and World Labs Li, both of whom argue that the next wave of AI will be grounded in physics rather than text alone.

Key Facts

Under the terms AMD disclosed, the transaction is all stock and valued at approximately $8.2 billion. It is expected to close by the end of 2026, subject to regulatory approvals and other customary closing conditions. After the close, the World Labs team will keep working on AI model research, and Li joins AMD as executive vice president and chief scientist, reporting to Su. Neither company disclosed how many employees would transfer.

The price is a steep markup on the startup last private valuation. Startup Fortune reported on September 29 that AMD backed World Labs $1 billion funding round just seven months earlier at a $5 billion valuation, and that the February 2026 round drew investors including AMD, Autodesk, Emerson Collective, Fidelity Management & Research Company, Nvidia and Sea. Autodesk separately said it invested $200 million. World Labs did not disclose the valuation at the time; Bloomberg had reported in January that it was in talks to raise at about $5 billion, and PitchBook put it near $5.4 billion.

Fortune reported on September 28 that World Labs had raised roughly $1 billion in total and that AMD shares were roughly flat in after-hours trading on Monday. Fortune also reported that AMD and World Labs began a deep technical partnership last year covering model training and inference optimizations on AMD GPUs, and that Li appeared as a guest at AMD CES presentation earlier this year.

TechCrunch reported on September 28 that World Labs justified the deal by saying AI development required close collaboration across model research, systems and compute, and that AMD believes understanding frontier workloads like those created at World Labs will shape its chip-making roadmap. The same report noted that world model remains a loosely defined term, and that Marble is pitched as a tool for creating entertainment experiences and simulated environments for robot training.

AMD said in its September 28 press release that building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving, and that the World Labs team brings exceptional research leadership and model expertise. Li said joining the company gives her team the resources and engineering depth to accelerate its research. On the product side, World Labs released Atlas on September 1, 2026, a world model built to reconstruct navigable 3D scenes from just a handful of photos and to eventually power robot training, following the arrival of Marble in November 2025.

Analysis

What this really means is that AMD has concluded it cannot design the next generation of AI chips from the outside looking in. The company own framing, that understanding frontier workloads will shape its chip roadmap, is effectively an admission that hardware plans are now written by model researchers. The $8.2 billion does not buy AMD a revenue stream of any consequence. What it buys is a seat inside the research loop, where decisions about memory, interconnect and inference efficiency are made before they harden into product requirements.

The bigger picture here is that physical AI has become the newest land grab in semiconductors, and the two companies best positioned in it are now on opposite sides of an ownership question. Nvidia holds the dominant position in AI training and inference hardware and has built Cosmos as its world-model platform for robotics and simulation, yet it was also an investor in the World Labs round of February 2026. Startup Fortune reported on September 29 that neither AMD nor World Labs disclosed how existing investors like Nvidia will be treated as the deal moves toward its close, a detail that determines whether a direct competitor keeps a financial stake in the technology AMD now intends to build its roadmap around.

The valuation is aggressive but not obviously irrational. Paying about $8.2 billion for a company that raised roughly $1 billion and last priced near $5 billion is a premium of well over 50 percent on the most recent reported mark, and it comes for a business whose commercial product sells paid plans topping out at $95 a month. On any near term revenue math, AMD is not buying a business. It is buying a two-year-old research organization, a founder with unusual name recognition, and model capabilities its own software stack has struggled to attract organically.

The execution risk sits in the same place as the strategy. The World Labs team is expected to keep doing model research after the close, so AMD is grafting a research culture onto a chip company rather than folding an asset into an existing unit, and the two firms have not said how many people are involved. Fortune reported on September 28 that AMD shares moved roughly flat in after-hours trading, which suggests investors read the deal as a strategic necessity rather than an immediate earnings event.

Why It Matters

For AMD, the stakes go beyond a single acquisition. The company has spent years fighting for a larger share of AI accelerators against a rival that also controls the software layer most developers rely on. Owning a model lab gives AMD an internal customer for its GPUs and a way to co-design silicon against workloads it understands at the source, instead of reacting to benchmarks published by others.

For the wider world-model field, the deal is a signal about consolidation. World Labs was one of several richly funded startups betting on world models, and it was led by the researcher most associated with the dataset that made deep learning practical. When a company that young is absorbed at that price, smaller labs working on spatial intelligence and robotics simulation gain a new reference point for what an exit looks like.

For Li, the move shifts her from running a startup to shaping the research agenda of a large public semiconductor company. She described it as a matter of resources and engineering depth, which is a candid way of naming the constraint most model labs hit: the ideas are rarely the bottleneck, the compute and the systems engineering around them are.

Next Up

The immediate step is regulatory review. AMD expects the transaction to close by the end of 2026, a short window for approvals in multiple jurisdictions and for the two companies to settle the treatment of existing World Labs investors, including Nvidia. Watch for disclosures about employee retention and about how the World Labs team will be organized inside AMD, both of which were left open at announcement.

The longer term question is whether Atlas, released on September 1, 2026, becomes a genuine foundation for robot training at scale. If it does, AMD will have bought its way into the workload that defines the next hardware generation. If it does not, the company will have spent $8.2 billion in stock on a research bet that its archrival is already funding from the other side of the table.