AMD Buys World Labs to Boost AI Chip Battle With Nvidia

AMD acquires World Labs for $8.2 billion

AMD has agreed to acquire the artificial intelligence startup World Labs in a deal valued at $8.2 billion. The companies said the transaction should be finalized by the end of this year, pending the usual regulatory reviews. This move marks AMD’s latest effort to strengthen its position in the fast-growing AI hardware market, where Nvidia currently holds a clear advantage.

World Labs founded by Fei-Fei Li in 2024

World Labs was launched in 2024 by computer vision expert Fei-Fei Li, along with researchers Justin Johnson, Christoph Lassner, and Ben Mildenhall. The startup raised $230 million in early funding, and AMD participated as one of the backers. From the outset, the company focused on building what it calls world models-AI systems designed to simulate and predict aspects of the physical world. These models differ from traditional large language models because they are trained on rich video and sensor data to understand how objects move, interact, and change over time.

One of the first public offerings from World Labs was a tool named Marble. Marble uses a technique called Gaussian splatting to create small, detailed three-dimensional scenes. Users can export those scenes as standard 3D assets, making them useful for filmmakers, game developers, and other creators who need virtual environments. Since Marble’s debut, the team has added other models and tools, such as Atlas, which shift the emphasis toward different applications while still relying on the core idea of predictive world simulation.

AMD seeks synthetic data to train robots via World Labs

A major motivation behind the heavy investment in world-model research is the need for synthetic data to train robots. By generating realistic, varied scenarios in software, companies can teach robotic systems to handle edge cases without risking real-world hardware. Nvidia has already built a strong foothold in this area with its Cosmos platform and related tools, giving it a lead over rivals. AMD, while it has released its own open-source world model called Micro-World, has not yet assembled a full software stack that can compete directly with Nvidia’s offering.

The acquisition brings not only World Labs’ technology under AMD’s roof but also its key research talent. According to the announcement from both firms, Fei-Fei Li will take on the role of Executive Vice President and Chief Scientist at AMD. Justin Johnson and Ben Mildenhall will continue to guide the World Labs team as it integrates into AMD, working alongside Li to form what the companies describe as a frontier research organization. This structure is intended to keep the group’s innovative culture while aligning its work with AMD’s broader product strategy.

AMD's focus on reasoning robotics simulation physical AI after acquisition

AMD’s own statement about the deal highlights a shift in focus toward areas such as reasoning, robotics, simulation, and what it calls physical AI. The company argues that as AI expands beyond pure language processing into these domains, the demands on computing hardware become more varied and complex. By gaining direct access to World Labs’ expertise in building advanced predictive models, AMD hopes to better anticipate how future workloads will evolve and to shape its upcoming chip designs and software stacks accordingly.

Observers note that the timing of the purchase is sooner than many expected for a startup that is only a few years old. However, the prior collaboration between AMD and World Labs made the move less surprising. The two companies had already worked together on earlier projects, giving AMD a window into the startup’s capabilities. With Nvidia’s dominance in AI accelerators showing little sign of weakening, AMD is betting that combining its chip engineering strength with World Labs’ research prowess will help it close the gap.

For developers and engineers who build AI-driven applications, the deal could mean new tools and libraries that emerge from the merged team. Those working on robotics, simulation, or content creation may eventually see AMD-optimized versions of world-model-based software that run efficiently on the company’s GPUs and CPUs. At the same time, competitors will be watching closely to see how quickly AMD can translate the acquisition into tangible products that challenge Nvidia’s current lead. The coming months will reveal whether the $8.2 billion investment pays off in the form of faster, more versatile hardware for the next wave of AI applications.

Frequently asked questions

What is AMD acquiring and for how much?

AMD is acquiring the artificial intelligence startup World Labs in a deal valued at $8.2 billion. The transaction is expected to close by the end of this year, pending regulatory approvals.

Who founded World Labs and what is its core technology?

World Labs was launched in 2024 by computer-vision expert Fei-Fei Li, together with researchers Justin Johnson, Christoph Lassner, and Ben Mildenhall. The startup focuses on building “world models”-AI systems trained on video and sensor data to simulate and predict how objects move, interact, and change over time.

What are some of World Labs’ publicly released tools mentioned in the article?

One of World Labs’ first public offerings is a tool called Marble, which uses Gaussian splatting to create small, detailed three-dimensional scenes that can be exported as standard 3D assets. The team later added Atlas, another tool that shifts the emphasis toward different applications while still relying on predictive world simulation.

What role will Fei-Fei Li play at AMD after the acquisition, and how does AMD plan to use World Labs’ expertise?

Fei-Fei Li will become Executive Vice President and Chief Scientist at AMD. AMD says the acquisition will give it direct access to World Labs’ world-model research to inform future chip designs and software stacks for reasoning, robotics, simulation, and physical AI, helping it compete with Nvidia’s lead.

Source: Ars Technica