AMD to Acquire World Labs for $8.2 Billion, with Fei-Fei Li Becoming Chief Scientist

On September 28, 2026, AMD announced it will acquire World Labs, a company known for research in spatial intelligence: AI that understands and generates 3D space. The deal is valued at about 8.2 billion USD (about 1.3 trillion yen) and will be paid entirely in AMD stock. World Labs leader Fei-Fei Li will join AMD as Executive Vice President and Chief Scientist, reporting directly to CEO Lisa Su. It is an unusual move for a chipmaker to bring an AI model team fully in-house.

※1 USD = 158 JPY (as of October 2, 2026)

All-stock deal, expected to close by year-end

According to the announcement, the acquisition is an all-stock transaction expected to close by the end of 2026, subject to regulatory approvals. In size, it is described as AMD's second-largest acquisition after the roughly 50 billion USD (about 7.9 trillion yen) purchase of Xilinx in 2022.

Past deals put the scale in perspective: ATI cost 5.4 billion USD (about 850 billion yen), ZT Systems 4.9 billion USD (about 770 billion yen), and Pensando 1.9 billion USD (about 300 billion yen).

AMD had also invested in World Labs' 1 billion USD (about 158 billion yen) funding round in February. The two companies have worked together on optimizing model training and inference, and World Labs' models have already run on AMD Instinct accelerators with the ROCm software stack. The acquisition deepens that existing relationship.

What World Labs does

Based in San Francisco, World Labs builds models that generate, reconstruct, and simulate interactive 3D environments. In late 2025 it released Marble, a commercial service that creates persistent 3D worlds. Its latest world model, Atlas, is positioned as the foundation for future versions of Marble.

Co-founders Justin Johnson and Ben Mildenhall will continue to lead the research. With the team kept intact, the direction of development is unlikely to change dramatically after the deal.

Why a chipmaker wants model researchers

Su said that building compute platforms for next-generation AI requires a deep understanding of how models are evolving. Li said AI becomes hobbled in efficiency unless model research, systems, and compute work more closely together.

World models and robotics simulation tend to stress memory capacity and bandwidth more than raw compute, which some see as a good fit for AMD's strength in high-capacity HBM memory. Analysts also point to a recent result in which AMD's MI355X outperformed NVIDIA's Blackwell Ultra in a video-generation comparison as having supported the decision.

It is also notable that Li will report directly to the CEO. Rather than placing her in one corner of the research organization, AMD appears to be signaling that world models sit at the core of its next-generation chip strategy.

The challenge: the market is still young

Challenges remain. AMD's lead in video generation may last only one benchmark cycle, and NVIDIA could regain the advantage as it optimizes. World models are also still largely at the research stage, with no standard evaluation framework like MLPerf. Whether video generation's cost structure carries over to world models has yet to be confirmed by real customer purchases.

While NVIDIA distributes its own models to stimulate GPU demand, AMD has chosen to bring model research in-house and design chips alongside it. Things to watch include when World Labs' models arrive in ROCm libraries and what memory specifications the next-generation MI400 will have.

Summary

AMD's acquisition of World Labs is an all-stock deal worth about 8.2 billion USD, targeted to close within the year. Li's appointment as Chief Scientist shows AMD placing world models at the center of its semiconductor strategy. Success will depend on whether the research translates into actual product design and customer adoption.