On September 28th, AMD officially announced that it had signed the final agreement to acquire the AI startup laboratory World Labs through an all-stock transaction worth 8.2 billion US dollars. This deal is AMD's second-largest acquisition in its history, following the $50 billion acquisition of Xilinx. It marks that the chip manufacturer is no longer solely focused on providing computing hardware but has incorporated model research capabilities into its technical framework, directly facing the new generation of computing load changes brought about by physical AI and spatial intelligence.
World Labs was founded by computer professor Li Feifei from Stanford University in San Francisco in 2024. Its core research direction is world models and physical AI. Its representative work is the Marble spatial intelligence model, which can generate persistent three-dimensional scenes with spatial consistency and interactivity based on a small amount of image and text input. Unlike current mainstream large language models, the technical focus of World Labs is not on text generation but on enabling AI to acquire knowledge of three-dimensional space, physical rules, and causal relationships, providing virtual training environments for scenarios such as robots, autonomous intelligent agents, industrial digital twins, and scientific simulations.
After the completion of the transaction, Li Feifei will serve as AMD's executive vice president and chief scientist, reporting directly to AMD CEO Su Zifeng. The other two co-founders of World Labs, Justin Johnson and Ben Mildenhall, will remain in their positions and maintain the original research team structure. The transaction is expected to be completed by the end of 2026 and still requires approval from multiple countries' anti-monopoly regulators. The World Labs will maintain independent operations before the transaction.
It is worth noting that AMD Ventures was already an early investor in World Labs. The two parties have previously carried out joint optimization at the GPU training and inference level, so this acquisition was not a temporary arrangement.
The core value of this acquisition is not simply "buying a set of model codes", but obtaining a leading model team and establishing a positive cycle of model-hardware collaborative design. In the past few years, AI chip competition has mostly revolved around the optimization of training and inference loads for large language models. The iterations of GPU architectures, memory bandwidth, and interconnection technologies have all served text-based large models. However, new loads such as physical AI and robot simulation have completely different computing characteristics - they require continuous rendering of three-dimensional scenes, simulation of physical dynamics, and high-frequency spatial reasoning, presenting new requirements for chip memory throughput, real-time rendering, and heterogeneous parallel computing power.
Su Zifeng stated in the official announcement that building the computing platform for the next generation of AI requires a deep understanding of the evolution of models. Li Feifei and the World Labs team have outstanding research leadership and model expertise. "We can leverage these insights to develop hardware, software, and system solutions that drive the next generation of AI and strengthen the open AI ecosystem."
Li Feifei believes that many current world model research is limited by computing infrastructure and hardware compatibility bottlenecks and remains at the laboratory stage. "Joining AMD will provide our team with resources and engineering strength, accelerating our research and helping us build the infrastructure needed for the next generation of AI." The World Labs team will retain research independence while deeply participating in AMD's overall AI strategy, promoting the collaboration of models, computing power, and system software.
From the competitive landscape, this acquisition will be AMD's differentiated layout in the AI computing power sector against major competitors, especially NVIDIA.
NVIDIA has continuously increased its investment in robotics, digital twins, and physical simulation ecosystems in recent years, leveraging GPU platforms and simulation software stacks to gain a first-mover advantage. In December last year and September this year, OpenAI spent 33 billion US dollars to acquire AI chip manufacturer Groq and open-source platform Hugging Face. Additionally, OpenAI and Meta also spent approximately 6.4 billion and 14 billion US dollars respectively in 2025 to acquire minority stakes in AI equipment startups io and Scale AI.
Previously, AMD's advantages lay in data center GPU hardware products, with relatively weak upper-level models and application ecosystems. Through the acquisition of World Labs, AMD entered the future high-potential field of physical AI, complemented its model capabilities in spatial intelligence and virtual simulation fields, improved the open AI ecosystem narrative, and expanded non-traditional large model markets such as robotics and industrial simulation.
However, the physical AI and world model fields are still in the early stage and there is a long period before they can achieve large-scale commercialization. In the short term, space models like Marble will not immediately bring significant revenue growth. This acquisition belongs to a strategic investment aimed at medium- and long-term computing power demand.
From the perspective of the entire semiconductor industry, AMD's acquisition of World Labs also reflects a new industry trend, that is, the AI computing power competition has shifted from simply competing in peak performance of chips to a full-stack competition of hardware, software stack, model capabilities, and industry scenarios. Future chip design will no longer be defined solely by hardware engineers, and the feedback from model researchers will become an important input for defining chip architectures. When AI moves from processing textual information to understanding the real physical world, the design logic of underlying computing power chips will also undergo a round of bottom-level reconfiguration.
Li Feifei immediately also posted an open letter on social media and began with a famous quote from "Ulysses" - "Come on, my friends, let's go and seek a newer world, for it is not too late now."
*Disclaimer: The article is from the internet. If there are disputes, please contact customer service.
