AMD to acquire World Labs in all-stock deal worth around $8.2 billion
The deal is an all-stock transaction, expected to close by the end of 2026, and Fei-Fei Li becomes AMD's new chief scientist under Lisa Su. The acquisition positions AMD against Nvidia's world-model ecosystem, but the terms beyond the headline figure remain unknown.
AMD announced on September 28, 2026 that it has entered into a definitive agreement to acquire World Labs, the AI lab led by Fei-Fei Li, in an all-stock transaction valued at approximately $8.2 billion. The deal is expected to close by the end of 2026, subject to regulatory approvals and other customary closing conditions (AMD press release).
Once the transaction is completed, Li becomes executive vice president and chief scientist at AMD, with a reporting line directly to CEO Lisa Su. Justin Johnson and Ben Mildenhall are, together with Li, set to continue leading the World Labs team after the acquisition (Unite.AI).
What World Labs actually builds
World Labs, headquartered in San Francisco, develops spatial intelligence models that generate, reconstruct and simulate interactive 3D environments from text, image and video inputs, in addition to technology for robotic learning and simulation (AMD press release). This is the technology AMD is now buying.
According to Crypto Briefing, the company has two central products: the model Marble, launched in November 2025, and Atlas, launched in September 2026, both aimed at gaming, robotics and virtual reality (Crypto Briefing). The same source reports that World Labs acquired the robotics simulation startup SceniX in July 2026, but this currently rests on a single reporting source.
World Labs itself describes Atlas as an "omni" world model, pre-trained from scratch as a multimodal autoregressive diffusion transformer operating directly on text, images, video and 3D, with camera-controlled video generation of up to one minute at 1440p, now entering early access with selected partners (Unite.AI). The model's performance claims, however, are largely the company's own evaluations, not independently verified.
The strategic logic
Su tied the acquisition directly to AMD's roadmap: "Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving," she said in the press release (AMD press release). The point is that hardware decisions are increasingly shaped by what kinds of models are being trained and run — and that by owning a model lab, AMD gains that insight internally rather than externally.
TechCrunch points to the more competitive reading: Nvidia already has a suite of open world models such as Cosmos, while AMD has so far only offered text- and video-based models publicly. With World Labs, AMD gets an entry into the world-model ecosystem, not just the chips (TechCrunch).
Li herself framed it in her announcement. According to TechCrunch, she wrote in her post that the company now wants to scale its efforts, broaden its reach and get "closer to the hardware" (TechCrunch). The core, in other words, is getting closer to the hardware — and it is hard to get closer to hardware than at a chipmaker.
This is not a relationship without history. World Labs itself states that the company began a deep technical partnership with AMD as early as 2025, starting from model training and inference optimization on AMD's GPUs, and that along the way the teams saw a natural fit between software, hardware, foundation models and applications (Unite.AI). Crypto Briefing additionally reports that AMD participated in World Labs' $1 billion round in February 2026, alongside Nvidia and Autodesk — giving AMD a close look at the technology before choosing to buy the whole company. The same source cites a seed round of $230 million in September 2024 and total funding of around $1.23 billion before the acquisition. These figures, however, are not verified through other sources in this evidence base, and should be read with that caveat (Crypto Briefing).
The market reaction
The stock market reaction on announcement day was negative: According to Blockonomi, AMD's share closed at $607.87, down 3.61 percent, after a sharp mid-morning wave of selling that, according to the same source, wiped out earlier gains (Blockonomi).
The structure itself — an all-stock deal — means AMD limits its cash use while potentially diluting existing shareholders. That is an analysis, not a fact stated by AMD, and the exact dilution effects depend on details the company has not yet presented.
What remains open
Several central details are still missing. The press release confirms only that the transaction is approximately $8.2 billion, in shares, and that the terms beyond that have not been disclosed. Regulatory approval remains outstanding, and although the company points to an expected closing by the end of 2026, there is no guarantee in that regard (AMD press release).
There is also some uncertainty in the public picture of the company itself: different sources are not fully in agreement on the number of founders behind World Labs — only Li, Johnson and Mildenhall are named in all sources. The funding figures and the SceniX acquisition mentioned above rest largely on a single reporting source, and the Atlas model's performance claims are, as described, largely the company's own evaluations, not independently verified.
Why it matters
Beyond the headline figure itself, this acquisition points to a broader development: it is no longer just about building fast chips, but about owning the whole stack — hardware, models and the surrounding application ecosystem. As chipmakers begin buying model labs, the lines between "chip company" and "AI company" grow steadily sharper.