Cloud-led Future · Intelligence-driven Hotline +86 21 6228 0217 中文 | EN

Chip Makers Buy Into Physical AI as AMD and Nvidia Close Major Deals

Overview

Two leading chip makers acquired model and platform assets in the same week. AMD completed its roughly $8.2 billion all-stock acquisition of spatial intelligence company World Labs, with Fei-Fei Li becoming executive vice president and chief scientist, reporting directly to Lisa Su. The team will continue developing spatial intelligence and world models. At CoRL 2026, Su argued that physical AI — robotics in particular — could reach a multi-trillion-dollar scale within years.

Nvidia announced the acquisition of open-source model platform Hugging Face alongside its fiscal 2027 third-quarter results. Revenue reached $115.8 billion, up 103% year on year, with data center revenue of $107.6 billion, up 110%, accounting for about 90% of the total. Vera Rubin NVL72 entered production, described as the fastest product ramp in company history. Fourth-quarter guidance of about $130 billion excludes China data center revenue. The board also authorized an additional $150 billion in buybacks.

Together, the deals signal that chip vendors are no longer content to sell compute: they are buying model capability and developer ecosystems outright.

Background & Interpretation

1. Technological Background and Evolution

Physical AI requires systems that understand three-dimensional space, predict motion and generate executable robot actions. Training demands high-bandwidth interconnects and large-scale simulation; inference pushes compute toward edge nodes. World models are therefore the connective tissue between data, simulation and control. Bringing them in-house lets chip companies create an internal loop between hardware workloads and software demand instead of waiting for third parties to fill scenario gaps.

2. Core Drivers and Underlying Mechanisms

Competition is moving from process nodes and interconnects to system-level contests spanning chips, software stacks, models and scenario loops. AMD’s route combines open hardware, open software and proprietary frontier models, using real spatial-intelligence workloads to feed back into Instinct and ROCm. Nvidia’s moat spans GPUs, CUDA, NVLink, the Cosmos world model and the GR00T robotics model. Acquiring an open-source platform makes the developer gateway itself a strategic asset.

3. Market Landscape and Value-Chain Impact

System-level competition carries tension. If in-house models are optimized first for their own chips, cloud providers and model companies may worry about being both referee and player. ROCm maturity, model openness and cloud instance availability will shape migration willingness. All-stock consideration also brings dilution that needs earnings to absorb. Nvidia’s ecosystem inertia and installed base remain hard to dislodge near term, while fragmented physical-AI scenarios leave room for open hardware and software.

Implications & Outlook

Closing the deals is only the beginning; the real test is roadmap coordination and developer adoption. Over the medium term, the global AI chip landscape is more likely to become multipolar, with cloud ASICs, open models and vertical models developing in parallel. Watch neutrality arrangements for acquired open platforms, the speed at which models adapt to new hardware, and actual compute procurement in robotics and industrial simulation.

Takeaways for Industry Participants

Chip companies should draw clear boundaries between open ecosystems and proprietary models, winning developers with verifiable performance and toolchains. Model and robotics firms should preserve multi-platform deployment to avoid lock-in. Cloud providers can exploit scenario fragmentation by differentiating in inference and edge supply.

This column compiles industry information and shares technical perspectives; it does not constitute investment advice.