Why Physical AI Demands Massive On-Device Compute
NVIDIA Cosmos VP Ming-Yu Liu explains how physical AI differs from digital models and why humanoids need powerful on-device compute.
Open models and the future of Physical AI with NVIDIA
Ming-Yu Liu, Vice President of Cosmos Lab at NVIDIA, joins Daniel Whitenack and Chris Benson to discuss the critical role of open foundation models and world models in advancing physical AI. He explains the differences between digital and physical AI, the architecture of NVIDIA Cosmos models for physical simulation and understanding, and how multi-agent robotics ecosystems will transform real-world automation.
Key takeaways
- Physical AI refers strictly to models deployed on physical devices that perturb the physical state of the environment to produce tangible, material results. Read more →
- Cloud language models hide hardware inefficiencies by batching hundreds of concurrent user requests; physical robots cannot aggregate requests and must run batch-size-one inference. Read more →
- Ming-Yu Liu leads NVIDIA Cosmos Lab to build foundation models that simulate physical physics rather than parse web text. Read more →
- Ming-Yu Liu, Vice President of Cosmos Lab at NVIDIA, defines an open model by three components: open weights hosted on Hugging Face, training frameworks on GitHub, and shared datasets. Read more →
- Closed APIs block internal inspection: Ming-Yu Liu points out that hosted endpoints prevent engineers from modifying model architectures or running local, zero-latency inference. Read more →
5 articles from this episode
Why Robotics AI Breaks on Batch Size One
NVIDIA Cosmos VP Ming-Yu Liu explains why physical robotics breaks cloud LLM inference economics and demands new edge architectures.
NVIDIA Cosmos: Why Physical AI Needs World Models, Not LLMs
Ming-Yu Liu explains how NVIDIA Cosmos merges simulation, video, and action data to train robots without real-world crashes.
NVIDIA's Open Model Play: Beyond Model Weights
Ming-Yu Liu explains why NVIDIA pairs open weights on Hugging Face with GitHub training code, cookbooks, and custom datasets.
Why NVIDIA Bets on Open Models Over Closed APIs
NVIDIA's Ming-Yu Liu explains why open weights beat closed APIs for custom physical AI workloads and hardware design.
More Practical AI episodes
- From AGENTS.md to Enterprise DeploymentSep 27 · 5 articles
- Computer-Use Agents and the Future of the Agentic InternetSep 13 · 4 articles