At today's CES2026 keynote, NVIDIA CEO Jensen Huang made a powerful statement, announcing that the robotics industry has officially entered its own "ChatGPT moment." This declaration marks the acceleration of AI technology from the digital world into the physical world with perceptual capabilities.

To drive this progress, NVIDIA released a series of open-source "physical AI" models. Among them, Cosmos Transfer2.5 and Cosmos Predict2.5 are fully customizable "world models," giving machines the ability to understand real-world physical properties and spatial relationships. To address the challenge of safety evaluation, these models can generate realistic synthetic data, providing a safe virtual training ground for robots before they enter real environments.

image.png

In terms of reasoning and control, NVIDIA introduced the Cosmos Reason2 visual language model, enabling machines to perform logical reasoning using common sense, just like humans. The humanoid robot field also saw major updates, with the new Isaac GR00T N1.6 model achieving precise control over the entire body of a humanoid robot by integrating Cosmos' reasoning capabilities.

On the hardware side, NVIDIA launched the Jetson T4000 module featuring the Blackwell architecture, which delivers four times the performance of the previous generation, providing powerful computing power for running physical AI. Additionally, through a deep collaboration with Hugging Face, NVIDIA significantly simplified the workflow from simulation training to physical deployment, comprehensively empowering the developer ecosystem.

Key points:

  • 🤖 Robotics is about to take off: At CES2026, Jensen Huang announced that the robotics industry has entered its "ChatGPT moment" and released multiple open-source physical AI models.

  • 🧠 Deep perception of the world model: The Cosmos series models give machines the ability to understand physical common sense and spatial relationships, supporting safe performance evaluation in virtual environments.

  • 🚀 Fourfold performance improvement: The new Jetson T4000 module achieves a performance leap thanks to the Blackwell architecture, supporting real-time operation of complex physical AI and humanoid robots.