CES 2025: Nvidia CEO Jensen Huang says company’s new AI models were trained on 20 million human hours

Nvidia is taking artificial intelligence (AI) to the next level with the launch of Cosmos, its latest family of AI core models. Announced During CEO Jensen Huang’s keynote at CES 2025 in Las Vegas, he says Cosmos is specifically designed to teach robots and machines to interact with the physical world. Cosmos was trained on 20 million hours of real footage of “people walking, moving their hands, manipulating things,” Jensen said. From industrial robots to humanoids to self-driving cars, the applications of this technology are vast and exciting.

What sets Cosmos apart?

While language models like ChatGPT or Bard learn to generate text by digesting mountains of written content, Cosmos focuses on the visual and physical realms. Huang explained that the artificial intelligence models were trained on 20 million hours of real footage of people performing everyday actions — walking, moving their hands, manipulating objects. “It’s not about creating art,” Huang said. “It’s about teaching AI to understand the physical world.”

During the presentation, Huang apparently demonstrated Cosmos in action, simulating a warehouse environment where boxes fall off the shelves. This type of realistic video creation is critical for training robots to recognize and respond to real-world situations, such as workplace accidents. Companies can also fine-tune Cosmos using their own data, tailoring it to their specific needs.

The influence of Cosmos is already felt in all industries. Humanoid robot startups like Agility Robotics and Figure AI, as well as self-driving car companies like Uber, Waabi, and Wayve, are using the technology. These companies hope to push the boundaries of what robots can achieve, whether it’s navigating complex environments or completing complex tasks.

Isaac to complete the Cosmos

To complement Cosmos, Nvidia also unveiled an update to its Isaac robot simulation platform. The new feature allows robots to learn new tasks with unprecedented efficiency. For example, builders can use several examples of a desired action, such as grabbing an object, and Isaac will generate a series of synthetic training data to refine the skill.

Huang’s keynote at CES showed digital renderings of 14 life-size humanoid robots from industry leaders such as Tesla, Boston Dynamics and Agility Robotics, highlighting Nvidia’s commitment to shaping the future of robotics. “With Cosmos and Isaac, we’re giving developers the tools to build robots that can truly understand and operate in our world,” Huang said.

Nvidia’s bold vision underscores its leadership in AI, combining innovation with practicality. The integration of artificial intelligence into robotics could redefine industries, from logistics and manufacturing to autonomous driving. With Cosmos, Nvidia is paving the way for machines to become more human—not in thought, but in movement.