Fact Checked by 
Nvidia is gearing up to transform robotics with artificial intelligence, according to CEO Jensen Huang. Speaking on Monday (Jan. 6), he shared an ambitious vision for the company’s next phase of growth, calling it a “multitrillion-dollar” opportunity.
During his keynote at the annual Consumer Electronics Show (CES 2025) in Las Vegas, Huang unveiled a slew of new products and collaborations in what he called the “physical AI” space. Highlights included advanced AI models for humanoid robots and a major partnership with Toyota, using Nvidia’s self-driving car technology.
Huang described physical AI as the next big leap for the technology, comparing its potential to the game-changing impact that large language models (LLMs) have had on generative AI.
He said: “The ChatGPT moment for general robotics is just around the corner.”
Solving the challenges of deploying robots at scale, Huang predicted, could lead to the birth of “the largest technology industry the world has ever seen.” According to Huang, the market for humanoid robots is expected to reach $38 billion over the next two decades. It appears as if companies such as Tesla and Figure AI are already ahead in the race with their own functioning bots.
In November, ReadWrite reported that the tech entrepreneur believes that not only will there be significant developments in physical AI, but also AI Scaling Laws will boost computing power by a million times in 10 years.
What is Nvidia Cosmos?
On Monday, Nvidia introduced a suite of foundational AI models on its Cosmos platform. The tools are available for free and give developers the flexibility to generate data and create their own models.
We’re so unfathomably back – @NVIDIAAI releases Cosmos: World foundation models (commercially permissive) 🔥
Models trained on over 20 MILLION hours of video can be used to generate dynamic, high quality videos from text, image, or video inputs 🤯
Available directly on Hugging… pic.twitter.com/4hevOvlqzW
— Vaibhav (VB) Srivastav (@reach_vb) January 7, 2025
According to Huang, Cosmos models work by taking prompts like text, images, or videos, and transforming them into virtual world states as video outputs. “Cosmos generations prioritize the unique requirements of AV and robotics use cases like real-world environments, lighting, and object permanence,” Huang explained. Over 20 million hours of video is said to have been used to train the model.
He added: “We created Cosmos to democratize physical AI and put general robotics in reach of every developer.” Consequently, it is open-license and ready to access on GitHub.
Top players in robotics and automotive, such as 1X, Agile Robots, Agility, Figure AI, Foretellix, Fourier, Galbot, Hillbot, IntBot, Neura Robotics, Skild AI, Virtual Incision, Waabi, and XPENG, have already embraced Nvidia’s Cosmos platform. Ridesharing giant Uber is also among the early adopters.
Featured image: Nvidia
