Nvidia Launches Cosmos WFMs to Advance AI Simulations

At CES 2025, Nvidia announced the release of its Cosmos World Foundation Models (WFMs), a new family of world models designed to predict and generate physics-aware videos. These models aim to revolutionize industries like robotics, autonomous vehicles, and physical AI research by creating controllable, high-quality synthetic data.

Key Features of Cosmos WFMs

Nvidia’s Cosmos WFMs are divided into three categories based on performance needs:

• Nano: Optimized for low-latency, real-time applications.

• Super: Provides robust performance for a variety of use cases.

• Ultra: Delivers maximum quality and fidelity, supporting high-end applications.

The models range in size from 4 billion to 14 billion parameters, where larger models generally exhibit superior performance and problem solving skills. Additional features include:

• Upsampling Model: A video decoder optimized for augmented reality.

• Guardrail Models: Designed to promote ethical AI use.

• Customizable Fine-Tuned Models: Tailored for applications like generating sensor data for autonomous vehicles.

These models were trained on an extensive dataset comprising 9,000 trillion tokens collected from 20 million hours of human interactions and data across real-world environments, robotics, and driving scenarios.

Applications and Industry Impact

Cosmos WFMs are particularly well-suited for:

• Robotics and Physical AI: Enabling realistic simulations to train robots in industrial and warehouse settings.

• Autonomous Vehicles: Generating synthetic driving data for enhanced training of self-driving systems.

• Entertainment and Media: Creating physics-based, high-quality videos for interactive experiences.

Several companies, including Uber, Waabi, and Fortellix, are already piloting Cosmos WFMs for autonomous driving and AI model development. Nvidia CEO Jensen Huang emphasized that Cosmos is purpose-built for industries requiring both rich data and powerful compute capabilities.

“Generative AI will power the future of mobility, requiring both rich data and very powerful compute,” Uber CEO Dara Khosrowshahi said in a statement. “By working with Nvidia, we are confident that we can help supercharge the timeline for safe and scalable autonomous driving solutions for the industry.”

“We really hope [Cosmos will] do for the world of robotics and industrial AI what Llama … has done for enterprise,” Nvidia CEO Jensen Huang said

Licensing and Controversies

While Nvidia touts Cosmos WFMs as “open models,” the licensing doesn’t fully align with the strictest definitions of open source. Nvidia has not disclosed complete details about the training data’s provenance or provided tools to recreate the models. Instead, the models are distributed under a permissive license that allows commercial use.

“Cosmos learns just like people learn,” the spokesperson said. “To help Cosmos learn, we gathered data from a variety of public and private sources and are confident our use of data is consistent with both the letter and spirit of the law. Facts about how the world works — which are what the Cosmos models learn — are not copyrightable or subject to the control of any individual author or company.”

Nvidia has faced scrutiny for its data collection practices, with critics raising concerns about the use of copyrighted materials like YouTube videos for training. The company maintains that its data use adheres to fair use laws, a legal area still being tested in courts.

What This Means

Nvidia’s Cosmos WFMs represent a significant step forward for AI applications that require realistic simulations. By enabling the generation of physics-based, synthetic data, these models can accelerate the development of advanced robotics, autonomous systems, and even augmented reality experiences.

Beyond their technical capabilities, Cosmos WFMs democratize access to world models by offering them under a permissive license, allowing developers of all sizes to innovate. This open approach could catalyze breakthroughs across industries, fostering collaboration and driving AI adoption at scale.

To start building using Cosmos, and to learn more, visit Nvidia’s website.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.