Nvidia has unveiled multiple new AI initiatives at CES 2025, centered around their Nemotron model families and Cosmos World Foundation Models. The company’s CEO Jensen Huang presented these developments during his opening keynote, introducing AI models designed to advance both enterprise and consumer applications. The Nemotron family includes language and vision models available as NIM microservices, which can be deployed across various computing platforms.
The newly announced Cosmos World Foundation Models (WFMs) focus on generating physics-aware videos and simulations. These models, ranging from 4 billion to 14 billion parameters, come in three sizes: Nano for real-time applications, Super for baseline performance, and Ultra for maximum quality. According to Nvidia, the models were trained on 9,000 trillion tokens from 20 million hours of various real-world data, though the company has not disclosed specific data sources.
Additionally, Nvidia introduced AI foundation models designed to run locally on RTX AI PCs, aiming to enhance digital human interaction, content creation, and development capabilities. The company announced that major PC manufacturers will launch NIM-ready RTX AI PCs, featuring these new technologies. The system will support various AI development frameworks and will be compatible with Windows 11 PCs through Windows Subsystem for Linux, with availability starting in February 2025.
Sources: VentureBeat, TechCrunch, VentureBeat