Nvidia unveils AI models for training tech

AI to better train robots and cars as well as new gaming chips dominated Nvidia CEO Jensen Huang's keynote speech at the CES 2025 conference as the world's second most valuable firm expounded upon its potential to expand its business.

Nvidia introduced what it calls Cosmos "foundation" models that generate photo-realistic video which can be used to train robots and self-driving cars at a much lower cost than using conventional data.

By creating what is known in the tech industry as "synthetic" training data, the models can help robots and cars understand the physical world similar to the way that large language models helped chatbots generate responses in natural language.

Users will be able to give Cosmos a text description that can be used to generate video of a world that obeys the laws of physics.

This promises to be much cheaper than gathering data as it is done today.

To train self-driving cars, for example, companies have fleets of vehicles that roam streets to gather data, and humanoid robots are often trained by having real humans repeat tasks over and over.

Huang, however, cautioned that the Cosmos models will need much more data before hitting their "ChatGPT moment".

"We really hope (Cosmos) will do for the world of robotics and industrial AI what Lama3 has done for enterprise AI," Huang said.

The new chips also have AI technology to help game developers generate more accurate human faces, an area where players are apt to notice even slightly unrealistic features.

The chips, which Nvidia calls its RTX 50 series, will range in price from US$549 to US$1,999, with top models arriving on January 30 and lower-tier models coming in February.

Nvidia said its mid-grade US$549 gaming chips will match the company's previous flagship chip, the RTX 4090 which sold for US$1,600.