NVIDIA has released a suite of open-source AI technologies across language, robotics, and healthcare. These tools, part of the Nemotron, Cosmos, Isaac GR00T, and Clara families, aim to enhance AI accessibility and foster innovation. They are being contributed to Hugging Face, allowing developers to leverage cutting-edge resources for specialized applications.

This article discusses the rapid evolution of AI infrastructure, focusing on the demand for advanced memory solutions like 16-Hi HBM and the implications for programming and robotics. It highlights how the increasing capabilities of AI models are outpacing current hardware, leading to a potential shift in how we leverage AI in various fields.

NVIDIA introduces Cosmos Policy, a new robot control system that enhances manipulation tasks by post-training the Cosmos Predict model. It combines robot actions, states, and success metrics into a unified framework, achieving top performance on benchmarks like LIBERO and RoboCasa. The article also announces an open hackathon for developers to experiment with these models.

Nvidia has launched its Jetson AGX Thor robotics chip module, priced at $3,499 for developers, aimed at enabling companies to create advanced robots. The chips, which are 7.5 times faster than previous models and equipped with 128GB of memory, are part of Nvidia's strategy to capitalize on the growing robotics market, although it currently represents only 1% of the company's revenue. Major companies like Amazon and Boston Dynamics are already utilizing these chips for their robotic applications.