NVIDIA is taking over the physical world: 19 game-changing announcements

At the latest GTC Taipei conference, Jensen Huang didn't come to talk about another 10% boost in gaming performance. He came to present the full infrastructure for what he calls the "Age of Agents." In a keynote featuring 19 different announcements, NVIDIA made it clear that it is no longer just a chip company – it is becoming the operating system of the physical world.

This shift, from software that creates text and images (generative AI) to systems that operate in the real world (Physical AI), is the most significant change we have seen in recent years. If, until now, AI was trapped inside a chatbot, today it is gaining a body, the ability to move, and responsibility for entire production lines.

Cosmos 3: The model that makes machines understand reality

One of the most fascinating announcements is Cosmos 3 – an open model designed to be the "brain" of robots and autonomous vehicles. Unlike standard language models, Cosmos 3 is an omnimodal model that understands motion, causality, and the laws of physics.

It comes in two versions: the Nano version (8 billion parameters) designed for local execution on workstations, and the Super version (32 billion parameters) for heavy cloud tasks. The goal here is to shorten robot training processes from months to a few days by accurately simulating the world before the robot even touches a physical object. You can find the models right now on Hugging Face.

Vera Rubin: The engine of AI factories

To run all these agents, you need a new kind of computing power. NVIDIA announced the Vera Rubin platform, which is entering full production. The highlight here is the Vera CPU, which is based on "Olympus" cores designed custom by NVIDIA.

Why does this matter? Because AI agents require a lot of "management" work – running code, calling external tools, and making logical decisions – tasks that traditional x86 processors are less efficient at. The Vera CPU delivers 1.8x better performance in AI agent and reinforcement learning tasks, making it the beating heart of the new "AI factories."

AI lands on the desktop: RTX Spark and DGX Station

NVIDIA isn't leaving the power only to the big data centers. It wants agents to run in your office and on your laptop.

• •RTX Spark: A new "super-chip" for Windows PCs that combines a Grace CPU and a Blackwell GPU. It brings 1 petaflop performance to a laptop, allowing you to run heavy AI agents locally and completely privately.
• •DGX Station for Windows: A powerful workstation (based on GB300) capable of running models of up to a trillion parameters (!) under the developer's desk.

The industrial revolution in Taiwan

The collaboration with TSMC and Foxconn is leveling up. NVIDIA introduced FOX – a blueprint for building an autonomous factory manager. It is an AI agent that connects to all machines, sensors, and operational data, making decisions in real-time to prevent malfunctions.

The medical world is not being left behind either: together with medical centers in Taiwan, NVIDIA is building AI agents that will help nurses and doctors manage clinical processes and diagnostics, which indicates that the technology is permeating every aspect of our lives.

Summary: Where are we headed?

NVIDIA's message is clear: AI is ceasing to be just "software that talks" and is becoming an infrastructure that runs the world. We will see more robots (including an open humanoid design for research that the company released), robotaxi vehicles that understand their environment better (thanks to the Alpamayo 2 model), and security systems like the BlueField-4 STX that will keep these agents from straying from their defined roles.

For developers and organizations, this is the time to stop thinking of AI as a tool for summarizing emails, and to start thinking of it as an employee – an autonomous agent capable of performing complex tasks in the real world.

For more information on the announcements, follow the official NVIDIA blog.