We are witnessing a paradigm shift that will make 2026 the year zero for a new species of machine. If until yesterday we marvelled at how software could write poetry or generate surreal images, what emerged from the CES in Las Vegas shifts the bar towards a far more tangible and complex frontier: Physical AI. We are no longer talking about bits that reorganise digital information, but artificial intelligence that understands, navigates and manipulates the physical world with an understanding that begins to blur the boundaries with biological perception.
Central to this revolution is the announcement of the new Vera Rubin architecture. Nvidia has decided to pay homage to the astronomer who confirmed the existence of dark matter with a system that promises to illuminate the dark sides of computational computing. The Rubin platform is not a simple incremental upgrade from Blackwell; it is a quantum leap made possible by what Jensen Huang called‘extreme code design‘. We are talking about a system where the Vera CPU and Rubin GPU work in such close symbiosis that they seem like a single organism, offering ten times the performance and memory management that redefines supercomputing. It is the engine needed to power not just chatbots, but entire digital factories and swarms of autonomous robots.
The most fascinating aspect, however, lies not in the silicon but in the software that animates it. The introduction of Cosmos, the new foundational model for the physical world, is the brain of this operation. During the presentation, which you can learn more about in the video we include in the article, Nvidia demonstrated an unprecedented type of visual analysis: not a simple detection of obstacles, but real reasoning in real time. The system for autonomous driving, called Alpa Mayo, does not just react to stimuli; it ‘thinks’ and verbalises its intentions before acting. Watching the demo, one notices how the AI breaks down complex scenarios into logical segments, assessing risks and trajectories with a cognitive transparency that self-driving cars’ black boxes have so far totally lacked. It is the difference between an unconditional reflex and a considered decision.
This intelligence does not remain confined to cars. The strategic partnership with giants like Siemens shows that the next step is theindustrial metaverse. Factories of the future will be designed, built and tested entirely within physically accurate simulations before a single real brick is laid. The humanoid robots, trained through the GR00T project in these synthetic environments, will arrive in our world already experienced, having lived millions of simulated lives in a few hours of computing.
We are facing the end of the experimental phase of generative AI and the beginning of its industrial and physical phase. No longer will we interact with AI only through a keyboard or voice command; we will meet it on the street, work with it side by side, and rely on its reasoning ability for our physical safety. If 2025 was the year AI learned to speak, 2026 will be remembered as the year it learned to move. And judging by the computing power deployed by Nvidia, it plans to run very fast.
