Nvidia puts artificial intelligence behind the wheel: CEO Jensen Huang also wants automotive

.

Jensen Huang always enters the stage as if he were about to take the stage at a concert. Black leather jacket, Silicon Valley rock star smile, ready joke. At Ces in early January, Las Vegas, he greets the audience like Elvis in Memphis and then tells you that 'the ChatGPT moment of physical artificial intelligence' is coming. Translation: not just chatbots, but machines that understand the real world. Not bad for a guy who built an empire selling graphics cards to gamers.

The protagonist is Jensen Huang, founder and CEO of Nvidia, who now also wants to take the wheel in the automotive industry. After having colonised data centres, generative AI and desktop supercomputers, Huang is now aiming at 'Physical AI': artificial intelligences that do not just generate text or images, but must drive, move, avoid obstacles and not run over pedestrians.

The key project is called Alpamayo. An Andean mountain name, technological Everest ambition. It is an open source family of models, simulators and datasets designed to train autonomous vehicles and robots. At its heart is Alpamayo 1, a VLA (vision, language, action) model with 10 billion parameters: it sees the road, interprets what is happening, thinks in steps and then decides what to do. It does not just react: it plans. It is the famous 'chain of thought', but applied to traffic. Here, Nvidia's strategy is clear: to become the nervous system of the automotive industry. Not just selling chips, but the entire cognitive infrastructure of the cars of the future. Huang has been repeating this for years with a certain Zen-like obsession: 'Simulation is what Nvidia does'. And in fact Alpamayo was not created to run directly on the car, but as a 'learning brain' from which manufacturers can derive lighter versions integrated into their software stacks.

The real conceptual leap, however, is another, and has an even more ambitious name: World Foundation Models. If LLMs learn correlations between words, these models learn the laws of the world. Space, gravity, collisions, cause and effect. They don't answer questions, but predict what happens if you act. It's an AI that thinks in three dimensions. Nvidia calls it Cosmo: a model that was not created to chat, but to simulate the real within photorealistic virtual environments built in Omniverse. Millions of fake accidents to avoid a real one.