NVIDIA pivots toward physical AI

Robotics is becoming Nvidia’s next big land grab. Not instead of data centers — on top of them. The shift is toward what Nvidia calls “physical AI,” meaning models that do not just generate text or images but perceive space, predict motion, and control machines in the real world. That story sharpened in early 2026, and by May 3 the market was treating it as more than branding, with Asian partner stocks rallying on fresh tie-ups and supply-chain roles. (thehindubusinessline.com) ### What does “physical AI” actually mean? Basically, it is AI for things that move and touch stuff. Nvidia’s stack here spans Cosmos world models, Isaac simulation tools, GR00T robot models, Omniverse digital twins, and Jetson edge modules that run inference inside machines rather than in a distant c(thehindubusinessline.com)warehouses, hospitals, and infrastructure. (blogs.nvidia.com) ### What changed this year? January and March were the real pivot points. At CES 2026, Nvidia released new open models, frameworks, and infrastructure for robotics and autonomous systems. Then at GTC on March 16, Jensen Huang pushed the message harder, saying physical AI had arrived and naming a long list of partners — ABB, FANUC, KUKA, YASKAWA, (blogs.nvidia.com) map. (nvidianews.nvidia.com) ### Why are Asian suppliers rallying now? Because the market is connecting Nvidia’s robotics push to the companies that actually build the pieces. On May 3, reporting tied recent rallies in LG Electronics, Nanya Technology, Huizhou Desay SV Automotive, and Pateo Connect to collaborations, supply-chain partici(nvidianews.nvidia.com)tics. (thehindubusinessline.com) ### Why is the 90% number such a big deal? Because it shows how much of Nvidia’s real-world execution now runs through Asia. Bloomberg’s compiled data, cited May 3, put Asian suppliers at about 90% of Nvidia’s production costs, up from roughly 65% a year earlier. That is not just more chips from Asia. (thehindubusinessline.com)on. (thehindubusinessline.com) ### Is this still a chip story? Yes — but the catch is that the chip is becoming the entry ticket, not the whole product. Nvidia is trying to do in robotics what it did in AI training: own the full stack. The moat is not only GPUs. It is simulation tools, model libraries, orchestration software, edge modules, and a common workflow that lets customers move from synthetic data to deployment without changing platforms. (blogs.nvidia.com) ### Why does simulation matter so much? Because training robots in the real world is slow, expensive, and dangerous. A chatbot can fail in a browser tab. A warehouse robot can crash into a shelf. Nvidia’s answer is Omniverse and Isaac — build a digital twin, generate synthetic data, stress-test policies, then ship the model to the edge. Think of it like a flight simulator for factories and robots, except the simulator is also the software pipeline. (blogs.nvidia.com) ### So what does this mean for startups? Do not try to out-Nvidia Nvidia at the silicon layer. The better opening is higher up the stack — software that helps companies deploy models into factories, fleets, warehouses, and supply chains; tools for evaluation, safety, orchestration, and vertical workflows; and apps that turn general robot capabili(blogs.nvidia.com)ike chip challengers and more like industrial software companies with robots attached. (blogs.nvidia.com) ### Bottom line Nvidia is no longer just selling compute for AI that lives on screens. It is building the platform for AI that acts in the world — and that shift is already moving money, suppliers, and product roadmaps across Asia. (thehindubusinessline.com)