The next decisive AI battle may not be fought on screens at all. It may unfold in factories, warehouses, hospitals, and streets, where machines must perceive, predict, and move rather than merely converse. NVIDIA has spent 2026 pushing exactly that vision of “physical AI.” At GTC in March, it unveiled new Cosmos world models, Isaac simulation frameworks, and Isaac GR00T N robot models, while highlighting a broad partner ecosystem that includes ABB Robotics, FANUC, Figure, KUKA, Skild AI, and others. On the same day, it introduced a Physical AI Data Factory Blueprint meant to automate the curation, augmentation, and evaluation of training data at scale—an especially significant step in robotics, where real-world data is costly, slow, and often hazardous to collect. (nvidianews.nvidia.com)
Alibaba, meanwhile, is moving rapidly from multimodal AI into embodied intelligence. On May 29, the Qwen team announced Qwen-VLA, a general-purpose vision-language-action model designed to turn visual understanding and language reasoning into continuous action and trajectory generation. Its paper reports striking benchmark results, including 97.9% on LIBERO, 73.7% on Simpler-WidowX, and strong out-of-distribution performance on real-world ALOHA robot tests. Then, on June 16, Reuters reported that Alibaba unveiled its first suite of AI models for robots. That push is now reflected in technical reports for Qwen-RobotManip, Qwen-RobotNav, and Qwen-RobotWorld, which target manipulation, navigation, and world modeling respectively. In other words, Alibaba is not presenting a single showpiece robot; it is assembling its own embodied-AI stack. (qwen.ai)
What makes this rivalry so consequential is that physical AI changes the scale of impact. A chatbot may save time; a capable robot can reshape labor, logistics, and industrial productivity. NVIDIA’s strength lies in its full-stack strategy: chips, simulation, world models, and a vast robotics ecosystem. Alibaba’s strength is different but formidable: a fast-evolving Qwen model family, cloud reach, and an increasingly coherent robotics research pipeline. The real contest, then, is not simply over who builds the smartest model, but over who can fuse data, simulation, hardware, and deployment into a workable industrial system first. After the era of talking machines, the era of acting machines has unmistakably begun. (nvidianews.nvidia.com)










