Investors warming to embodied AI

Alibaba Cloud has joined a 2 billion yuan, or about $290 million, funding round for ShengShu, a Chinese start-up building artificial intelligence models meant to simulate the physical world. (cnbc.com) ShengShu announced the Series B round on April 10 and said the money will go toward a “general world model.” Reuters reported the round was led by Alibaba Cloud and valued at about $292.6 million. (reuters.com) The company is best known for Vidu, an artificial intelligence video generator, but it said the new model is supposed to connect digital scenes with real-world uses such as autonomous driving and robots. Bloomberg reported the round also drew backing from Baidu Ventures and Luminous Ventures. (bloomberg.com) A world model is software that tries to predict what happens next in a scene, more like a flight simulator than a chatbot. Nvidia says its own “world foundation models” are designed to generate predictive video worlds for robots and other “physical artificial intelligence” systems. (nvidia.com) That pitch has gained traction as companies run into the limits of large language models trained mostly on text. CNBC reported ShengShu said multimodal data such as vision, audio and touch can capture how the physical world works more naturally than text-heavy models alone. (cnbc.com) The idea is spreading beyond one company. Google DeepMind said in August 2025 that Genie 3 could generate interactive environments in real time at 24 frames per second, while Nvidia has been building tools to generate robot-centric simulations and synthetic training data. (deepmind.google) (developer.nvidia.com) Investors are also backing the hardware side of the same bet. In September 2025, Alibaba Cloud led a nearly 1 billion yuan funding round for X Square Robot, which said the deal was Alibaba Cloud’s first investment in embodied intelligence. (scmp.com) Not everyone agrees that better simulation will quickly turn into capable robots. Google said Project Genie still has limits in realism and character control, a reminder that generating convincing environments is easier than building systems that can reliably act inside them. (blog.google) For now, the money is moving toward models that can see, predict and rehearse action, not just answer prompts. ShengShu’s new round shows investors want the next wave of artificial intelligence to learn physics as well as prose. (reuters.com)