Alibaba Cloud led a $290 million Series B funding round for ShengShu, the developer of the AI video generation tool Vidu, CNBC reported on Thursday.

TAL Education and Baidu Ventures also participated in the investment.

The deal is one of a series of examples showing a shift toward betting on the development of “world models” based on video and physical scenarios, reflecting recognition of the limits of text-focused large language models (LLMs).

ShengShu will use the funds to develop a “general-purpose world model” that connects digital worlds based on gaming and AI video with physical worlds based on autonomous driving and robots.

ShengShu said a general-purpose world model built on multimodal data such as vision, audio and touch captures how the physical world works more naturally than LLMs. ShengShu founder Zhu Jun (주준) said the goal is to connect perception and action so AI systems can consistently model and predict real-world behavior.

According to CNBC, ShengShu’s latest model, Vidu Q3 Pro, ranks in the top 10 in the field of AI models that generate video from text and images. ShengShu launched Vidu globally months before OpenAI unveiled Sora.

Alibaba is expanding investments in related startups. Last month, it invested $50 million with Baidu Ventures in Tripo AI, which generates 3D digital models from photos. In September last year, it invested $60 million in PixVerse, which launched an AI world model. Alibaba has also released open-source AI video generation models and unveiled a model for powering robots in February this year. World models are seen as particularly important in robotics, analysis shows. Kevin Kelly (케빈 켈리), a co-founder of U.S. IT magazine Wired, said building human-level AI requires three things: reasoning, understanding the physical world and continual learning. He said world models would be the key breakthrough.