Alibaba Bets on ‘World Models’ as Next Frontier Beyond ChatGPT-Style AI

Alibaba is doubling down on the future of artificial intelligence with a major investment in a new class of models designed to go beyond text-based reasoning. Through its cloud division, the company is leading a 2 billion yuan ($290 million) funding round in ShengShu, a startup developing advanced AI systems that aim to replicate real-world environments more accurately.

Unlike traditional AI systems such as ChatGPT, which are primarily trained on text, ShengShu is building “world models” that rely on multimodal data — including video, audio and physical interactions. The goal is to create AI that not only understands language but can also interpret and predict real-world behavior, bridging the gap between digital environments and physical systems.

At the center of this effort is ShengShu’s Vidu platform, an AI video generation tool that has already ranked among the top models globally. The company’s broader ambition is to develop a “general world model” capable of connecting perception and action — a critical requirement for technologies such as autonomous driving, robotics and embodied AI systems.

The investment reflects a broader industry shift as developers confront the limitations of large language models (LLMs). While LLMs have driven rapid progress in conversational AI, they struggle to model physical reality, making them less effective for applications that require spatial awareness or interaction with the real world.

Alibaba’s move also intensifies competition in China’s rapidly evolving AI landscape. Companies like Kuaishou and ByteDance have launched their own AI video tools, while global players continue to explore similar directions. Earlier initiatives by Alibaba — including investments in startups like Tripo AI and PixVerse — show a consistent push toward building AI systems grounded in physical space rather than purely textual data.

Beyond video generation, world models are seen as a foundational technology for the next wave of AI applications. From humanoid robots to industrial automation, these systems require an understanding of how the physical world behaves — something current AI models only partially achieve.

Industry experts suggest that replicating human-like intelligence will require a combination of reasoning, real-world understanding and continuous learning. While large language models have made significant strides in reasoning and knowledge, world models could fill the critical gap in physical awareness.

As competition intensifies and investments accelerate, the race to build more holistic AI systems is entering a new phase — one where success may depend not just on processing language, but on understanding the world itself.