Alibaba Cloud is investing heavily in a fundamentally different type of artificial intelligence, one designed to understand and simulate the physical world rather than process text. The Chinese tech giant led a 2 billion yuan ($290 million) investment in ShengShu, the startup behind the AI video generation tool Vidu, signaling a major industry shift away from large language models (LLMs) like ChatGPT toward what researchers call "world models" .

This move reflects a growing recognition among AI developers that the current generation of text-based chatbots has fundamental limitations. While LLMs excel at language tasks, they struggle with understanding how the physical world actually works, how objects move, and how actions produce consequences. World models, by contrast, are built on video and real-world scenarios, allowing AI systems to learn the rules of physics and causality in ways that pure language training cannot achieve .

What's the Difference Between World Models and Language Models?

The distinction matters because it determines what AI can actually do in the real world. Large language models are trained primarily on text data, making them powerful at answering questions, writing essays, and engaging in conversation. But they lack something fundamental: an intuitive understanding of how physical systems behave. A language model might know the definition of "gravity," but it doesn't truly understand how objects fall or how a robot should move through space .

World models, by contrast, learn from video and multimodal data including vision, audio, and touch sensors. This approach allows AI to develop a more natural understanding of cause and effect in the physical world. ShengShu, the three-year-old startup receiving Alibaba's investment, explicitly stated this advantage in its announcement .

"ShengShu believes that a general world model, built on multimodal data such as vision, audio, and touch, more naturally captures how the physical world works than large language models," the company said in a statement.

ShengShu, AI Video Generation Startup

The founder of ShengShu emphasized the practical goal driving this technology: connecting perception with action. "We aim to connect perception and action," stated Zhu Jun, founder of ShengShu, "allowing AI systems to better model and predict real-world behavior consistently" .

of ShengShu

Why Does This Matter for Robotics and Autonomous Systems?

The investment timing reveals why major tech companies are suddenly prioritizing world models. These systems are essential for robotics because robots need to understand the physical world in ways that pure language processing cannot provide. A humanoid robot, for example, must predict how its arm will move when it reaches for an object, how much force to apply, and what will happen if it miscalculates. Language models alone cannot provide this understanding .

According to Kevin Kelly, co-founder of the tech magazine Wired, AI will ultimately need three capabilities to replicate human intelligence: reasoning, understanding of the physical world, and continuous learning. While LLM-powered chatbots have created the knowledge element, world models represent the critical breakthrough area needed for the next phase of AI development .

ShengShu's latest model, Vidu Q3 Pro, released in January, ranks among the top 10 AI models for generating videos from text and images according to Artificial Analysis. The company launched Vidu globally before OpenAI made its now-shuttered Sora tool widely available, positioning ShengShu as an early leader in this space .

How Alibaba Is Building an Ecosystem Around World Models

• Direct Investment in Video Generation: Alibaba led the $290 million Series B funding round for ShengShu, with participation from TAL Education and Baidu Ventures, specifically to develop a general world model bridging digital and physical domains.
• 3D Model Generation: Alibaba and Baidu Ventures led a $50 million investment in Tripo AI, a platform that generates 3D digital models from photographs and is developing its own world model grounded in physical space.
• Interactive Video Control: Alibaba led a $60 million investment in PixVerse, which released an AI world model allowing users to direct how videos unfold during generation, demonstrating real-time physical understanding.

Beyond these direct investments, Alibaba has released free, open-source AI models for video generation and, in February, launched a model specifically designed for powering robots. This multi-pronged approach suggests the company views world models as foundational infrastructure for the next generation of AI applications .

ShengShu has also announced strategic partnerships with companies developing embodied AI, which refers to systems like humanoid robots that physically interact with the world. These partnerships span industrial, commercial, and home settings, indicating that world models are moving from research into practical deployment .

The competitive landscape includes other major players. Chinese short-video companies Kuaishou and ByteDance have released competing AI tools for video generation, though Alibaba's broader ecosystem approach and substantial capital commitments suggest a longer-term strategic bet on world models as the foundation for future AI capabilities .

This shift represents more than just a new investment trend. It signals that the AI industry is moving beyond the text-based paradigm that dominated the past two years, toward systems that can understand and simulate physical reality. For robotics, autonomous driving, and other applications requiring real-world interaction, this transition may prove as significant as the rise of large language models itself.