appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Minxiao Chang

Artificial intelligence has spent years learning to generate words, images and videos. Now, the industry’s attention is shifting towards “world models” – AI systems designed to simulate how the physical or digital world changes over time.
While the term was originally associated with physics simulations and robotics, a growing number of Chinese companies are applying the concept more broadly. The technology remains in its infancy, with little consensus on what a world model should ultimately look like.
However, China has emerged as one of the most active testing grounds for these competing approaches.
Here’s what you need to know about the next frontier of AI.
What is a world model – and why the sudden buzz?
World models have quickly become one of AI’s hottest buzzwords, with big tech names jumping into the fray.
Nvidia recently introduced the latest update to its robotics world model platform Cosmos, AI “godmother” Li Fei-fei’s World Labs is developing systems to understand and simulate the physical world, and Alibaba Group Holding has entered the race with Qwen-AgentWorld, calling it the first native language-based world model. Alibaba owns the South China Morning Post.
The renewed interest reflects the rapid rise of AI agents and robotics, where simply generating text or video is no longer enough. Increasingly, AI systems are expected to predict how environments will respond to actions before they are taken.
Unlike conversational AI like ChatGPT, which predicts the next word in a sentence, a world model predicts the next state of a physical environment.
What makes the current wave distinct is that companies are building their products based on fundamentally different interpretations of what a “world” should be.
Companies like Google, Nvidia and World Labs remain focused on modelling physical reality, forecasting how objects move, collide or interact.
Alibaba’s Qwen-AgentWorld applies the same logic to digital environments, predicting how web browsers, coding terminals and operating systems will respond to an AI agent’s commands.
Rather than describing a single technology, “world model” has become an umbrella term encompassing physical reality, interactive virtual spaces and digital software environments.
World models could become a foundational layer of AI, supporting applications ranging from robotics to enterprise workflows, according to a Goldman Sachs report on AI published in June.

Where is China gaining traction?
Globally, the technology is in its early stages, with no company establishing a dominant commercial framework yet, according to Cao Yanpei, chief scientist at Chinese 3D AI company Vast. “The whole field is still exploring,” he said.
However, China was already moving beyond research labs and into real-world deployment across three primary sectors, according to Zhu Zheng, chief scientist at GigaAI.
Autonomous driving was seeing the strongest commercial traction, Zhu said, with firms using world models as simulation engines and synthetic data generators.
Because self-driving vehicles continuously generated real-world operational data, they created a tight feedback loop for model training, said Chen Bihan, a Singapore-based AI practitioner working in AIoT and AI data. China’s extensive logistics networks also provide an ideal environment for data collection.
Meanwhile, in the embodied AI and robotics sector, developers increasingly rely on world models to simulate environments, generate training data and map out control policies for physical robots. Examples include Alibaba’s Qwen-RobotWorld, Ant Group’s LingBot-World and SenseTime’s Kairos series.
Tech giants are also leveraging world models for interactive AI-generated content (AIGC). Alibaba’s HappyOyster allows users to explore and alter virtual worlds in real time, while Kuaishou’s Kling AI and ByteDance’s SeedDance push the boundaries of realistic and interactive video generation.

Does China have an edge in world models?
GigaAI’s Zhu estimated that Chinese world models currently lagged their US counterparts by about three to six months – a significantly smaller gap than seen in large language models. “At the foundation-model level, China and the US are relatively close,” he said.
Still, researchers caution against framing the global competition as a simple geopolitical race. Instead, the real distinction lies in research priorities and data availability.
US developers leaned towards general-purpose foundation models and simulation tools, while Chinese companies often iterated directly within industrial settings – such as manufacturing, logistics and autonomous driving – where real-world operational data could be collected continuously, said Wang Xiaofeng, algorithm partner at Giga AI.
For now, though, the race is less about determining which country is ahead, and more about discovering which technical approaches will prove most useful as world models transition from experimental research into practical, everyday applications.


Philippine tribe vows to stand ground against US-led Pax Silica: ‘listen to us’

As the US weighs restrictions on cloud computing, how will China’s AI sector adapt?

Chinese AI chipmaker Hygon plots expansion from data centres to robotics

For China’s hi-tech IPOs, it’s Hong Kong vs the mainland – but how much does it matter?

Artificial intelligence (AI)

The next frontier of AI: how ‘world models’ are simulating reality and virtual spaces

GigaAI specialises in embodied intelligent models. Photo: Handout

Alibaba is pushing Qwen models to the physical world with its Qwen-Robot Suite. Photo: Handout

RynnBrain is an embodied foundation model based on Alibaba’s Qwen3-VL. Photo: Handout

Glass crashes slashed? Ant Group embodied AI unit claims breakthrough in robot sensing

A wheeled humanoid robot developed by Robbyant is unveiled at the company’s booth at the Inclusion Conference on the Bund on September 11, 2025. Photo: Wency Chen

An engineer trains a humanoid robot in household tasks at X Square Robot’s facilities in Shenzhen, in southern China’s Guangdong province, on May 22. Photo: AFP

Google DeepMind director Cao Liangliang makes a boomerang return to Hong Kong

Machine learning and computer vision expert Cao Liangliang says his appointment as chair professor of data science and artificial intelligence at PolyU is “a full-circle moment for me”. Photo: Handout

Tech

Tech Trends

News

China Future Tech

Increasingly, AI systems are expected to predict how environments will respond to actions before they are taken


Increasingly, AI systems are expected to predict how environments will respond to actions before they are taken.


The next frontier of AI: how ‘world models’ are simulating reality and virtual spaces

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Increasingly, AI systems are expected to predict how environments will respond to actions before they are taken

Increasingly, AI systems are expected to predict how environments will respond to actions before they are taken