China’s edge over US in AI world models: abundant data, faster deployment, executive says

China’s world models are benefiting from early integration with the country’s strong industrial base, which gives the domestic ecosystem greater momentum than that of the US, according to an executive at a leading Chinese world model start-up.World models simulate 3D environments and physical dynamics, and are expected to help train the next generation of physical artificial intelligence applications, including robots and autonomous vehicles.Wang Xiaofeng, an algorithm partner at GigaAI, said Chinese world models were increasingly being deployed in industrial and governmental applications, while their US counterparts were focusing more on AI-assisted gaming and design.The Beijing-based start-up’s latest world model, GigaWorld-1, outperformed models from Google and Nvidia across visual quality, physics adherence and 3D accuracy, according to WorldArena, an industry benchmark that evaluates the capabilities of world models.Wang, a PhD graduate from the Institute of Automation at the Chinese Academy of Sciences who led research and development of GigaWorld models, said China held an advantage over the US in developing breakthroughs because of its abundance of real-world data.“What the embodied intelligence field lacks right now is large-scale, structured data,” he said. “China’s advantage lies in its extensive industrial scenarios and the ability of government-led platforms to organise data collection at scale, creating a more comprehensive data source to support robotics development.”Founded in 2023, GigaAI is one of China’s earliest ventures to experiment with world models. Photo: HandoutGigaAI’s founder and CEO, Huang Guan, holds a PhD in automation from Tsinghua University and previously worked at Microsoft Research Asia and China-research-institute" class="entity-link entity-organization" data-entity-id="116620" data-entity-type="organization">Samsung China Research Institute.China had a far greater scale of robot deployments than the US, with faster maintenance cycles and more data from factories that could be used to support further training of world models, Wang said.Investor excitement has followed. GigaAI recently completed a 1 billion yuan (US$146 million) funding round just weeks after securing a similar amount in March, underscoring growing domestic investor interest in an emerging area with potential to accelerate physical AI developments.Founded in 2023, GigaAI was one of China’s earliest ventures to experiment with world models. It was now generating tens of millions of yuan in annual revenue, according to a source who requested anonymity as they were not authorised to speak. The company has partnered with electric vehicle giants including Li Auto, Xpeng and BYD to develop vision-based autonomous driving systems.Further ReadingUS investors have also shown strong appetite for world model start-ups. AI pioneer Li Fei-Fei’s World Labs and Turing award winner Yann LeCun’s AMI Labs have both secured funding rounds of around US$1 billion in the first quarter of 2026, underscoring intensifying investor competition in the sector.World models are also finding application in autonomous driving, with Google DeepMind announcing a partnership with Waymo in February to use its Genie world models to train its self-driving fleet. Tesla said it was applying world models to train its humanoid robot Optimus.In China, Alibaba Group Holding’s mapping arm Amap launched its world model research efforts in January. Tencent Holdings also open-sourced multiple world models last year, allowing developers to generate and interact with 3D environments from single image or text prompts. Alibaba owns the South China Morning Post.Questions remain, however, about whether world models can be powerful enough to simulate the complexities of the physical world to train safe and reliable physical AI applications, as well as their commercial viability.