Jiangsu plans 100+ physical AI datasets and 10+ world-model datasets by end of 2028
In short
Jiangsu has issued an action plan for building high-quality datasets for physical AI and world models in 2026-2028. By the end of 2028 it aims for datasets covering more than 100 key scenarios for physical AI and more than 10 for world models.
Read the full story 2 min read
Jiangsu has issued the Jiangsu Provincial Action Plan for High-Quality Datasets for Physical AI and World Models (2026-2028). Yicai, citing a message from the Jiangsu Provincial Data Bureau, reported that the data bureau, the provincial development and reform commission, the science and technology department and the industry and information technology department formally issued the plan. [ 7 ]
The plan sets an overall goal for the end of 2028. Based on the "1650" industrial system, the "51010" strategic emerging industry clusters and the "6+6" industry AI application needs, Jiangsu is to build physical AI datasets covering more than 100 key scenarios. For the frontier of world models, it is to build more than 10 datasets with causal consistency, spatiotemporal continuity and verifiable applications. These are meant to help models improve physical cognition, real-time deduction and iterative optimisation. [ 1 , 4 , 7 ]
The plan also aims to break through key technologies, including multi-source heterogeneous semantic alignment and fusion processing, ontology modelling, and multimodal data synthesis and simulation enhancement. It calls for a set of standards for data collection, labelling and evaluation, and for typical application scenarios in which data drives physical AI. [ 1 , 4 , 7 ]
On technology, the plan says to review the key bottlenecks in collecting, processing, circulating and applying physical AI and world-model data. For world models it lists multimodal spatiotemporal alignment, cross-domain adaptation and transfer, and causal consistency modelling, and it says to explore a technical system for evaluating and labelling large models for these fields. [ 3 , 5 ]
Another section covers provincial state-owned enterprises. It says they should play a leading role, improve data classification and grading management, and build a data resource system across the whole business chain. Key enterprises are to mine scenario data in fields such as energy and road networks and build physical AI datasets. The plan also calls for open sharing mechanisms that offer licensed use to small and medium-sized firms, universities and startup teams in accordance with rules, to lower the cost of obtaining and using data. [ 2 , 6 ]
Why it matters
The documents describe a provincial target and a list of technical and standards-setting tasks, issued jointly by four provincial departments. They give no budget or enforcement detail, so the practical effect is not yet stated.
Key facts
- The Jiangsu Provincial Action Plan for High-Quality Datasets for Physical AI and World Models (2026-2028) was issued recently. [ 1 , 4 , 7 ]
- Yicai, citing a Jiangsu Provincial Data Bureau message, reported that the data bureau, the provincial development and reform commission, the science and technology department and the industry and information technology department issued the plan. [ 7 ]
- The plan targets physical AI datasets covering more than 100 key scenarios by the end of 2028. [ 1 , 4 , 7 ]
- It also targets more than 10 world-model datasets with causal consistency, spatiotemporal continuity and verifiable applications. [ 1 , 4 , 7 ]
- The plan calls for standards for data collection, labelling and evaluation. [ 1 , 4 , 7 ]
- The plan says key enterprises are to mine scenario data in fields such as energy and road networks and build physical AI datasets. [ 2 , 6 ]
Confirmed by several sources
- Jiangsu's 2026-2028 action plan sets a goal of physical AI datasets covering more than 100 key scenarios and more than 10 world-model datasets by the end of 2028. [ 1 , 4 , 7 ]
- The plan includes work on key technologies such as multi-source heterogeneous semantic alignment and fusion processing, ontology modelling, and multimodal data synthesis and simulation enhancement. [ 1 , 3 , 4 , 5 , 7 ]
Still unclear
- The exact date the plan was issued Outlets say only that it was issued recently or on a recent day; no document gives a date.
- Funding and enforcement for the plan None of the documents mention a budget or how targets will be monitored.
What local media are saying
Timeline, local time
- CLS reported the issuing of the action plan and its 2028 targets. [ 1 ]
- CLS reported the plan's provisions on provincial enterprises and on key technology work. [ 2 , 3 ]
- Jiemian published the same targets and provisions. [ 4 , 5 , 6 ]
- Yicai reported that four provincial departments issued the plan, citing the Jiangsu Provincial Data Bureau. [ 7 ]