Guotai Haitong: AI Moves Toward the Physical World, Data Services Usher in New Opportunities

date
09:12 16/11/2026
avatar
GMT Eight
As the data services market expands, companies with data resources, large-scale production, and platform service capabilities are expected to benefit.
Guotai Haitong released a research report stating that physical AI is accelerating its development, demand for high-quality data continues to grow, and data supply has become a key factor constraining model development. As the data services market expands, enterprises with data resources, large-scale production capabilities, and platform service capabilities are expected to benefit. Focus on two major directions: real data collection and processing, and simulation synthesis and spatial data. 1) Real data collection and processing: enterprises with real-scenario resources, multimodal data accumulation, and mature collection and processing capabilities are expected to benefit; 2) Simulation synthesis and spatial data: enterprises with high-fidelity physics engines, 3D scene generation, and spatial data resource advantages are expected to benefit. Guotai Haitong's main views are as follows: Physical AI and world models are accelerating, and data demand continues to grow AI is extending from the digital world to the physical world, with model capabilities gradually expanding from environmental perception to physical state prediction, Siasun Robot&Automation motion generation, and complex task execution. Google, NVIDIA, Tencent, and other vendors are accelerating their layout in world models, driving the accelerated integration of VLA and world models. Model training data is also expanding from traditional text, images, and video to multimodal interaction data such as real operation trajectories, joint states, force perception, and tactile feedback. On the data supply side, publicly available internet data, bodyless data, simulation-synthesized data, and real-machine data each have their own advantages and jointly support model training. In 2025, the global physical AI and world model data services market size was approximately RMB17.75 billion, and it is expected to reach RMB131.23 billion by 2030, with a CAGR of approximately 49.2% from 2025 to 2030. The China market is expected to grow from RMB8.28 billion to RMB64.99 billion over the same period, with a CAGR of approximately 51.0%. Domestic and overseas vendors are accelerating their layout, and a professional data service system is gradually taking shape Overseas vendors are mainly leveraging their existing AI data services, expert networks, and model evaluation capabilities to expand into the physical AI field. Among them, Scale AI launched the Physical AI Data Engine, laying out Siasun Robot&Automation teleoperation, first-person data collection, and multimodal annotation; Surge AI, Mercor, micro1, and Handshake AI are expanding services around expert training data, model evaluation, reinforcement learning environments, and real-scenario data. Domestically, Guanglun Intelligent focuses on physical AI data and simulation evaluation infrastructure, forming a full-stack product system covering human behavior data, simulation synthesis, simulation evaluation, and deployment feedback. Data resources, cost efficiency, and platform ecosystem build long-term competitive barriers In terms of data resources, mature embodied models require real data at the tens of millions of hours level, while current global high-quality real data is only at the hundreds of thousands of hours level. Enterprises with stable data sources, rich scenario coverage, and standardized quality control capabilities have greater advantages. In terms of cost efficiency, real-machine collection costs are relatively high, and methods such as UMI, EGO, and simulation synthesis are expected to reduce data production costs. AI agent-assisted collection is also expected to reduce manual input and improve data output efficiency. In terms of platform ecosystem, data services are upgrading from single collection and annotation to full-chain services such as data management, model training, evaluation, and deployment feedback. Enterprises with cross-embodiment adaptation, platform integration, and industrial chain collaboration capabilities are expected to enhance customer stickiness and sustained service capabilities. Risk warnings: Physical AI industry development falls short of expectations; progress in world models and simulation technology falls short of expectations; changes in data collection technology routes; intensifying industry competition and commercialization falling short of expectations; data security, copyright, and compliance risks.