According to a report by Jiemian News today, the Institute of Artificial Intelligence at the China Academy of Information and Communications Technology, together with Humanoid Robot (Shanghai) Co., Ltd. and the Embodied Intelligence Testing Laboratory, jointly released the Research Report on Embodied Intelligence Training Grounds (2026). The report points out that the training-ground industry chain has taken initial shape, exhibiting an overall inverted-triangle structure characterized by active upstream technology supply and downstream applications that still need to be expanded.
Training grounds require substantial investment in fixed assets. Construction costs cover multiple areas, including scenario development, data collection and hardware equipment, computing power, networks, storage, and management systems. Daily operating expenses are also considerable.
Current commercial services mainly center on selling data products, while providers are exploring services such as model and hardware training and validation, joint research and development, and testing and verification. Selling data products generates revenue most quickly, but relying solely on data sales makes it difficult to cover the heavy fixed-asset investment, resulting in a long payback period. Most training grounds were built only recently, so there is an urgent need to establish diversified revenue structures through platform-based transactions, expanded value-added services, and approaches such as “training as a service.”