The technical route for embodied AI remains undetermined at this stage. Nevertheless, as robot bodies enter mass production, the industry's focal point has shifted from merely manufacturing robots to the ongoing production and iterative enhancement of robot capabilities, with the mass production of capabilities now at the heart of development. To realize this mass production of capabilities, three critical areas must be addressed: enhancing model iteration efficiency, ensuring a steady stream of continuous data production, and deploying capabilities across different robot bodies and scenarios. Presently, during phases aimed at boosting R&D efficiency and facilitating large-scale deployment, embodied AI companies are finding common ground in their demands for underlying infrastructure. Their primary concerns revolve around optimizing model training and inference efficiency, managing full-link data processing, and adapting to and deploying across various robot bodies. A full-stack infrastructure can consolidate the common facets of capability production into a reusable foundation. This, in turn, enables companies to concentrate on models, products, and scenarios, thereby boosting the efficiency of capability generation and facilitating the swift validation, replication, and iteration of skills.