AWS targets physical AI deployment gap

AWS's move comes as generative AI has largely lived in the digital realm, but physical AI extends intelligence into robots and equipment that must act in unpredictable environments. Traditional industrial automation struggles with deviations, whereas physical AI systems can learn from experience. The article underscores that the model alone is not enough—data, compute, simulation, and deployment operations are equally important. For instance, RLWRLD's robotics foundation model for five-fingered dexterity relies on hundreds of terabytes of factory data, and Config's pipeline collects 20,000 hours of action data monthly yet still faces data-diversity issues. AWS's strategy is to provide a common foundation across the lifecycle, from cloud-scale training to real-time edge inference, rather than becoming a robot maker. The company acknowledges the ecosystem is still evolving and does not yet offer a single branded simulation service. The path to deployment remains complex, but AWS aims to reduce the bespoke effort involved, potentially making physical AI more accessible to enterprises.