KDDI and ELYZA launch physical AI venture, aim to deploy at least 1,000 robots by fiscal 2030

KDDI and ELYZA established ELYZA Robotics to run their physical AI business, and the company began operations on Oct. 1, 2026. The companies plan to invest ¥30 billion in research and development and build a specialist organization of more than 300 people by fiscal 2030. They also aim to deploy more than 1,000 robots in the field by that fiscal year.

KDDI and ELYZA say they will build a system that goes beyond model development to include field data collection, remote monitoring and coordinated operation of multiple robots. For businesses considering deploying robots, the plan underscores the need to design communications, operations and data use alongside hardware selection.

The targets through fiscal 2030 are ¥30 billion in research and development investment, a specialist organization of more than 300 people and deployment of more than 1,000 robots in the field. However, the companies have not disclosed a breakdown of the investment or its allocation by fiscal year. The announcement also does not specify the types of robots or how they will be allocated across industries. The plan’s substance will depend not only on its scale but also on how operations connect research and development with deployment. The announcement centers not on manufacturing robots in-house, but on a structure for adapting different types of robots to specific tasks and deploying them in the field alongside models and systems. In addition to developing VLMs and VLAs, the company will work on collecting work data, reproducing work sites in simulators, distributed inference and remote monitoring. The plan to feed data from the field back into research and development appears intended to create a cycle in which information from deployed robots informs subsequent development, rather than ending with model creation. The announcement materials do not provide specific details on data formats, rights or safety management. A report by Physical AI News, KDDI to install Rubin at Osaka Sakai data center, expanding inference infrastructure for physical AI, covered developments on the computing infrastructure side. In this case, the “Digital Belt” and MCC are intended to serve a layer closer to robots’ field operations. The articles cover separate announcements, but both point to the importance of systems that include communications, monitoring and real-world robot operations, in addition to computing resources. While NTT Docomo develops “Physical AI collaboration platform” linking AI agents and robots focuses on motion management and permission controls, ELYZA Robotics’ announcement covers foundation models, field data and the operation of multiple robots. There is no indication that the two systems will directly interoperate; they can be compared only in terms of the layers they address. In industry, the company is expected to handle robot selection and adaptation, integration with on-site systems, and data collection as a unified service. It cites applications in settings including retail, logistics and manufacturing, but has not yet disclosed actual utilization rates, task-completion levels or the operational burden after deployment. Questions to watch include which robot models and tasks will be covered first in the 1,000-robot plan, what conditions will enable a shift from proof-of-concept projects to ongoing operations, and how field data will be reused for training. Without those details, it is not possible to assess how the investment and deployment targets will translate into results in real-world operations.

ELYZA Robotics will develop foundation models for robots, including VLMs that process images and language, and VLAs that generate robot actions. Its research will also cover collecting work data recorded from a human perspective and technology for reproducing work sites in simulators. Alongside communications networks, the company will use the “Digital Belt,” which combines a nationwide low-latency network with AI computing resources, and the Mobility Control Center (MCC), a remote monitoring hub. The business has three pillars: research and development; integrating robot systems into business operations; and developing platforms to support research and development and field operations. The company will conduct proof-of-concept projects in a wide range of settings starting in 2026, followed by phased deployment. It plans to develop two types of platforms: one to support model development and the accumulation of field data, and another to support distributed inference, remote monitoring and coordinated work by multiple robots. The plan is to feed data and insights from the field back into research and development.