KDDI to Deploy Rubin at Osaka Sakai Data Center, Expand Inference Infrastructure for Physical AI

newsroom.kddi.com reported on October 27, 2026, that KDDI plans to deploy Rubin at its Osaka Sakai data center and expand its inference infrastructure for physical AI. We have not been able to verify the primary source, so please refer to the original for details.

KDDI says it will provide an inference environment where customers can choose Gemini, Gemma or proprietary enterprise models based on their needs while storing and managing sensitive data in Japan. The plan is relevant to companies that prioritize data processing locations and dedicated environments, but the published information does not clarify specific terms for physical AI use or details of security measures.

The announcement centers not on the unveiling of robots or autonomous-control technology, but on an environment for running inference with multiple AI models at a domestic data center and upgrades to its computing infrastructure. The planned expansion of available models starting in December 2026 and the introduction of Rubin in fiscal 2027 suggest that KDDI intends to develop model options and processing infrastructure in parallel. Physical AI is cited as a potential application, but the announcement does not specify how the environment would connect to control systems for robots or drones, or how processing would be divided between on-site computers and the data center. On performance, the announcement cited increases in memory capacity, bandwidth and FP4 computing performance, as well as up to 10 times the inference throughput. However, the throughput figure was published by NVIDIA and is based on a comparison with HGX B200 under specific model and operating conditions. It should not be taken to mean that the same processing performance will be achieved in actual enterprise operations or robot deployments. Shorter construction times through containerized data centers and lower cooling power consumption through free cooling are also design benefits cited by KDDI, and should be distinguished from operational results after deployment. An article we published, KDDI and RLWRLD Showcase Physical AI Collaboration Case (October 4, 2026), introduced the companies’ work. The latest announcement can be seen as KDDI outlining, from a different angle than an individual collaboration case, its vision of using an inference environment for physical AI as well. However, published information does not confirm any specific connection between the two initiatives or whether they will use a shared model or data platform. The business focus is on providing an environment for managing data in Japan, access to multiple models and computing infrastructure as an integrated offering. This could be an option for companies with requirements on where data is stored and processed, but whether domestic storage and separation between users can prevent information leaks—and to what extent—requires separate verification. A Nikkei article identified measures to prevent information leaks as a challenge, but KDDI’s official announcement did not provide details of specific security measures. Going forward, it will be necessary to confirm the number of GPUs, available capacity, pricing, specific terms for uses such as robotics, and operational safety and security measures.