KDDI to expand AI inference capabilities at Osaka Sakai Data Center

KDDI has announced an expansion of the AI inference environment at its Osaka Sakai Data Center. Starting in December 2026, the facility will increase its range of available AI models to include open-weight models such as Google’s ‘Gemma’, alongside the existing ‘Gemini’ frontier model. This expansion aims to provide a secure environment where companies can process sensitive data within Japan while selecting models tailored to specific requirements, including proprietary models run in isolated environments.

Early partners for this environment include Ricoh and the Medical Engineering Institute (MEI). Ricoh intends to test the effectiveness of secure generative AI utilization without the need to procure and maintain its own GPU servers, while MEI plans to utilize specialized medical AI models for big data analysis.

To further enhance computational power, KDDI plans to introduce NVIDIA’s next-generation ‘HGX Rubin NVL8’ GPU platform in fiscal year 2027. This hardware offers significant performance improvements over the previous Blackwell generation, including up to 10 times the inference throughput. To support this, KDDI will implement containerized data center facilities and direct liquid cooling technology, which is designed to reduce cooling-related power consumption by up to 90% through the use of free cooling.