Noetra began full-scale research and development on a Japan-developed multimodal foundation model on July 16, 2026, backed by 44 investing companies and organizations. Sony is one of four core member companies and investors alongside SoftBank, NEC and Honda; Noetra is not identified as a Sony subsidiary. The project is aimed first at Japanese-language reasoning and AI agents, then at systems that process text, images, video and audio for robotics and other physical-AI uses.
A 27,500-GPU system is planned for June 2028
Noetra says development will initially use AI computing infrastructure operated by providers in Japan. A much larger system is planned in collaboration with Nvidia: approximately 27,500 Rubin GPUs, with construction scheduled to begin in April 2027 and operations expected in June 2028. The announcement describes that infrastructure as a plan, not a facility that is already built or available to researchers.
The R&D organization draws engineers from the core investors and other participants, including Japan's National Institute of Advanced Industrial Science and Technology and Preferred Networks. That structure gives Noetra access to model-development and industrial expertise without making any one core investor the sole owner or operator described by the release.
The roadmap starts with Japanese reasoning before adding video and audio
From fiscal 2026, the company aims to build a reasoning foundation model for AI agents and natural-language processing, with advanced Japanese understanding, logical reasoning and instruction following. Its fiscal 2028 target is an omni-modal model that can process language, images, video and audio together. By fiscal 2030, Noetra wants what it calls real-world native AI, designed to understand physical properties such as spatial relationships and operate in real environments.
Sony links the research to entertainment and semiconductors
Sony Group chief executive Hiroki Totoki says the company expects knowledge from the joint development effort to apply to entertainment as well as semiconductors used in physical AI. The announcement does not name a PlayStation game, film, music product or Sony device that will use the model, so those applications remain broad areas of interest rather than announced products.
No model or consumer service is available yet
Noetra says models will be offered externally and released in stages as research and real-world implementation progress. It does not publish model sizes, training-data details, evaluation results, licensing terms or a date for the first external release. The July announcement therefore establishes an organization, infrastructure plan and development timetable—not the performance or safety of a finished system.
