软银新闻介绍日本 Noetra 物理 AI 项目与未来算力规划

On July 16, 2026, Japan's Ministry of Economy, Trade and Industry (METI) hosted a press conference to explain its physical AI policies and initiatives. Government officials, leading experts and representatives from participating companies took the stage to discuss policy priorities and the R&D framework for advancing the deployment of physical AI in the real world.

Noetra Corp.—a newly formed company that lists Sony Group Corporation, SoftBank Corp. (TOKYO: 9434), NEC Corporation, and Honda Motor Co., Ltd. as core members and investors—is leading the development of a Japan-developed multimodal foundation model in collaboration with numerous other Japanese companies. Noetra representatives were joined at the event by Japan’s National Institute of Advanced Industrial Science and Technology (AIST) and the New Energy and Industrial Technology Development Organization (NEDO).

Building trustworthy AI with a natural understanding of the real world

In his address, Satoshi Nohara, Director-General of METI’s Commerce and Information Policy Bureau, explained that as generative AI and AI agents continue to advance, AI is expanding beyond cyberspace into real-world environments, such as factories and warehouses, making it possible to solve real-world challenges with robots and mobility systems.

METI's vision calls for deploying physical AI in real-world environments, where it can constantly collect operational data to further improve AI models and create a continuous loop of learning and deployment.

This real-world data includes information critical to industrial competitiveness, such as equipment operating conditions, production workflows, the movement of people and objects, as well as image, video, audio and sensor data. Establishing an environment in which this data can be securely utilized within Japan’s borders will provide the foundation for developing homegrown AI, Nohara explained. By creating AI rooted in Japan's own real-world environments—rather than relying solely on models developed outside the country—METI’s initiative will expand the range of AI options available to Japan-based companies.

Core to this vision is the development of multimodal foundation models capable of processing diverse forms of data. Named after the initials of “ F oundation for R eal-world O mni- N ative T rustworthy I ntelligence ＆ A lignment,” Nohara introduced the “FRONTia Project” as a METI-driven collaborative effort, which as a stated goal of building trustworthy AI with a natural understanding of the real world.

Laying a foundation to support physical AI applications across a wide range of industries in Japan

In this context, Neotra officially launched R&D efforts to develop a Japan-developed multimodal foundation model that will serve as the foundation for AI-enabled robots and physical AI.

Noetra and AIST were selected for NEDO's “Multimodal Foundation Model Development Project for AI Robots and Physical AI” and will conduct R&D under the program. Sony Group, SoftBank, NEC and Honda Motor are participating as Noetra's core member companies and investors, and Noetra also received investments from a total of 44 companies and organizations in various industries, led by the manufacturing sector, that are promoting AI adoption.

Noetra will develop foundation models that support physical AI applications across a wide range of industries, while AIST will advance research on the advanced technologies required for those foundation models in collaboration with research institutions in Japan and around the world. METI and NEDO will support these R&D efforts through a governance framework that monitors progress and periodically reviews the project's course.

Noetra to build core AI that serves as a “teacher” for other AI systems

Hironobu Tamba, President and CEO of Noetra, explained that engineers spearheading AI development in Japan assembled at Noetra to advance the project in collaboration with AIST.

Describing the AI that Noetra aims to build, Tamba said it will serve as the core AI that acts as a “teacher” or “mentor” for other AI systems. He emphasized that Noetra's approach begins with the expertise developed on the front lines of Japanese industry, using insights gained from real-world operations to improve AI models, and to reflect advancements back into the field to create a continuous cycle of innovation.

Daisuke Okanohara, who is leading model development at Noetra as the Joint R&D Project Director, described the company's vision as developing “AI that can observe, understand and work in the real world.” To achieve this, Noetra plans to develop both reasoning foundation models and generative foundation models in parallel.

As part of its foundation model roadmap, Noetra will develop AI with core capabilities that include advanced Japanese language understanding, logical reasoning and instruction following. Starting in phases from the year ending March 31, 2027 (FY2026), the company plans to build a reasoning foundation model that will serve as the core of AI agents and natural language processing. By FY2028, it aims to develop an omni-modal foundation model capable of seamlessly processing text, images, video and audio. Looking ahead to FY2030, Noetra will pursue the realization of “Real-world Native AI” capable of understanding physical properties such as spatial awareness and designed for deployment in real-world environments.

In addition, with support from NVIDIA, Noetra plans to begin construction of AI computing infrastructure equipped with approximately 27,500 NVIDIA Vera Rubin GPUs in April 2027. The infrastructure, optimized for agentic AI workloads centered on large-scale foundation models, is scheduled to begin operations in June 2028.

While taking R&D progress and real-world deployment into consideration, the models developed by Noetra will be released in stages and be made available for external use. Tamba said Noetra aims to provide Japan with an alternative to overseas AI models, and that the company plans to release models that can be used safely and reliably on a year-by-year basis.

Collaborating with leading global research institutions to apply state-of-the-art AI

In his presentation, Masahiro Hamasaki, Deputy Director of the Artificial Intelligence Research Center (AIRC) at AIST noted that realizing physical AI will require more than just developing foundation models; it will also depend on advances made across a broad range of research fields, including natural language processing, computer vision, speech, robotics and AI agents.

He explained that the project will launch joint research initiatives in collaboration with leading researchers from Japan and around the world. By working with researchers in the United Kingdom, France, Germany, Canada, the United States and Japan, AIST aims to advance state-of-the-art AI research and apply those results to the development of Noetra's foundation models.

Building on Japan's strengths to accelerate the adoption physical AI

Physical AI remains an emerging field that requires continued R&D investment. At the same time, physical AI presents a significant opportunity for Japan to build on its fundamental strengths in manufacturing, real-world operational data and robotics. The event demonstrated how a public-private partnership, with support from research institutions, can help accelerate the adoption of AI in the real world.

(Posted on July 24, 2026)
by SoftBank News Editors