Noetra, backed by core member companies Sony, SoftBank, NEC, and Honda Motor, launched full-scale R&D on July 17, 2026, for a Japan-developed multimodal foundation model designed to power AI-enabled robots and physical AI. The initiative has drawn investment from 44 companies and organizations, predominantly from the manufacturing sector, all aligned on building sovereign AI infrastructure for Japan's industrial base.
The R&D organization centers on engineers seconded from its four core investors, the National Institute of Advanced Industrial Science and Technology (AIST), Preferred Networks, and other participating organizations. Noetra will combine the model development expertise these groups have built over years of AI work to accelerate the program.
Computing infrastructure and the Nvidia partnership
Initial development will run on AI computing infrastructure operated by Japan-based providers. For the heavier training workloads ahead, Noetra plans to build a dedicated cluster equipped with approximately 27,500 Nvidia Rubin GPUs - Nvidia's latest graphics processors optimized for agentic AI workloads tied to large-scale foundation models. Construction starts in April 2027, with operations expected by June 2028.
The Rubin GPU architecture, which Nvidia positions for data center AI workloads, signals the scale of compute Noetra expects to need. For IT infrastructure teams, a buildout of this size represents one of the largest single AI training clusters planned in Japan, with direct implications for networking, storage, and power provisioning at facilities that host it.
A three-phase roadmap to physical AI
Noetra's development timeline spans three stages. Starting in phases from fiscal 2026 (ending March 31, 2027), the first target is a reasoning foundation model built for AI agents and natural language processing - equipped with advanced Japanese language understanding, logical reasoning, and instruction following.
By fiscal 2028, the plan shifts to an omni-modal foundation model that can process text, images, video, and audio together. The goal is AI that works across data types rather than requiring separate models for each modality. Looking ahead to fiscal 2030, Noetra aims for what it calls "Real-world Native AI" - models that understand physical properties such as spatial awareness and can deploy in real-world environments like factory floors and robotics platforms. This progression from language to multimodal to embodied AI reflects the broader direction of Generative AI and LLM research, where frontier models increasingly handle diverse input types.
Models will be released externally in stages, tied to R&D progress and real-world implementation readiness.
What leadership is saying
Hironobu Tamba, president and CEO of Noetra, said: "For Japan to become a global leader in physical AI, it is essential to develop multimodal foundation models that will strengthen the nation's industrial competitiveness while helping address societal challenges and creating new value. Working together with our partners, Noetra will build these multimodal foundation models while striving to realize trusted AI infrastructure that supports the transformation of Japan's industries and society."
Junichi Miyakawa, president and CEO of SoftBank, framed the effort around data sovereignty. "In a society that coexists with AI, the data held by Japan's industries and businesses will be a key source of competitive strength. Creating an environment in which that data can be securely utilized within Japan is essential to strengthening the country's industrial competitiveness." He said SoftBank expects Noetra to play a central role in building that environment and will support the effort with AI infrastructure and other resources.
Why this matters for IT and development
For teams managing AI infrastructure, Noetra's 27,500-GPU cluster will demand expertise in large-scale distributed systems, high-bandwidth interconnects, and power-constrained data center design - skills that are in short supply globally. The phased model releases also create a concrete timeline for developers working on robotics, manufacturing AI, and agent-based systems to plan integration work around known milestones rather than speculative roadmaps.
The emphasis on sovereign AI - keeping Japan's industrial data within domestic infrastructure for training - mirrors trends in other markets where regulated industries require onshore compute. IT leaders supporting multinational operations should watch how Noetra's AI for IT & Development patterns evolve, particularly around data governance and model access controls, as these will set precedents that influence procurement and compliance decisions across Asia-Pacific manufacturing networks. The 44-company consortium also signals that AI infrastructure decisions are shifting from purely IT-led initiatives to cross-industry collaborations, a model that reshapes how technical teams justify and scope large-scale compute investments.
Your membership also unlocks:
