Woofun AI reports that Japan officially launched the "Noetra" domestic multimodal foundational model project on July 16, featuring participation from 44 entities including Sony Group, SoftBank, NEC, and Honda. The initiative, aligned with the Ministry of Economy, Trade and Industry's FRONTia Project, aims to develop a Japan-made foundational model for physical AI applications in manufacturing and robotics rather than conversational interfaces.

The infrastructure will utilize an NVIDIA Vera Rubin NVL72 rack and DSX platform architecture, incorporating approximately 27,500 Rubin GPUs and 13,750 Vera CPUs with a 140-megawatt power capacity. Construction is scheduled to begin in April 2027, with operations commencing in June 2028. The development roadmap targets a reasoning core for AI agents by fiscal 2026, a fully multimodal model by fiscal 2028, and real-world native AI understanding spatial attributes by fiscal 2030.