Japan's Noetra Partners with Sony, SoftBank and 44 Companies to Advance Sovereign AI, Plans to Deploy ~27,500 NVIDIA Rubin GPUs

0xBroomberg
Published todayAbout 9 min read

Japanese AI firm Noetra has launched the country's largest sovereign-AI initiative with Sony, SoftBank, NEC, Honda and 44 corporate backers, planning a compute cluster armed with 27,500 Nvidia Rubin GPUs to build a full-stack AI capability — from chips to models — on home soil.

01

What exactly is Japan trying to build?

Noetra's goal is a complete domestic AI chain — from compute infrastructure to multimodal foundation models — built and operated in Japan.
This means → Japan no longer wants to rely solely on American models and cloud compute; it wants to own the full stack, hardware through model weights.
In plain terms = Japanese companies have mostly called OpenAI and Google APIs; this time they are training their own model from scratch.
02

Who is paying and who is building?

Four core members: Sony, SoftBank, NEC and Honda — spanning entertainment, telecom, IT and automotive.
The broader coalition numbers 44 investors, including manufacturing heavyweights (Fanuc, Yaskawa, Mitsubishi Electric, Tokyo Electron), Japan's three mega-banks (MUFG, Mizuho, SMBC), and tech firms such as Rakuten and Sakana AI.
This reflects a national-scale bet: sovereign AI is not one company's project but a collective wager by manufacturing, finance and energy.
03

Where will the compute come from?

Noetra plans to partner with Nvidia to build a cluster housing roughly 27,500 Rubin GPUs — Nvidia's next-generation architecture after the current Blackwell line, not yet in mass production.
Construction is set for April 2027; the cluster should go live by June 2028.
Until then, model development will run on existing infrastructure from Japanese AI-compute providers.
This means → the team faces a roughly three-year window where it must prove the fundamentals on limited compute before the big cluster arrives.
04

A three-phase model roadmap — is it realistic?

Phase 1 (FY 2026): ship a reasoning foundation model focused on Japanese-language comprehension, logical reasoning and instruction-following — solving the pain point that Japanese-language AI still lags.
Phase 2 (FY 2028): a full multimodal foundation model handling text, image, video and audio.
Phase 3 (FY 2030): "real-world native AI" — models that understand spatial perception and physical properties, designed to operate in the real world (think robotics and autonomous driving).
In plain terms = first, build a smart Japanese chatbot; then teach it to see pictures and video; finally, let it walk into the physical world.
05

What is the real test for Japan's sovereign AI?

The critical milestone is June 2028 — whether the compute cluster ships on time determines if phases two and three can start on schedule.
A 44-member alliance offers breadth, but the bigger the coalition, the higher the coordination cost. The actual R&D team is staffed by engineers seconded from core members, supported by Japan's AIST research institute and Preferred Networks.
This means → funding and political will are in place, but delivering the hardware and a working model within three years is the sole yardstick for whether this path holds.

Content is for reference only, not financial advice.

Japan's Noetra Partners with Sony, SoftBank and 44 Companies to Advance Sovereign AI, Plans to Deploy ~27,500 NVIDIA Rubin GPUs · nashnova