Japan's Industrial Giants Are Building a 1-Trillion Parameter AI Model — SoftBank, Sony, Honda, and 6 Others Just Formed a Company
SoftBank, Sony, Honda, NEC, and five major Japanese financial institutions have formed a joint venture to build a domestically developed 1-trillion-parameter multimodal AI model by end of decade, explicitly to keep Japanese industrial data inside Japan.
Original sourceJapan's largest corporations have joined forces in what may be the most ambitious national AI project outside the United States and China. SoftBank, Sony, Honda, NEC, MUFG, SMBC, Mizuho, Nippon Steel, and Kobe Steel announced a joint venture today to develop a 1-trillion-parameter multimodal foundation model with a focus on physical AI for manufacturing robotics.
The venture is applying for approximately 1 trillion yen (roughly $6.7 billion) in Japanese government support under the AI economic security initiative. The training infrastructure will be housed in a converted LCD factory in Osaka, repurposed into a high-density compute facility. The target completion date for the flagship model is 2030, with smaller domain-specific models planned earlier.
The stated motivation is data sovereignty: Japanese automotive, manufacturing, and financial data has been flowing to US-based AI systems for years. The coalition argues that physical AI for industrial robotics — the specific application where Japan has the most to gain — requires training on proprietary sensor, control, and production data that companies are unwilling to share with foreign AI providers.
The scale of industrial coordination is unusual even by Japanese corporate standards. Getting Honda and Nippon Steel to co-fund an AI model alongside SoftBank's venture arm represents a level of cross-sector alignment that typically requires years of government negotiation. The urgency is clearly driven by watching Chinese industrial AI advance while Japanese manufacturers have lagged in AI adoption.
Whether a model trained on Japanese manufacturing data can compete with foundation models trained on orders of magnitude more diverse data is the central technical question. The focus on physical AI for robotics — a narrow but high-value domain where Japan has genuine expertise — may be the coalition's smartest design decision.
Panel Takes
The Builder
Developer Perspective
“The physical AI focus is the right call for Japan's comparative advantage. Japanese manufacturers have world-class robotics data that US foundation models have never been trained on. If this coalition can actually execute — a big if for Japanese consortium projects — the resulting model could genuinely outperform GPT-5 on industrial automation tasks.”
The Skeptic
Reality Check
“Japan has announced major technology consortiums before and they rarely ship on time or on spec. The 2030 target is four years away — an eternity in AI development. By the time this model is ready, the US and China will have released multiple generations beyond it. National AI sovereignty projects tend to produce expensive government reports, not competitive foundation models.”
The Futurist
Big Picture
“This is the opening move in a three-way AI race that most Western observers haven't fully processed. US, China, and now a Japan-anchored industrial coalition are each betting on different training data moats. Japan's bet is physical-world manufacturing data. If physical AI is indeed the next wave after language AI, Japan's timing may be exactly right.”