Four of Japan's most recognizable technology companies — SoftBank, NEC, Honda, and Sony Group — have formed a new joint venture to develop a domestic AI foundation model at trillion-parameter scale, in a clear bid to close the capability gap with US and Chinese frontier labs.
The new company, reported in Japanese press on April 12, gathers roughly 100 AI engineers from the founding shareholders. An executive who previously led AI development operations at SoftBank will serve as president. Beyond the four anchor partners, additional minority investors include Nippon Steel, Kobe Steel, Mitsubishi UFJ Bank, Sumitomo Mitsui Banking Corporation, and Mizuho Bank — a coalition that spans heavy industry, electronics, automotive, and finance.
A trillion-parameter target with state backing
The venture's first technical milestone is to build a foundation model with about 1 trillion parameters, a scale aligned with the largest publicly disclosed frontier models. To get there, Japan's Ministry of Economy, Trade and Industry plans to provide roughly 1 trillion yen in support over the next five years for AI development through programs tied to the initiative.
SoftBank and NEC are expected to lead foundation model R&D, while Sony and Honda will focus on deploying the resulting models into automotive, robotics, gaming, and semiconductor products. That division of labor reflects a deliberate choice: rather than chase chatbots, the consortium is positioning Japan around what it sees as a structural advantage.
Betting on physical AI
Local coverage frames the strategy bluntly. "The U.S. and China lead in AI development, but Japan is believed to have an advantage in physical AI," the Yomiuri Shimbun reported, capturing the consortium's pitch that the country's manufacturing base, industrial robotics, and automotive depth are natural distribution channels for embodied models.
That framing matters because it signals where the project will measure success. Sony's gaming, imaging, and semiconductor businesses, Honda's mobility and humanoid programs, and NEC's enterprise systems each provide real-world deployment surfaces that pure-play model labs lack. The consortium's stated objective is to translate that advantage into leadership in physical AI by 2030.
Sovereign AI moves from talking point to balance sheet
The alliance lands in a year when sovereign AI has moved from policy paper to capital allocation. Governments from the UK and France to South Korea and India have announced national model programs, but few have assembled an industrial consortium of this composition: a telco-investor, a flagship electronics maker, a global automaker, and a systems integrator, all funded in part by the country's largest banks.
Whether trillion-parameter scale alone is enough to close the gap with OpenAI, Anthropic, Google, or Chinese labs that are training on far larger compute footprints is an open question. But the structure of the bet is striking. Japan is not trying to out-spend US hyperscalers on raw compute — it is trying to out-deploy them in the physical world, with a single consortium controlling the model, the hardware, and the end-customer touchpoints at once.
For enterprise buyers in Asia, the practical question over the next twelve months is whether this venture ships a usable model — and on what licensing terms — before the gap with frontier US labs widens further.
