Japan Allocates Up to ¥1 Trillion to Support SoftBank-Led AI Foundation Model
nashnova research
Japan announced up to ¥1 trillion (~$61.6 billion) for a nine-company consortium called Noetra, led by SoftBank, to build a domestic AI foundation model. The bet is Japan's largest-ever AI industrial policy move — and its core wager lands on "physical AI," training models to control robots using real-world factory data.
¥1 trillion — how does the money flow, and who gets it?
First-year allocation is ¥387.3 billion; the rest comes in installments over five years (from FY2026), tied to results. This means → the government is not writing a blank check — it is paying on performance.
The consortium, Noetra, includes SoftBank, Honda, NEC, and Sony, plus the government-affiliated National Institute of Advanced Industrial Science and Technology (AIST).
By mid-July, participating investors will expand to 44 companies — 28 manufacturers (autos, electronics) and 16 non-manufacturers (finance, logistics, telecom). In plain terms = this is not a SoftBank solo project — it pulls in half of Japan's industrial base to share cost and share the model.
What performance is Noetra targeting, and when will it ship?
The plan calls for a working foundation model by 2027, benchmarked against Alibaba's Qwen series and other leading global models.
But sources say initial scale will likely match where Qwen stood before 2025. This means → even if Noetra delivers on time, the model will arrive at least two years behind the global frontier.
By 2031, the model aims for full multimodal capability — processing text, audio, and video simultaneously — plus understanding of spatial and physical properties.
Why bet on "physical AI"?
Noetra's core differentiator is physical AI — AI that directly controls machinery and robots, not just generates text or images.
Japan's logic: its manufacturers hold vast stores of real-world equipment operation and inspection data, a scarce fuel for training robotic AI. This reflects an attempt to offset Japan's late start in language models by leveraging its strongest asset — precision-manufacturing shop-floor data.
Daiwa Research Institute senior researcher Miho Tanabe flags the key challenge: digitizing the tacit knowledge locked in factory workers' hands and instincts. In plain terms = decades of expert craft still live in human heads, not in databases — converting that into AI-trainable data is itself a major hurdle.
How large is the policy package beyond the model?
The government will centrally manage rights to the foundation model, ensuring broad access for supply-chain manufacturers, and will subsidize AI computing infrastructure costs.
A revised AI robotics strategy goes further: ¥10.5 trillion in public-private investment in physical AI through FY2040, with a target of deploying 10 million autonomous AI robots by 2040.
Economy Minister Ryosei Akazawa said: "We will build and strengthen a physical-AI and robotics data foundation that leverages Japan's advantages."
What is the real backdrop behind this move?
The immediate trigger: in mid-June the U.S. government ordered Anthropic to suspend access to its Claude Mythos 5 model for foreign nationals. This means → even allied nations can lose access to frontier AI models overnight.
China, through state-led initiatives, already leads the world in physical-AI patent filings. Without a domestic model, Japan is caught between the two — unable to secure cutting-edge U.S. models, unable to match China's industrialization pace.
In plain terms = the ¥1 trillion buys not just a model but an insurance policy against AI supply disruption. Whether that policy pays out depends on what Noetra actually delivers when the model goes live in 2027.
市场有风险，内容仅供研究参考，不构成投资建议。