Japan is throwing its industrial weight behind a homegrown challenger to US and Chinese dominance in artificial intelligence.
The government announced this week it will fund Noetra, a new consortium of SoftBank Corp., Sony, NEC and Honda, to build a national AI foundation model in partnership with the National Institute of Advanced Industrial Science and Technology (AIST).
The move reflects mounting concern in Tokyo that Japan’s industrial competitiveness and national security could suffer as AI advances rapidly in the US and China.
Noetra President Hironobu Tanba is a senior executive officer at SoftBank Corp. known for his advocacy of Made-in-Japan AI models. “Being a sovereign AI becomes even more important when AI plays a core role in business,” he said.
“Dependence on overseas LLMs [large language models] carries not only the concern of a company’s confidential information being unintentionally transferred abroad, but also serious risks related to business continuity itself. For example, if an AI is controlling a manufacturer’s production line, and its use is suddenly restricted due to changes in foreign laws or international affairs, the factory’s lines could be shut down,” Tanba said.
On June 30, Noetra and AIST announced that their joint proposal entitled “Research and Development of Physical AI Foundation Technologies for Real-World Native Applications” had been selected under the “Development of Multimodal Foundation Models for AI Robots and Physical AI” program run by the New Energy and Industrial Technology Development Organization (NEDO).
Minister of Economy, Trade and Industry Ryosei Akazawa announced the decision after a Cabinet meeting held the same day. Noetra will receive 387.3 billion yen (US$2.4 billion) in subsidies this fiscal year and about 1 trillion yen ($6.1 billion) over five years.
Noetra does not aim to compete with OpenAI or other developers of advanced general-use AI models.
Instead, Akazawa said, “Japan’s path to success lies in leveraging data accumulated in areas such as health care for the elderly, disaster response, manufacturing sites and the decommissioning of the Fukushima No. 1 nuclear power plant. … By building and expanding data infrastructure for physical AI and robotics — fields in which Japan can capitalize on its strengths — we aim to take the lead globally.”
In addition to its four founding members, Noetra is targeting the participation of — and investment from — other Japanese AI developers and about 40 companies from the manufacturing and non-manufacturing sectors, including autos, electronics, logistics, telecom, IT and finance. The idea is to establish a cross-industry framework extending from R&D to deployment in Japan and overseas.
The foundation model is intended to be multimodal, incorporating data, images, video, audio and physical properties to perform tasks such as recognition and reasoning. To make it work, Noetra will need to integrate critical industrial data in a way that does not compromise the trade secrets of its members.
At the same time, access to AI-related technology should be a powerful incentive for participation, particularly regarding competition with China. As Chinese companies move up the value chain in one industry after another, their Japanese competitors cannot afford to fall behind in physical AI. Even Noetra’s leaders are not trying to go it alone.
If this were all there was to it — another Japanese industry-bureaucracy combine — a great deal of skepticism might be in order. But there are other AI developers in Japan, and two of them, Sakana AI and Preferred Networks, have recently made headlines of their own.
On June 22, Japanese venture company Sakana AI released its Fugu LLM AI service. Fugu is a “multi-agent orchestrator” that combines and coordinates the work of other LLMs to deliver the best possible result depending on the task, “without any single-vendor dependency or the complexity of a traditional multi-agent system.”
Sakana AI claims that Fugu has outperformed Anthropic’s Fable 5, Google’s Gemini 3.1 Pro and OpenAI’s GPT 5.5 in specific applications such as planning, coding and testing, and that it has outperformed Anthropic’s Mythos Preview in answering graduate-level science questions.
Fugu is expensive — as much as five times more expensive than Claude Opus 4.8, according to one assessment — and much slower, but its adaptable specialization and cross-checking reportedly result in fewer hallucinations and other mistakes. It also has value as an alternative to U.S. and Chinese models, and it is a work in progress.
Co-founded in July 2023 by two former AI researchers at Google and an executive with experience in e-commerce and AI, Sakana AI has also developed Japanese-language and image-generation models. It has numerous Japanese and foreign investors, including Nvidia, Khosla Ventures, NEC, Fujitsu and the ITOCHU general trading and investment company.
On June 2, applied computer science company Preferred Networks announced an alliance with Mitsubishi Heavy Industries (MHI) to jointly develop AI technologies for MHI’s national security and social infrastructure product portfolios. MHI is Japan’s largest defense contractor.
Established in 2014 by two computer scientists from the University of Tokyo, Preferred Networks (PFN) is a vertically integrated developer of AI semiconductor devices, computing infrastructure, generative AI and specific AI solutions. It has also created an LLM called PLaMo that is trained on Japanese data and offers a cloud service for AI workloads running on its MN-Core series of processors.
Over the past 12 years, projects in manufacturing (autos, industrial robots, etc.), energy, materials and chemicals, life sciences (pharmaceuticals, drug discovery), logistics, public infrastructure and other sectors have given Preferred Networks extensive domain knowledge that it now brings to MHI. Drone and counter-drone systems are likely to be among the first targets of their collaboration.
Preferred Networks has also begun joint research into physical AI with Toyota Motor’s Frontier Research Center. After the launch of Preferred Networks’ next-generation AI processors in 2027, the two companies plan to conduct tests with robots in applications requiring high-speed inference. Robotics technology developed with Toyota should also inform PFN’s work with MHI.
On June 22, Preferred Networks released a new AI model for corporate users that is optimized for the Japanese language. PLaMo 3.0 Prime reportedly sells for less than half the price of comparable models from OpenAI and other English-based competitors.
The initiatives of Noetra, Sakana AI and Preferred Networks fit with Prime Minister Sanae Takaichi’s promotion of 17 strategic economic sectors, which projects 10.5 trillion yen ($64.5 billion) in public and private investment in physical AI by 2040. In what is usually regarded as a contest between the US and China, Japan intends to be a serious contender.
Follow this writer on X: @ScottFo83517667
