Japan Backs SoftBank’s Noetra for National AI Model Push

Japan’s industry ministry said on June 30 that it had picked Noetra Corp., a SoftBank-backed venture, to lead a national project that aims to build a “physical artificial intelligence” model for robots and machines, with public support worth up to 1 trillion yen (roughly $6.2 billion) over five years. Industry Minister Ryosei Akazawa announced the selection after a Cabinet meeting, and the New Energy and Industrial Technology Development Organization, known as NEDO, is set to formalize the contract by late August.

Noetra, formerly known as Nihon AI Kiban Moderu Kaihatsu, or Japan AI foundation model development, was put together earlier this year to apply for that exact NEDO call. It beat out a small field because it was already lined up with the four companies that built it, plus a research partner in the government.

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. We will compete in areas where we have big data. 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.

Akazawa said that at a news conference on June 30, framing the project as a deliberate bet on Japanese-owned industrial data rather than a chase of general-purpose AI led by U.S. firms such as OpenAI. The distinction matters: Japan’s plan is to compete on the ground where robots actually work, not in the general-purpose chatbot race where the United States and China already hold a lead.

The Roster That Already Has Skin in the Game

Noetra is built on a tight founding circle. SoftBank Corp., NEC Corp., Sony Group Corp. and Honda Motor Co. set up the company in April and each holds more than 10 percent, with the four founders keeping a majority stake between them. A SoftBank executive serves as president. The lineup is not accidental: SoftBank and NEC bring proprietary generative AI work, Honda plans to deploy the model first in autonomous vehicles, and Sony’s role covers robotics and gaming hardware.

Behind those four sits a layer of capital from Japan’s industrial and financial core. The three megabanks, MUFG Bank, Sumitomo Mitsui Banking Corp. and Mizuho Bank, are investors, alongside Nippon Steel Corp. and Kobe Steel Ltd. Fujitsu Ltd. and Rakuten Group Inc. are weighing investments of their own, and the count of participating firms is expected to exceed 40. Preferred Networks Inc., the deep-learning specialist with existing ties to Toyota, is set to collaborate on the technical side.

What separates this consortium from the average industry partnership is how the roles are split. SoftBank and NEC carry the foundation model build. Honda and Sony take the model into their own products and services. The megabanks and steelmakers get access to the data flowing through the project as investors, without necessarily building the model themselves. Each group gets a different kind of upside, and the structure was designed that way.

Why Physical AI, Not General-Purpose Models

Physical AI is the term the industry has settled on for models that let robots and machines operate autonomously in the real world, as opposed to generative AI, which produces text and images in digital space. The Asahi Shimbun defined the split cleanly: physical AI “enables robots and machines to operate autonomously in real-world settings.” That is the workload Japan wants to own, because the data needed to train such models is largely produced in Japanese factories, on Japanese plant floors, by Japanese operators.

The starting position is unusually strong. According to METI data cited in industry coverage, Japanese manufacturers account for about 70% of the global industrial robotics market, and five of the world’s ten largest industrial robot manufacturers are Japanese. The government wants to convert that installed base into training data. METI announced in March 2026 that Japan aims to capture a 30 percent share of the global physical AI market by 2040.

The choice to specialize is also a recognition of where Japan is not winning. The United States and China lead general-purpose AI development, and Japan has publicly said it does not want to chase them on that ground. Akazawa’s data-rich sectors, including health care for the elderly, disaster response, manufacturing, and the decommissioning of the Fukushima No. 1 nuclear power plant, are exactly the kind of workloads where physical, not language, capability is what matters.

Noetra and AIST, the National Institute of Advanced Industrial Science and Technology, plan to release a foundation model “as early as this fiscal year,” with an improved version every fiscal year after that. That release cadence is one of the few concrete delivery dates attached to the project.

The 387.3 Billion Yen That’s Actually Wired Up

The headline number is 1 trillion yen. The number sitting in METI’s budget outline for fiscal 2026 is 387.3 billion yen, and the gap between the two is where the real story lives. The five-year window runs from fiscal 2026 to fiscal 2030, and the contracts on the table now cover the first stretch. Additional aid after fiscal 2026 depends on results, year by year.

Jiji Press, the wire service that broke the contract details, reported that the ministry “plans to provide additional aid to the project after the first year.” Japan’s plan to release a multimodal AI model every fiscal year is the pace the contracts are tied to, which makes the foundation model release schedule and the year-end review moments for Noetra the same thing.

The Asahi Shimbun pointed out the obvious constraint: “government support will total 1 trillion yen over five years… that sum falls short of the AI investment of a single major U.S. tech company.” One trillion yen over five years is a serious commitment in Japan, and a fraction of what a single U.S. frontier lab is spending in 2026 alone.

• Total public support over five years: 1 trillion yen (~$6.2 billion)
• Fiscal 2026 outlay: 387.3 billion yen
• Project window: fiscal 2026 to fiscal 2030
• Japanese share of global industrial robotics: about 70%
• Government’s stated physical AI market target by 2040: 30% global share

Why a Few Names Capture the Real Upside

When a government commits to a single integration layer for pilots, data formats, and safety checks, suppliers and customers tend to build around it, because that is where the contracts and the compatibility show up. Noetra is being positioned as that layer. The named adopters and partners stand to capture more direct revenue from it than the broader wave of Japanese “AI-linked” stocks.

One executive at a company that plans to invest in Noetra told the Asahi Shimbun that access through investment is the practical payoff. “This is something that will effectively conclude in five years, but the ability to gain access to information through investment is significant,” the executive said, on the condition that the company was not named. That is the trade the consortium is selling: data and visibility in exchange for capital and adoption.

Market reaction to the consortium’s April launch was not a clean rally. SoftBank shares closed down approximately 2 percent on the day of the joint venture announcement, with TradingKey attributing the move to “fragmented requirements of various companies” and the difficulty of fitting a single foundation model to Honda’s autonomous driving work and Sony’s robotics focus at the same time. The same analyst flagged the inconsistent data standards across Japanese manufacturing as a near-term integration problem.

The Open Questions That Could Reshape the Field

The Asahi Shimbun asked the question Japanese industry watchers have been asking since the consortium was announced: “whether the companies involved can deepen collaboration beyond traditional boundaries without repeating the failures of past ‘all-Japan’ alliances in sectors such as semiconductors and liquid crystal displays.” That history is the closest analogue, and it ended with the government writing down most of its investment.

This is something that will effectively conclude in five years, but the ability to gain access to information through investment is significant.

An executive at a company that plans to invest in Noetra said that, on the same condition of anonymity. The five-year window is the test: either the consortium produces a working national model that Japanese industry actually builds around, or it joins the list of grand Japanese technology alliances that did not.

Another open question is who actually owns the model. The NEDO application framework positions the foundation model as public infrastructure to be opened up to Japanese industry at large, not a corporate asset held by the investors. That structure gives Honda and Sony early-mover know-how, but also opens the door to competitors getting the same model once it ships. For investors expecting exclusive competitive advantage, there is a gap between that expectation and the way the contracts are written.

TradingKey argued that the Q3 2026 METI budget approval will be the first pass-fail moment, “directly determining whether subsidies can be implemented in a timely manner.” Each year after that carries the same gate, with the consortium either clearing it and getting the next tranche or losing the runway.

How SoftBank Squares Two Bets at Once

SoftBank’s position in Noetra looks contradictory only if you read it from the outside. SoftBank led OpenAI’s $40 billion funding round in 2025, and is now the standard-bearer for a domestic AI project positioned as independent of the U.S. AI ecosystem. The contradiction dissolves once the two efforts are placed on different layers. Noetra is a sovereign capability for workloads where data confidentiality and physical control make foreign dependency a liability; OpenAI is the broader frontier play.

The numbers underline the gap. TradingKey reported that SoftBank holds approximately a 13 percent stake in OpenAI, with cumulative investments exceeding $60 billion. The NEDO commission to Noetra is up to 387.3 billion yen per year, and the entire five-year sum is roughly one-sixth of SoftBank’s OpenAI exposure. Noetra is not where the big dollars are. It is where the industrial data and the government contracts are.

For markets, the practical question is what the trillion-yen headline actually buys. The plan is a series of year-by-year approval gates, with the consortium’s foundation model release schedule tied to those gates. The named partners, including SoftBank, NEC, Honda, Sony, the three megabanks, Nippon Steel and Kobe Steel, capture visibility and data flow as long as they stay on the schedule. Everyone else in Japanese AI is watching the same schedule, with no contract attached.

SoftBank’s actions are a hedge. If OpenAI keeps compounding, the domestic bet paid for itself with public money. If physical AI becomes the standard layer Japanese industry builds around, Noetra holds the position OpenAI’s general-purpose model does not. The bet covers both outcomes without forcing a choice between them.

What changes the calculus is the annual aid decision after fiscal 2026, and that decision will turn on whether Noetra ships a working multimodal model this fiscal year and keeps improving it year over year. The trillion-yen headline is the ceiling. The 387.3 billion yen in fiscal 2026 is the floor. Everything between them is decided gate by gate.

Frequently Asked Questions

How much is Japan actually committing to the Noetra project?

Japan’s industry ministry said it would provide 387.3 billion yen in fiscal 2026, with the total potentially reaching 1 trillion yen, roughly $6.2 billion, over five years if each annual review clears. The 387.3 billion yen figure is what is wired up in the current budget outline. The remaining years depend on milestone reviews.

Which companies are part of the Noetra consortium?

SoftBank, NEC, Sony Group and Honda founded the company and each hold more than 10 percent, with the four together holding a majority stake. The three megabanks (MUFG, Sumitomo Mitsui Banking, Mizuho Bank), Nippon Steel and Kobe Steel are investors. Fujitsu and Rakuten are considering investments, and the count of participating firms is expected to exceed 40.

What is physical AI and why is Japan focusing on it?

Physical AI refers to models that let robots and machines operate autonomously in real-world settings, as distinct from generative AI that produces text and images. Japan’s bet is that its installed base of factory robotics and its accumulated process data give it an advantage in physical AI that it does not have in general-purpose AI, where the United States and China lead.

When will Noetra release its first foundation model?

Noetra and AIST plan to release a foundation model as early as fiscal 2026, with an improved version every fiscal year after that, using data obtained from manufacturers and other companies.

How does Japan’s AI spending compare to US tech companies?

The Asahi Shimbun reported that the 1 trillion yen commitment over five years “falls short of the AI investment of a single major U.S. tech company,” underscoring that Japan’s public outlay is large by domestic standards and modest relative to private U.S. frontier-lab budgets.