Japan’s Noetra: The Physical AI Consortium Behind the 10 Million Robot Plan
Japan Formally Commissioned a Physical AI Consortium Targeting 10 Million Robots by 2040 — Here Is How It Works
Japan’s Ministry of Economy, Trade and Industry formally selected Noetra — backed by SoftBank, Sony, NEC, and Honda — to build a domestic foundation model for physical AI. The government has attached a target of 10 million AI-equipped robots across 18 sectors by 2040. The programme’s design is more specific than the headline suggests, and more honest about its constraints.
How this was reported: the facts in this article come from BigGo Finance’s full translation of METI’s June 30 announcement, the Asia Times analysis, AI Weekly’s structural breakdown of the funding, Finimize’s stage-gate reporting, and the talap.org programme design summary. Minister Ryosei Akazawa’s quotes are reproduced from Ary News and confirmed across multiple independent Japanese and international sources. Wangdoo has separately covered Sakana AI — Japan’s other significant AI initiative, which Asia Times explicitly connects to the same national strategy as Noetra.
What Happened on June 30
Japan’s Ministry of Economy, Trade and Industry (METI) formally announced on June 30 2026, following a Cabinet meeting, that it had selected two organisations to build a domestic multimodal foundation model for physical AI. The selection was conducted by NEDO — Japan’s New Energy and Industrial Technology Development Organisation — through a public competition that ran from March 24 to April 22, 2026, receiving fifteen applications and selecting two.
The first recipient is Noetra, a purpose-built company founded by SoftBank Corp., NEC, Sony Group, and Honda. Its president is Hironobu Tamba, a senior executive officer at SoftBank Corp. and a long-standing advocate of domestically developed AI models. The second recipient is the National Institute of Advanced Industrial Science and Technology (AIST) — Japan’s principal national research laboratory — which will focus on foundational and long-horizon research in collaboration with Japanese and international research centres.
The programme’s official name is the “Programme for Developing Multimodal Foundation Models for AI Robots and Physical AI.” It runs from fiscal year 2026 through fiscal year 2030.
WION (English-language international news) on Japan’s push into physical AI as a response to its labour shortage — May 21 2026. Independent production, not affiliated with Wangdoo or METI.
What Noetra Is Building and Why Physical AI Specifically
The goal is a multimodal foundation model — one that can process language, images, video, and sensor data together — so that robots can interpret real environments and act within them. The word “physical” is doing specific work in the programme’s framing: Japan is not trying to build a general-purpose language model to compete with ChatGPT or Gemini. Minister Akazawa was explicit about this when confirming the programme: “Noetra does not aim to compete with OpenAI or other developers of advanced general-use AI models.”
The distinction matters because it defines where Japan thinks it has a structural advantage. It does not have the compute scale, the English-language training data, or the research talent concentration that the American frontier labs have built. What it does have — and what Akazawa specifically cited — is proprietary industrial data that no foreign AI lab can replicate through licensing:
The companies involved bring specific data assets that the programme design explicitly requires. METI’s announcement confirms the four founding members and their general roles; the analysis site tech-insider.org adds more specific detail, reporting that Honda brings autonomous-driving data from existing programmes, Sony Group contributes imaging and sensor fusion capacity, NEC contributes from its existing AI model infrastructure, and SoftBank Corp. provides compute and engineering capacity. AIST contributes the national research layer — independent of commercial interests — and ensures the resulting model is usable across the broader Japanese ecosystem rather than locked to any consortium member’s product roadmap.
The 10 Million Robots Target — What It Actually Means
The 10 million figure is the most attention-grabbing element of the June 30 announcement, and it warrants careful reading. The government increased the target from 16 sectors to 18 sectors on June 30, adding food service, food manufacturing, and healthcare to the programme’s scope. These are the three sectors where Japan’s labour shortage is most acute and where the demographic case for robotic substitution is clearest.
Japan holds approximately 70% of the global industrial robotics market by METI’s own data — a position it has held for decades through companies such as Fanuc, Yaskawa, and Kawasaki. The 10 million target is not a bet on building new robotics manufacturing capability from scratch. It is a bet on adding an AI layer — the physical foundation model — to hardware Japan already builds and exports globally. If Noetra delivers a usable model, Japan’s existing robot manufacturers are the immediate deployment base, not a future aspiration.
By 2040, the total physical AI market is estimated at approximately 60 trillion yen ($370 billion) by McKinsey modelling cited in METI’s strategy documents. Japan’s stated target is capturing 30% or more of that market. The AI robotics segment specifically — manufacturing, logistics, and service robots with AI capabilities — is estimated at roughly 20 trillion yen within that total.
The Funding Structure — What the ¥1 Trillion Actually Is
The headline figure requires the same careful reading as the robot target. The ¥1 trillion over five years is a conditional ceiling, not a committed disbursement. The first two fiscal years are locked — ¥387.3 billion has been allocated from GX Economy Transition Bonds for FY2026. After that, the programme goes through annual stage-gate reviews. If Noetra misses its milestones, METI can stop funding without obligation to continue.
What the stage-gate structure means in practice
Japan’s previous large national technology consortia have a mixed record. The Fifth Generation Computer Project, run from 1982 to 1992 with ¥54 billion in investment, produced significant research but failed to translate into commercial products — the demo applications were not built around real use cases with real users, a widely cited reason for its limited industrial impact. The stage-gate structure here is designed to prevent that failure mode: if the model does not produce usable outputs on a defined milestone schedule, the funding does not continue. Citi’s analysis, reported by tech-insider.org, noted that the effective “new money” in the programme is closer to ¥600–700 billion rather than the full ¥1 trillion headline, because some commitments consolidate pre-existing NEDO allocations. That is a meaningful caveat for anyone evaluating the programme’s actual scale.
How Noetra Connects to Japan’s Broader AI Strategy
Noetra is one piece of a larger picture. Prime Minister Sanae Takaichi’s promotion of 17 strategic economic sectors projects a total of ¥10.5 trillion ($64.5 billion) in combined public and private investment in physical AI by 2040. Noetra’s ¥1 trillion represents the government-funded foundation model layer. The private investment target — from the 44 companies Noetra is seeking to bring into the consortium — is expected to constitute the larger share. Finimize confirmed the named early partners include NEC, Honda, Sony, Mitsubishi UFJ Financial Group, Sumitomo Mitsui Financial Group, Mizuho Financial Group, Nippon Steel, and Kobe Steel, spanning manufacturing, finance, and heavy industry.
The Sakana AI model — the Tokyo-based lab co-founded by a Transformer paper co-author and a former Google Brain researcher that Wangdoo covered earlier this month — is explicitly named alongside Noetra in Japan’s sovereign AI strategy, as is Preferred Networks, whose engineers The Japan Times confirms will join the project alongside SoftBank engineers. These are not competing initiatives. They represent different layers of the same national AI stack: Sakana at the research and commercial model layer, Preferred Networks at the industrial robotics application layer, and Noetra at the sovereign physical AI foundation layer backed by state capital.
What Noetra Has and Has Not Committed To
The programme’s publicly confirmed commitments are specific:
- Noetra will release a first physical AI foundation model within fiscal year 2026 — meaning by March 2027
- Annually improved versions will follow, built on accumulated manufacturer data
- The model will be multimodal — combining language, image, video, and sensor inputs
- The resulting model will be available across the broader Japanese ecosystem, not exclusively to the four founding members
- AIST will independently pursue longer-horizon fundamental research in parallel
What is not publicly committed: the parameter count of the model, the specific compute infrastructure beyond SoftBank’s existing capacity, the exact metrics for the stage-gate milestones, or the list of companies beyond the four founding members that have confirmed participation. The 44-company target is a goal, not a signed roster.
My Take — Mr Wangdoo
The most interesting sentence in the June 30 announcement is the one about Fukushima. The decommissioning of the Fukushima No. 1 nuclear plant is a decades-long project that will require autonomous robotics operating in environments no human can safely enter. When the minister cites it alongside elder care and factory automation as Japan’s proprietary data advantage, he is making a specific claim: Japan has real-world physical environments with real-world consequences that no amount of simulation or synthetic data generation can replicate. That is a genuine competitive moat for physical AI that the American labs are not positioned to challenge directly.
Whether the consortium can execute on that advantage is a different question. Japan’s national technology consortia have produced genuine breakthroughs — the industrial robot industry itself emerged partly from government-backed research in the 1970s — and spectacular failures. The stage-gate structure, the competitive selection process through NEDO, and the explicit framing of Noetra as a specialist tool rather than a general AI competitor are features that differ from the design of Japan’s previous national tech consortia. Whether those design differences are sufficient to avoid the same failure modes is a question the first milestone will begin to answer. The first milestone — a usable model by March 2027 — will be the clearest early signal of whether the design translates into execution.
Frequently Asked Questions
What is physical AI?
Physical AI refers to AI systems designed to perceive and act in the physical world rather than operating purely in digital environments. A language model that answers questions is not physical AI. A robot that uses cameras and sensors to understand its environment and manipulate objects within it is physical AI. The key technical challenge is that the real world is messier than text — lighting changes, objects are in unexpected positions, and actions have consequences that cannot be undone. Noetra’s goal is a foundation model that combines language, image, video, and sensor inputs specifically to handle these real-world conditions rather than a general-purpose text model.
Who is Noetra and why were they selected?
Noetra is a purpose-built company — formerly called Japan AI Foundation Model Development — created specifically for this programme. Its four core founders are SoftBank Corp., NEC, Sony Group, and Honda. NEDO selected it through a public competition that received fifteen applications. Noetra’s competitive advantage in the selection was the combination of its members’ proprietary physical data — Honda’s robotics and automotive data, Sony’s imaging and sensor data, NEC’s AI research infrastructure — which no other applicant could match. Its president, Hironobu Tamba, is a SoftBank Corp. senior executive officer who has publicly advocated for domestically built Japanese AI models.
Is ¥1 trillion guaranteed or conditional?
Conditional. The first-year allocation of ¥387.3 billion is locked, funded through GX Economy Transition Bonds. After the first two years, the programme goes through annual stage-gate reviews. If Noetra fails to hit defined milestones, METI is not obligated to continue funding. The ¥1 trillion figure is the maximum envelope over five years, not a committed disbursement. Citi’s analysis suggests the effective new money in the programme is closer to ¥600–700 billion once pre-existing NEDO commitments are netted out.
How does Noetra relate to Sakana AI and Preferred Networks?
They operate at different layers of Japan’s AI ecosystem rather than competing. Sakana AI is a private research company building collective intelligence models with commercial applications across finance, defence, and research automation. Preferred Networks is Japan’s leading industrial AI company, specialising in robotics and automotive AI. Both have personnel seconded into the Noetra consortium. Japan’s government explicitly frames all three as complementary components of a national AI strategy rather than alternatives. The Noetra programme funds the sovereign physical foundation model layer; Sakana and Preferred Networks operate at the commercial application layer built on top of that infrastructure.
What is Japan’s timeline for the 10 million robot target?
The target is 10 million AI-equipped robots deployed across 18 sectors by 2040 — fourteen years from the programme’s start. The 18 sectors now include food service, food manufacturing, and healthcare, expanded from the previous 16 sectors on June 30. Japan’s existing industrial robotics sector — where Japanese manufacturers hold approximately 70% of the global market — provides the deployment base. The Noetra foundation model is intended to be the AI layer added to robots that Japan already builds. The 10 million target is a binding government deployment target, not a market forecast, meaning it reflects planned public-sector procurement as well as private industrial uptake.
Sources
- Japan Launches Domestic Physical AI Initiative: Noetra Secures ¥387.3 Billion in First Year — BigGo Finance (full translation of METI June 30 2026 announcement — primary source)
- Japan rallies tech-giant alliance to build sovereign AI — Asia Times, July 2026 (Noetra president quote, Sakana/Preferred Networks connections confirmed)
- Japan pledges up to $6.16B to SoftBank-led AI consortium — AI Weekly (GX Economy Transition Bonds, structural analysis, SoftBank OpenAI stake context)
- Japan Picks SoftBank-Backed Noetra To Build A National AI Model — Finimize (stage-gate structure, milestone mechanism, market implications)
- Physical AI as a national bet: what lies behind Japan’s new programme — talap.org (NEDO competition details, 15 applications, April 22 selection date, programme structure)
- METI to Provide 1 Trillion Yen to SoftBank Alliance for Domestic AI Development — News On Japan (Japanese primary media confirmation of Noetra formation and AIST partnership)
- Japan announces aid for domestic AI development project — The Japan Times, June 30 2026 (primary English-language Japanese news source — confirms Preferred Networks engineers joining the project)
