Why Japan Can Come Back in Physical AI: Government Policy, the Latest News, the Macro Numbers, and Why TIMEWELL Is Moving Into Physical AI

Hello, this is Ryuta Hamamoto from TIMEWELL.

On July 16, 2026, at a government-hosted event in a Tokyo hotel, Prime Minister Takaichi said in a video message that a project to build a Japanese foundation model for physical AI would be the banner of "Japan's comeback."¹ On the same stage, NVIDIA's Jensen Huang declared that "the next industrial revolution starts here, in Japan." This is a country that everyone agrees has fallen well behind the United States and China in generative AI. So why the strong words? Is it cheerleading, or is there a real path to winning? This long read tries to answer that question using primary sources only.

Let me state my position up front. I think Japan can come back, on conditions. But the conditions are not the old self-image of "we are good at building robots." I will look squarely at the numbers where Japan is losing, then read the substance of the major policies, the news from this summer and autumn, and the qualitative trends behind the numbers. At the end, I will declare that TIMEWELL is moving into physical AI too. If you want to check first whether your own shop-floor data is in a state AI can use, the AI readiness check takes about five minutes.

What happened on the day they called "Japan's comeback"

The July 16 event was officially named by the Ministry of Economy, Trade and Industry (METI) as an "event to communicate Japan's physical AI policy to the world."¹ Its centerpiece was FRONTia, a project to build a Japanese multimodal foundation model that will serve as the development base for AI robots and physical AI. It is run by a consortium of Noetra Inc. and the National Institute of Advanced Industrial Science and Technology (AIST), from fiscal 2026 to 2030. The models it develops will be "weight-available": trained weights are provided to qualifying companies, not released without restriction.²

What struck me in the Prime Minister's message was where she placed the path to winning. She called physical AI "a technology field that is Japan's path to victory, where the critical data accumulated in industrial sites such as manufacturing, logistics, and infrastructure, which are Japan's strengths, becomes the source of competitiveness," and said the government assumes 80 trillion yen of public and private investment through fiscal 2040, including semiconductors.¹ Placing the path to winning in "data accumulated on the shop floor" rather than "robot hardware" is consistent across every government document we will look at.

Minister of Economy, Trade and Industry Akazawa was blunter. "Japan's path to winning lies in being a super-aged society and a disaster-prone country," he said, and he pointed to the decommissioning site at Fukushima Daiichi as a place where only Japan can accumulate data on physical AI operating under extreme conditions.¹ It is an argument that reads weaknesses as advantages. Whether that reading actually holds is the central question of this article.

Huang's talk called Japan's industrial knowledge "national intelligence, a national asset," and continued: "Takumi, kaizen, kanban, genba. Japan has built industrial knowledge into its way of life over generations. That knowledge must not be lost." He then said, "Japan should not outsource its national intelligence. Japan itself needs to build, develop, protect, and deploy 'Japan AI.'"¹ In NVIDIA's announcement the day before, he put it this way: "Japan invented modern manufacturing. Now, it has the opportunity to reinvent it for the age of intelligent industries," calling this "a once-in-a-generation opportunity for Japan."³

Honestly, it is natural for the head of a company that sells semiconductors to praise a customer country. So I decided to check the numbers and the documents rather than the words.

Start with the numbers Japan is losing on

Before talking about a comeback, here are the numbers where Japan is currently behind. Skip this section and the article becomes cheerleading.

According to the International Federation of Robotics (IFR) World Robotics 2025 report, 542,000 industrial robots were installed worldwide in 2024, more than double the figure ten years earlier. China accounted for 295,000 of them, 54 percent of the global total. China's operational stock exceeded 2 million units, the largest of any country. Japan installed 44,500 units, down 4 percent, with an operational stock of 450,500.⁴ China's stock is about 4.5 times Japan's.⁵

The other thing you cannot overlook is the reversal inside China. In 2024, Chinese manufacturers sold more in their home market than foreign suppliers for the first time, taking a 57 percent domestic share, up from about 28 percent a decade earlier.⁴ Japanese and European robot makers supported the Chinese market for a long time, and that structure has already changed. In May 2026, the IFR reported that China's 15th Five-Year Plan places robotics at the heart of its industrial system, with the aim of pivoting AI research "towards physical applications."⁵

By robot density, the number of robots in operation per 10,000 manufacturing employees, South Korea stands at 1,220, Germany at 449, Japan at 446, the United States at 307, and China at 166.⁶ Japan is near the top, but it is far behind Korea and roughly level with Germany. The self-image of a "robot superpower" is accurate on the production side and increasingly outdated on the adoption side.

On top of this comes the gap in generative AI. Yutaka Matsuo, who chairs the government's expert panel on AI strategy, acknowledged on July 16 that while GPU capacity and talent development have advanced, "in terms of competing with the world's most advanced models, the situation remains difficult," and called physical AI "a fight we simply cannot lose."¹ "Cannot lose" is a phrase that only comes from someone who has already lost once.

Then there are humanoids. In 2026, footage of Chinese humanoid robots dancing on Lunar New Year television and running a half marathon in Beijing went around the world. The IFR's assessment is sober: "Despite these impressive public presentations at staged events, the actual capabilities in real-world production scenarios are currently limited to demonstrators or pilot projects."⁵ For Japan, that is a relief and, at the same time, a sign that the window is not long.

Three reasons a comeback is still possible

Having looked at the losing numbers, here are the three reasons I still think a comeback is possible.

The first is shop-floor data. Government documents use this phrase repeatedly, and I read it as "Japan's factories have kept the ground-truth data AI needs to learn from, for longer and in finer detail than anywhere in the world." The Japan Growth Strategy states that "Japan has abundant 'field data' across public and private sectors, in industry, medicine, and logistics," and that by accumulating and training on this high-quality data, Japan will "establish a competitive advantage in physical AI and go out into the world."⁷ The AI Basic Plan (Phase II) goes further and specifies the order: "Vertical AI, which is being adopted first, accumulates the experience and knowledge of each workplace, including tacit knowledge, as data; physical AI, which follows, executes that data and AI judgments in the real world through machines and equipment."⁸ The generative AI race was a contest over web text, data anyone could collect. The physical AI race is a contest over data that can only be collected inside a factory. That is where Japan's accumulation finally becomes an asset.

The second is the manufacturing and components base. According to the IFR, Japan is "the world's predominant robot manufacturing country," representing 38 percent of global robot production.⁹ The government's investment roadmap notes that Japan holds roughly a 70 percent share of the industrial robot market (about 0.8 trillion yen) and strong competitiveness in key components such as motors and reducers.¹⁰ Even if physical AI is a contest of "brains," the brain can only act on the world through a "body." Holding the body buys time to catch up on the brain. The same roadmap, though, notes that Japan's share of the service robot market (about 2.8 trillion yen) is only a little over 10 percent.¹⁰ The strength is concentrated in industrial robots and has not reached the segments where the market is expanding.

The third is demand itself. The Recruit Works Institute's "Future Forecast 2040" projects that Japan's labor supply shortfall will reach 3.415 million workers in 2030 and 11.004 million in 2040.¹¹ That number is the background to Matsuo's remark that physical AI is "an extremely important theme for Japan, which faces labor shortages."¹ The AI Robotics Strategy positions Japan as a "country facing challenges ahead of others" and aims to turn the latent demand from labor shortages into real deployment ahead of the rest of the world.¹² In a country where the reason to buy a robot is business continuity rather than cost reduction, adoption moves at a different speed. Akazawa's line, "it looks like a crisis, but that is exactly the path to winning," is this third reason in a sentence.

Put the three together and a comeback scenario emerges. With shop-floor data at the core, on top of a domestic components and manufacturing base, deploy first against the urgent demand of labor shortages, improve models with the data collected there, and expand to other sectors and overseas. Government documents call this the "cycle." Next, let us see how the government has designed it.

The major policies on one page

Since the start of 2026, physical AI policy documents have come out one after another. Read them in the wrong order and the picture does not form, so here they are in a single table.

Document or program	Date	What it decided
AI Robotics Strategy12	March 26, 2026 (partly revised May 27)	Capture 20 trillion yen and more than a 30 percent global share of a multi-purpose robot market expected to reach about 60 trillion yen by 2040. Deploy 10 million AI robots domestically by 2040 under implementation roadmaps for 18 sectors
AI Basic Plan (Phase II)8	July 14, 2026, Cabinet decision	Names "vertical AI" and "physical AI" implementation, or AX, as Japan's path to winning. Assumes public and private investment through fiscal 2040 of 23.1 trillion yen in vertical AI, 10.5 trillion yen in physical AI, and 68.0 trillion yen in semiconductors
FRONTia2	Launched July 16, 2026	Noetra and AIST develop a Japanese multimodal foundation model for physical AI (fiscal 2026 to 2030), provided on a weight-available basis
Japan Growth Strategy and investment roadmap710	July 21, 2026, Cabinet decision	Assumes cumulative domestic investment of more than 370 trillion yen through fiscal 2040 across 17 strategic fields and 62 products and technologies. Economic ripple effect of physical AI put at 144.4 trillion yen
GENIAC and AIRoA1314	Data released July 31; awards announced September 9, 2026	About 5,000 hours of robot motion data and a foundation model released publicly. Thirteen organizations selected for R&D on multi-purpose robot foundation models

What I find most important in the AI Robotics Strategy is not the targets but the explicit change in approach. The strategy declares a departure from the conventional method of "advancing technology development and demonstration first, then encouraging adoption," and instead supports supply and demand together to "expand demand and supply simultaneously through a cycle with shop-floor data at its core."¹² This is also a reflection on thirty years of Japanese robot policy. The failure was having the technology and not getting it onto the shop floor. This time, the plan is to avoid that by putting it there first.

The near-term tasks are concrete. For actions that can be executed with relatively simple recognition and judgment, "patrolling" and "moving things," the strategy selects eight priority tasks: outdoor and indoor inspection, outdoor and indoor transport, cleaning, receiving and palletizing, handling, and welding, painting, and machining, with a move into areas requiring complex judgment and dexterity from around 2030.¹² Behind the footage of dancing humanoids, what the government targets first is warehouse transport and factory inspection. Unglamorous, and the right order.

The 18 sectors are manufacturing (high-mix, low-volume), shipbuilding, logistics, construction and civil engineering, building, infrastructure maintenance, retail, lodging, food service and food manufacturing, healthcare, nursing care, security, agriculture, forestry, waste processing, disaster response, policing, and defense.¹² The strategy also states that public demand in areas such as disaster response, construction, and defense will be used as "anchor tenancy" to secure continuous demand. The state becomes the first customer, by design.

The AI Basic Plan (Phase II) places this strategy within national AI policy as a whole. The heading "Japan's path to winning is AX through the implementation of vertical AI and physical AI" appears as is, and on physical AI it says concretely: "Develop physical AI foundation models domestically to strengthen the capability to develop robot foundation models. Foster robot OEMs, strengthen the design of critical components such as motors, reducers, sensors, and batteries, and strengthen the supply chain."⁸ The same document says Japan "must avoid excessive dependence on specific countries or companies" and uses the phrase "open AI sovereignty." It points in the same direction as Huang's "should not outsource its national intelligence."

And then the money. The AI Basic Plan cites an estimate from the Council on Economic and Fiscal Policy and the Japan Growth Strategy Council that faster AI adoption will lift Japan's total factor productivity (TFP) growth by 0.2 percentage points.⁸ The investment roadmap puts the economic ripple effect of physical AI investment through fiscal 2040 at 144.4 trillion yen.¹⁰ Press reports say METI's budget request for fiscal 2027 also prioritizes AI, semiconductors, and robots. More than the size of the numbers, what I am watching is that the investment assumptions are written on a horizon "through fiscal 2040." Whether this really escapes the pattern of single-year supplementary budgets can be verified a few years from now.

If you want to read the government's full picture across all 17 fields, I have summarized it in Reading the Japan Growth Strategy 2026 in full.

The latest news, and the qualitative trends it reveals

Policy is the vessel. Whether anything is in it is confirmed by what companies do. Here is the news from July to September 2026 in order.

On July 15, NVIDIA announced its physical AI initiatives in Japan. More than twenty companies, including AIRoA, FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, Kubota, NEC, SoftBank Corp., Sony Group, and Yaskawa Electric, said they intend to join the NVIDIA Cosmos Coalition to build open world models together, and Fujitsu announced it would lead a business study with FANUC, Yaskawa, and Kawasaki on a collaborative control platform.³ Japan's industrial robot makers lining up on the same platform is symbolic. So is the fact that the platform for the brain is provided by an American company.

On July 31, the AI Robot Association (AIRoA) released roughly 5,000 hours of robot motion data covering 68 tasks, collected with GENIAC support, along with a model built by continued pretraining of the open-source VLA model π0.5.¹³ The AI Robotics Strategy states that a beta version of AIRoA's Japanese robot foundation model is planned for open-source release around June 2027.¹² Releasing data to attract developers goes straight at one of the reasons Japan fell behind in generative AI.

On September 3, Hitachi expanded HMAX, its physical AI solution suite for social infrastructure, and announced that a "Physical AI FDE Team" would work alongside customers on site, connecting the extraction of shop-floor knowledge, ontology construction, accumulation in a data platform, AI use, and operations end to end.¹⁵ A major Japanese company putting FDE (Forward Deployed Engineer) at the center of its physical AI strategy was a big moment for us. I wrote about what an FDE is in a separate article; in short, it is "the person who sits next to the shop floor and turns tacit knowledge into a form AI can read." Hitachi's announcement shows that the entrance to physical AI is not buying a robot but organizing shop-floor knowledge.

On September 8, Arm announced Arm Total Design for Physical AI, with more than 80 participating companies including AWS, Hugging Face, NXP, Siemens, and Unitree Robotics. Arm estimates physical AI will be a 200 billion dollar annual compute opportunity in the 2030s and proposed a Robotics Capability Framework to define robot capabilities in a common language, the way SAE levels do for driving automation.¹⁶ Read it as a chip design company coming to define the "language" of robots.

On September 9, GENIAC announced the selection of 13 organizations, including Telexistence, Preferred Networks, KDDI, Mercari, Highlanders, and Hitachi Construction Machinery, for R&D on multi-purpose robot foundation models. Its data ecosystem program lists consortia such as Kawasaki Heavy Industries with FANUC, Yaskawa, and Osaka University; DMG Mori with AIST; and Komatsu with AIST.¹⁴

Regions are moving too. The city of Kitakyushu declared itself a "Physical AI Coexistence City" on May 25, 2026, and adopted a strategy on August 21.¹⁷ A steel and auto town has made physical AI its next industry.

Line up the news and three qualitative trends appear that the numbers do not show.

The first is the shift from tightly coupled to loosely coupled systems. The investment roadmap states that with the arrival of physical AI, the center of gravity in AI competition is moving "from a competition centered on 'scale,' built on web data and compute, to a competition of 'integration and operational capability,' which takes in shop-floor data, optimally integrates AI and robotics, and continues on-site deployment and improvement while ensuring reliability and safety," and expects industry structure to shift from "tightly coupled" systems in which software and hardware are built as one to "loosely coupled" systems that combine the best modules for each use.¹⁰ The strength of Japan's robot industry was tightly coupled craftsmanship. The warning is that this strength, left as is, becomes a weakness.

The second is the shift from a competition of scale to a competition of operations. Generative AI was decided by scale: parameters, compute, and data volume. Physical AI is decided by operations: putting a system on the shop floor, keeping it running without breaking, and learning from failures to improve. When the AI Robotics Strategy says "run the whole process at high speed and scale it up,"¹² this operational capability is what it means. It is exactly what Japanese manufacturing has called kaizen, and it is the knowledge Huang said "must not be lost."

The third is the entrance: making tacit knowledge AI-ready. Hitachi put FDE at the center, GENIAC set up "AI-readiness of manufacturing data" as a program, and the AI Basic Plan specified the order from vertical AI to physical AI. Three separate parties point at the same entrance. Turn shop-floor knowledge into data before buying robots. That is the common answer visible this autumn.

Three pitfalls that would break the comeback scenario

So far I have written the hopeful side. The same primary sources also show how it could break.

The first pitfall is an over-emphasis on humanoids. In the summer of 2026, news about Japanese humanoid robots came almost weekly. But the IFR's analysis invokes the principle that "form follows function," noting that the human body is not suited to certain tasks, that conventional industrial robots with fewer joints are faster and more reliable, and that industrial robots are therefore likely to remain the backbone of high-speed, precision manufacturing.⁵ The government's eight near-term tasks are transport and inspection. If investment flows to the appearance of humanoids and data collection for warehouse transport robots is pushed back, the cycle will not turn. Japan has experience from the 1990s and 2000s of astonishing the world with bipedal walking and failing to build a market. The answer for not repeating that is already written in the government documents.

The second pitfall is where the data goes. If shop-floor data is the path to winning, then where that data accumulates and whose models it makes smarter is the win or loss itself. Japan's major companies joining NVIDIA's Cosmos Coalition is rational for speeding up development. At the same time, as long as the world model platform belongs to a foreign company, the handling of data that flows into it is decided by contract. FRONTia chose the weight-available model, which does not release trained weights without restriction, explicitly because the models "contain data unique to Japan."² The AI Basic Plan's "open AI sovereignty" means not closing off but keeping the ability to choose. Which cloud, which region, holds the data from your factory's sensors, and whose training is it used for? Neglect this check and you hand your one advantage, shop-floor data, to someone else with your own hands. I go into this in Information leakage risks in the physical AI era.

The third pitfall is people. The AI Robotics Strategy acknowledges the shortage of people who span both AI and robotics and lists expanding and upgrading the system integrator (SIer) base as a measure.¹² Deploying 10 million robots by 2040 means, by simple division, more than 700,000 units a year that someone has to install, teach, and keep repairing on the shop floor. The people who deploy and operate robots on site are in far shorter supply than the people who build them. I think this is the biggest bottleneck. And I think this is exactly where companies like ours come in.

TIMEWELL is moving into physical AI too

From here, this is a declaration. TIMEWELL will move into the field of physical AI.

The reason is in the primary sources this article has walked through. The government placed the entrance to physical AI at "vertical AI that accumulates tacit shop-floor knowledge as data." Hitachi gave the role of guarding that entrance a name: FDE. With ZEROCK, our AI agent for manufacturing, we have been reading drawings, organizing the basis for quotes, and turning shop-floor knowledge into a form AI can use. And we have already publicly committed to the FDE style of sitting next to the customer and putting working software in place. In other words, we are already standing at the "first lap" of the physical AI cycle. Going on to the next lap, handing the organized shop-floor knowledge to the side that executes it through machines, is a natural extension.

We have decided three things about how we will go. First, we will not change our entrance. We will not build robot hardware. We enter from the AI-readiness side: organizing manufacturing drawings, quotes, work procedures, and exception handling into a form that robot foundation models and automated equipment can read. That is the place the government calls "vertical AI, which is being adopted first." The data we organize there will be kept in a state that can be handed to any robot and any model.

Second, we will build loosely coupled. We will align our designs from the start with the shift "from tightly coupled to loosely coupled" that the investment roadmap anticipates. Models can be swapped, data can be taken out, equipment can be chosen again. Not locking customers to a specific robot maker or AI vendor is our commitment. It is also an extension of the fact that we have built our own products without depending on any single model.

Third, we will manage where the data goes, inside Japan. ZEROCK runs on servers in Japan. As we move into physical AI, we will keep customers in control of where their shop-floor data resides and whose training it is used for. You could call it implementing "open AI sovereignty" inside a single company's product.

To be honest, we are not a robotics company, and we cannot deliver a working robot today. We will publish the concrete shape of our services as each part is ready. We are declaring this first anyway, because we want to run that first lap together with the companies that read this article and decide to "organize shop-floor knowledge before buying robots." What ZEROCK does on manufacturing shop floors is summarized on the product page.

Summary

We have read, from primary sources only, whether Japan can come back in physical AI. The losing numbers are clear: China accounts for 54 percent of annual industrial robot installations, its operational stock is about 4.5 times Japan's, Japan trails Korea by a wide margin on robot density, and in generative AI the competition with frontier models "remains difficult." The grounds for a comeback were clear too.

• Shop-floor data. The high-quality data accumulated on Japan's industrial sites, unlike web text, can only be collected there
• The manufacturing and components base. Thirty-eight percent of global robot production, roughly a 70 percent share of the industrial robot market, and competitiveness in motors and reducers
• Demand. A projected labor shortfall of 11 million workers by 2040, which turns adoption into a matter of business continuity

The government has set targets of 20 trillion yen and 10 million robots by 2040 in the AI Robotics Strategy, assumed public and private investment through fiscal 2040 of 10.5 trillion yen in physical AI and 68.0 trillion yen in semiconductors in the AI Basic Plan (Phase II), started building a Japanese foundation model through FRONTia, and released 5,000 hours of data through AIRoA. On the corporate side, NVIDIA and Arm are gathering Japanese industry onto their platforms, Hitachi has built its physical AI approach around FDE, and GENIAC has selected 13 organizations. The qualitative trends are three: from tightly coupled to loosely coupled, from scale to integration and operations, and an entrance at making tacit knowledge AI-ready.

The pitfalls are three as well: investment skewing toward the look of humanoids, failing to check by contract where shop-floor data goes, and a shortage of people who deploy and operate robots on site. Whether the comeback happens depends, in my view, on avoiding these three.

TIMEWELL is moving into physical AI. Our entrance is making shop-floor knowledge AI-ready, our method is loose coupling, and the data stays managed in Japan. With those three commitments, we are looking for companies to run the first lap with. If you want to start by taking stock of your drawings, quotes, and work procedures, and find out how readable your shop-floor data is to AI today, let's talk. Buying the robot can wait until after that.

Footnotes

1. Report on the event to communicate Japan's physical AI policy (METI GENIAC Magazine, held July 16, 2026, published August 6). Remarks by Prime Minister Takaichi, Minister Akazawa, Jensen Huang, and Yutaka Matsuo are from METI's Japanese-language record; English renderings are the author's ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. FRONTia project site (METI and NEDO program; consortium of Noetra Inc. and AIST). The weight-available policy and the "contains data unique to Japan" rationale are from the project overview ↩ ↩² ↩³

3. Japan's Robotics and Manufacturing Leaders Build on NVIDIA Cosmos to Advance Physical AI Frontier (NVIDIA, July 15, 2026) ↩ ↩²

4. World Robotics 2025 Report (International Federation of Robotics, September 25, 2025) ↩ ↩²

5. China Makes AI-Powered Robots Core of National Strategy (IFR, May 5, 2026) ↩ ↩² ↩³ ↩⁴

6. US Robot Industry Returns to Double Digit Growth (IFR, June 18, 2026). Robot density figures by country are from this release ↩

7. Japan Growth Strategy (Cabinet Secretariat, Cabinet decision of July 21, 2026, Japanese). The "field data" passage and the assumption of more than 370 trillion yen in cumulative investment across 62 items through fiscal 2040 are from this document ↩ ↩²

8. AI Basic Plan (Phase II) (Cabinet Office, Cabinet decision of July 14, 2026, Japanese). An English translation is also available. The investment figures (23.1, 10.5, and 68.0 trillion yen) and the 0.2-point TFP estimate are from the document's footnote ↩ ↩² ↩³ ↩⁴

9. Japan's Automotive Industry Installs 13,000 Robots (IFR, July 15, 2025). The "38% of global robot production" figure is from IFR President Takayuki Ito ↩

10. Public-Private Investment Roadmap for Key Products and Technologies in the 17 Strategic Fields (Cabinet Secretariat, July 21, 2026, Japanese), section on "Physical AI (especially AI robots)" under AI and semiconductors. Roadmap PDF ↩ ↩² ↩³ ↩⁴ ↩⁵

11. Future Forecast 2040: The Dawn of the Limited-Labor-Supply Society (Recruit Works Institute, March 28, 2023). Labor supply shortfall estimates of 3.415 million in 2030 and 11.004 million in 2040 ↩

12. AI Robotics Strategy (Inter-ministerial Liaison Council on AI Robotics, decided March 26, 2026, partly revised May 27, Japanese). A summary version is on the same site ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

13. Public release of the robot motion dataset and robot foundation model developed under the fiscal 2025 GENIAC program (AI Robot Association, August 3, 2026) ↩ ↩²

14. GENIAC kickoff, interim, and results presentations by selected organizations (GENIAC, September 9, 2026) ↩ ↩²

15. Hitachi expands HMAX to accelerate the social implementation of physical AI (Hitachi, Ltd., September 3, 2026, Japanese) ↩

16. Arm brings the ecosystem together to build and define the next phase of physical AI (Arm, September 8, 2026). The 200 billion dollar figure is Arm's own estimate ↩

17. Physical AI initiatives (City of Kitakyushu, Japanese). Declaration on May 25, 2026; strategy adopted on August 21, 2026 ↩