Japan's $100 Billion Market Opened Up by General-Purpose Robots

The final major transformation of the Japanese robotics industry, as pointed out by McKinsey

August 15, 2026

Japan was once called a "robot superpower." If you visit a factory, you will see robots from FANUC, Yaskawa Electric, Kawasaki Heavy Industries, Mitsubishi Electric, Nachi-Fujikoshi, Denso, and others in motion. If you include speed reducers, motors, sensors, bearings, and precision machined parts, Japanese companies have built a massive supply chain that supports the global robotics industry. In the 1970s, Waseda University developed "WABOT-1," one of the world's first full-scale humanoid robots, and since then, through Honda's ASIMO, AIST's HRP series, and Toyota's Human Support Robot, Japan has long shown the world the future of "machines that move like humans." However, in 2026, if you list the companies at the center of the global humanoid and general-purpose robot race, the landscape has changed significantly. In China, numerous companies including Unitree, AgiBot, and UBTECH have entered the mass production race, while in the United States, Tesla, Figure AI, Agility Robotics, and others are attempting to form a new industry that fuses AI and robotics. And in that competition, the names of Japanese companies are almost nowhere to be found.

This is not a story about "Japanese robot technology disappearing." In fact, the problem is the opposite.

Japan still possesses world-class hardware technology. Nevertheless, the axis of competition in the robotics industry itself has shifted from hardware to AI, software, data, and mass deployment, and the ways of winning that Japan was good at are beginning to no longer apply.

This issue was addressed head-on in McKinsey & Company's 2026 analysis, "Japan's $100 billion opportunity in general-purpose robotics". The future McKinsey envisions is extremely large. The general-purpose robot market, which was less than $1 billion in 2025, will grow to approximately $370 billion by 2040. And the Ministry of Economy, Trade and Industry has set a goal for Japanese companies to capture 30% of the global general-purpose robot market by 2040, or

approximately $111 billion

If converted at 160 yen to the dollar, that is approximately 17.76 trillion yen. It is literally a "$100 billion market." This is a part that is easily misread, so I will clarify it first. The market forecast of $370 billion is McKinsey's estimate, but the figure of 30%—$111 billion—is not a share promised to Japan by McKinsey. It is a policy goal set by the Ministry of Economy, Trade and Industry that McKinsey is citing in its report. However, more important than this figure is the industrial structural transformation that McKinsey is demanding of Japanese companies. From "companies that make and sell robots" to

"companies that deploy robots in large numbers to the field, collect data, train AI, and continuously improve performance through software."

The same applies to parts manufacturers. From companies that sell motors, speed reducers, and sensors as individual items to

"companies that supply integrated modules such as smart joints, safety systems, and perception/control stacks."

Furthermore, instead of succeeding in the Japanese market and then going overseas, to

"Global Native" companies that design products with the global market in mind from the start.

What McKinsey is demanding is not just an improvement in robot products. It is replacing the "OS" of the Japanese robotics industry itself.

To the reader—Can "Robot Superpower Japan" remain a protagonist for the next 15 years?

In 2040, the global general-purpose robot market will reach approximately $370 billion. If Japanese companies can capture 30%, that scale will be approximately $111 billion—exceeding $100 billion.

However, this is not a future promised to Japan. Japan, which once led the world in robot density and drove global industrial robot production, is now facing a new "AI x Robot" competition led by China and the United States. What determines the outcome is not just the performance of motors, speed reducers, and robot arms. AI, software, field data, mass production capability, deployment speed, and the data flywheel where deployed robots continue to learn are becoming new competitive advantages. Even so, Japan has the robotics industry of FANUC, Yaskawa Electric, Kawasaki Heavy Industries, Mitsubishi Electric, etc., precision parts, sensors, and control technology, world-class manufacturing sites, and the power of "suriawase" (fine-tuning) cultivated over many years. So, can Japan convert these assets into competitiveness in the era of physical AI? In this article, focusing on McKinsey's "Japan's $100 billion opportunity in general-purpose robotics," while layering in the 2026 Japanese government "AI Robotics Strategy," NEDO's foundation model development, and IFR robot statistics, we will decipher whether Japan will end as a "past robot superpower" or lead the global robot industry again in 2040.

Table of Contents

Part 1: What is happening in the robotics industry?

Chapter 1: What McKinsey's "Japan's $100 Billion Market" means \ Chapter 2: What are general-purpose robots? — The shift to physical AI \ Chapter 3: The 2040 $370 billion market and the value shift in the robotics industry

Part 2: Why Japan Became a Robot Superpower

Chapter 4: Why Did Japan Become a Robot Superpower? From WABOT to ASIMO\Chapter 5: "Suriawase" (Integration) and the Components Industry: A Massive Asset Remaining in Japan

Part 3: Japan's Paradox

Chapter 6: The Paradox of Robot Superpower Japan: Humanoids and Patent Competition\Chapter 7: Robot Density: Why "Japan 4th" and "Japan 5th" Coexist\Chapter 8: The Essence of Declining Production Share: What Japan Lacks is Deployment Speed, Not Technology

Part 4: The US, China, and Japan: Three Robot Strategies

Chapter 9: China's Hardware-first and the US's AI-centric Approach\Chapter 10: Japan's Third Way: Reliable Physical AI and a New Form of "Suriawase"

Part 5: The Major Transformation Forced Upon Robot Manufacturers

Chapter 11: Robot Manufacturers Moving from "Manufacturing" to "Data Companies"\Chapter 12: Robot OS, OTA, and Fleet Management: Software Foundations That Grow After Shipment\Chapter 13: Redefining RaaS and SI: From One-off Sales to Continuous Services

Part 6: A Massive Opportunity for Japanese Component Manufacturers

Chapter 14: From Components to Modules: The Winning Strategy of Smart Joints\Chapter 15: Safety, Action Stack, and Edge AI: Making Japanese Hardware Intelligent

Part 7: The Japanese Government's AI Robotics Strategy

Chapter 16: Why a National Strategy Became Necessary in 2026: The Challenge of a 20 Trillion Yen Market\Chapter 17: 10 Million Units Domestically, 18 Fields: Making Social Implementation a National Project\Chapter 18: Domestic Foundation Models and Field Data: Building a Japanese-style Data Flywheel

Part 8: Markets Where Japan Can Win

Chapter 19: Manufacturing, Logistics, and Construction: The Three Fields Where a Massive Market Will Emerge First\Chapter 20: Nursing and Medical Care: What Role Will Robots Play in a Society with Labor Shortages?\Chapter 21: Agriculture, Food, Disaster Relief, and Defense: Turning Japan's Unique Real-world Environments into Competitive Advantages

Part 9: Management Reforms Necessary for Japanese Companies

Chapter 22: Management Reform: Mass Deployment, Business Sandboxes, M&A, and Talent Acquisition\Chapter 23: Becoming a Global Native Company: Assuming the Global Market from the Start\Chapter 24: Can the Government Become the "First Customer"? Government Demand and Regulatory Sandboxes

Part 10: Can Japan Become a Robot Superpower Again?

Chapter 25: Conditions for Realizing a $100 Billion Market: Japan's True Winning Strategy\Chapter 26: Structural Changes in the Robot Industry in the 2030s\Chapter 27: Conclusion: It Is "Learning Speed" That Will Determine the Next Robot Superpower

Chapter 1: What McKinsey's "Japan's $100 Billion Market" Signifies

The original McKinsey report, "Japan's $100 billion opportunity in general-purpose robotics," was published in English on May 26, 2026. About two and a half months later, on August 4, the Japanese version, "The $100 Billion Market for Japan Opened Up by General-Purpose Robots," was released by the McKinsey Japan office, with introductions in Japanese media following even later. The Japanese version is also an official McKinsey translation, not secondary information. However, if there are discrepancies in figures or expressions, the English original takes precedence. In fact, as with the number of patents mentioned later, there are places in the Japanese version where qualifiers present in the English original have been dropped. It was co-authored by Ani Kelkar, Christian Jansen, Erik Sparre, Hiroshi Odawara, and Kenji Nonaka, with Julien Descamps and Takuya Hata joining, and is summarized as the view of McKinsey's Industrials Practice. What I would like to focus on here is the authors' affiliations. Ani Kelkar is in Boston, Christian Jansen is in Hamburg, and Julien Descamps is in New Jersey. And Erik Sparre is a partner, Hiroshi Odawara and Kenji Nonaka are senior partners, and Takuya Hata is an associate partner, all of whom are affiliated with the Tokyo office.

Four out of seven members are based in Tokyo. This is not a report discussing Japan from the outside, but a proposal written by people who are looking at the inside of Japan's manufacturing industry. The PDF version is 14 pages long. What is important is that it is not merely a market size forecast, but a discussion of Japan's OEMs, parts manufacturers, software, data, capital allocation, human resources, and government cooperation as a single industrial transformation. Furthermore, the first important point in understanding the McKinsey discussion is that this report does not claim that "Japan's robot industry is over." It is quite the opposite. Japan still possesses a massive industrial foundation of precision machinery, motors, speed reducers, sensors, bearings, machine tools, control devices, production technology, quality control, and supply chains. McKinsey estimates that Japan's supply chain alone can supply almost all the parts necessary for humanoids. The question is what kind of "product" to make of them. Until now, Japanese companies have captured the global market by making high-performance motors, high-precision speed reducers, high-quality sensors, and robot arms that are resistant to failure. However, in the era of physical AI, the competition will not be decided solely by "which company's motor is the most precise."

After a robot is deployed to the field, how fast does it learn? How many hours does it take to master a new task? Can it predict failures in advance? Can it manage robots from multiple manufacturers collectively? Can it add capabilities just through software updates? Can it be managed when the number increases to 100, 1,000, or 100,000 units? This is where the new competitive arena lies. In other words, robots are...

changing from "finished machines" to "continuously learning computing platforms."

Chapter 2: What are General-Purpose Robots? — The Shift to Physical AI

When people hear "general-purpose robot," many imagine a humanoid. However, the two are not synonymous. Humanoid is a term that mainly describes body shape, meaning a robot with a form close to a human, such as two legs, two arms, a torso, and a head. On the other hand, the essence of a General-purpose Robot lies in its "capability," not its "form." Conventional industrial robots have been designed for the purpose of repeating specific processes such as welding, painting, palletizing, and assembly with high precision. In contrast, general-purpose robots use AI to perceive the environment, judge situations, and learn new tasks. Today, they carry boxes. Tomorrow, they arrange products on shelves. The next day, they install parts. In another place, they clean. It is important that their use can be changed through software and learning in this way. Just as smartphones integrated phones, cameras, maps, music players, and payment terminals into one device, general-purpose robots are approaching "general-purpose computers for physical work" that can switch between multiple physical tasks via software.

The generative AI revolution began inside computers. Generative AI such as ChatGPT has become able to handle not only text but also images, audio, video, and programs. However, the activity domain of AI was basically the digital world. What is beginning next is...

Physical AI

It perceives the real world through cameras, LiDAR, tactile sensors, force sensors, audio, etc., and moves machines based on those results. In physical AI, the output of the AI does not end with text or images. Physical actions themselves, such as "walking," "holding," "pushing," "turning," "carrying," and "assembling," become the output. Here, the robot is not just an automatic machine, but...

a body through which AI acts on the real world.

The Ministry of Economy, Trade and Industry also positions the advancement of physical AI, which integrates image, audio, video, and sensor information to understand the real world and executes physical tasks based on that understanding, as an important structural change in its 2026 AI Robotics Strategy.

Chapter 3: The $370 Billion Market in 2040 and the Value Shift in the Robot Industry

The most eye-catching figure in McKinsey's forecast is...

$370 billion.

As of 2025, the global general-purpose robot market is less than $1 billion. However, it is projected to grow to approximately $370 billion by 2040. The average annual growth rate is about 70%. In contrast, the conventional industrial robot market is expected to grow at an annual rate of only about 3%, reaching approximately $80 billion by 2040. In other words, in 2040, the composition will be approximately $80 billion for conventional robots and $370 billion for general-purpose robots, with general-purpose robots accounting for the majority of the combined market of approximately $450 billion. This is the essence of the industrial structural transformation. Even if Japanese companies maintain a high share in the current industrial robot market, if the growth market itself moves elsewhere, their relative presence in the global robot industry will decline.

However, one must not misunderstand here. Industrial robots will not be completely replaced by humanoids. If you are welding the same spot hundreds of thousands of times in an automobile factory, a dedicated industrial robot is more likely to be faster, cheaper, more accurate, safer, and more durable. Versatility comes with a cost. Therefore, there is rationality in the forecast that a conventional robot market of about $80 billion will remain even in 2040. What changes is the "range of work that can be automated." Conventional robots required the environment to be adapted to the robot. With physical AI, conversely...

the robot adapts to the human environment.

This creates the possibility for robots to enter huge labor markets that could not be automated in the past.

Due to the physical AI revolution, the center of value in the robot industry will change significantly. Previously, the flow was: Hardware -> Control Device -> System Integration -> Delivery. From now on, it will be a circular industry: AI Foundation Model -> Robot Foundation Model -> Simulation -> Learning Data -> Perception/Control -> Hardware -> Fleet Management -> OTA -> Maintenance -> Field Data -> Re-learning. The field data acquired at the end returns to the AI again. In other words, the industrial structure changes from a "straight line" to a "loop." This is extremely important.

Chapter 4: Why Did Japan Become a Robot Superpower? — From WABOT to ASIMO

The strength of Japan's robot industry is no coincidence. Since the period of high economic growth, Japan has had a concentration of huge manufacturing industries such as automobiles, electrical machinery, electronic components, machine tools, and semiconductors. Factories needed automation, and robot manufacturers grew in response to that demand. User companies and robot companies repeated improvements in close proximity. As a result, FANUC, Yaskawa Electric, Kawasaki Heavy Industries, Mitsubishi Electric, Denso, Nachi-Fujikoshi, and others grew into world-class robot manufacturers. This can also be interpreted as a huge "domestic data flywheel." However, what was circulating at that time was not AI learning data like it is today. Field know-how, quality information, customer requirements, failure information, and knowledge of process improvements went back and forth between user companies and manufacturers, which led to product improvements. Japanese companies continued to improve their hardware by utilizing this information circulation.

Japan was once at the forefront of global humanoid robot research. In the early 1970s, WABOT-1 was developed at Waseda University. From the 1990s onward, Honda advanced bipedal walking research, announcing ASIMO in 2000. At that time, for people around the world, "humanoid robot" was synonymous with Japan. Significant achievements were also accumulated in research infrastructure, such as AIST's HRP series. Nevertheless, according to McKinsey, no Japanese companies are among the top 10 current humanoid suppliers. This does not mean that Japan's research achievements were meaningless. The problem is that

Research → Product → Mass Production → Deployment → Data → AI Improvement

they were unable to build an industrial loop of this kind.

Chapter 5: "Suriawase" (Coordination) and the Parts Industry—A Huge Asset Remaining in Japan

One of the terms McKinsey highlights as a strength of Japan is

"Suriawase" (Coordination)

Japanese companies have excelled not only at optimizing machinery, electricity, control, materials, processing, and quality individually, but also at finely coordinating each of them to optimize the system as a whole. The automotive industry is a typical example. The same applies to robots. In the AI era, the objects of coordination increase further. Mechanical sensors, GPU/SoC, AI, firmware, communication, safety, cloud, and battery applications must all be integrated. Therefore, Japan's "culture of coordination" will not disappear. Rather, the question is

whether it can evolve from hardware coordination to "AI x Hardware x Data" coordination

is being asked.

This is the most important part of the argument for Japan's revival. McKinsey analyzes that Japanese companies alone can supply almost all the parts necessary for humanoids. For example, precision motors, speed reducers, linear guides, force sensors, torque sensors, encoders, bearings, power supplies, materials, and processing technologies. Particularly important in the future may be harmonic strain wave drives, planetary roller screws, linear guides for robots, 6-axis force sensors, and tactile sensors. In other words, it is not that Japan "lacks the parts." The problem is

whether to sell them as parts or evolve them into intelligent modules

is the issue.

Chapter 6: The Paradox of Japan as a Robot Powerhouse—Humanoids and Patent Competition

McKinsey presents this effectively as the "Japanese Paradox." The country that made the world's first humanoids. The country with one of the world's largest industrial robot companies. The country that can make almost all parts domestically. According to McKinsey, as of 2010, Japanese companies accounted for 5 of the top 10 industrial robot manufacturers in the world. Yet, their presence in the next-generation humanoid competition is weak. The speed at which the gap is widening is easy to understand by looking at the mass production track record of Chinese companies. In its report, McKinsey mentions that China's AgiBot has produced its 10,000th humanoid. However, this figure is already old. On June 28, 2026, AgiBot officially announced that its 15,000th robot had left the production line. Note that "cumulative production volume" and "actual number of units shipped and in operation" are not the same. Still, it is an important milestone showing the speed of expansion in mass production capacity. The speed at which Chinese production lines run is faster than the speed at which reports are printed. This itself symbolizes the subject of this article. The reason cannot be explained by a simple lack of technical capability. Japan has a strong tendency to "make a perfect product before releasing it to the market."

However, with AI robots, releasing to the market, failing, collecting data, improving, and redeploying becomes research and development itself. This difference is extremely significant.

According to McKinsey, the annual number of patent applications related to humanoids is approximately: China 7,700, USA 1,560, Japan 1,100. The original McKinsey source cites an article from the South China Morning Post dated December 21, 2025, and what is shown here is the annual number of applications, not the cumulative total. This point is important. Cumulative patents represent past technological accumulation, while annual applications show the current momentum of R&D activities. The gap between China and Japan is about 7 times, which is one indicator that R&D activities in China in the field of physical AI and humanoids are very fast. Note that the Japanese version of the McKinsey report does not translate the qualifier "annually," simply stating "number of patent applications." I note this because it is a point easily misunderstood as cumulative figures if one only reads the Japanese version. Of course, one cannot judge technical capability based solely on the number of patents. However, it is one indicator showing that Chinese companies are moving toward research, commercialization, and mass production at a furious pace.

Even more important is the difference between the research stage and the commercialization stage. While many Japanese companies are in the R&D stage, China and the US are beginning to deploy robots into actual factories and logistics sites. In physical AI, this "difference in the number of deployments" eventually becomes a "difference in the amount of data." And the difference in the amount of data becomes the difference in AI performance.

Chapter 7: Robot Density—Why "Japan 4th" and "Japan 5th" Coexist

This point requires caution in how statistics are read. McKinsey notes that Japan's manufacturing robot density retreated from world number one in 2009 to 5th in 2024. On the other hand, in "Robot Density Surges in Europe, Asia, and Americas" published by the IFR in April 2026, Japan is placed 4th with 446 units, following South Korea (1,220), Singapore (818), and Germany (449). In this series, China is not in the top 10, and a value of 166 units is shown using the new manufacturing employment figures from the National Bureau of Statistics of China. However, in another official IFR compilation, "Automation is a Cornerstone of Modern Manufacturing," for the same year 2024, South Korea is 1,220, Singapore 818, China 567, Germany 449, and Japan 446, making Japan 5th. In other words, as of 2026, two series with different treatments of China's manufacturing employment figures coexist on the official IFR website.. Robot density is calculated as "number of operating robots / number of manufacturing employees x 10,000." In a country with a huge manufacturing population like China, changing the denominator's statistical series causes the ranking to shift significantly. Therefore, here, without forcing a conclusion on whether "Japan is 4th" or "Japan is 5th," I will clearly state
4th in the IFR series published in April 2026, and 5th including China's 567 units in another official compilation based on World Robotics 2025. However, what is really important is not the ranking. At the end of 2024, the world's industrial robot operating stock was about 4.66 million units, with China having over 2 million and Japan 450,500. Of the 542,000 new installations in 2024, China accounted for about 295,000, or 54% of the world total. Japan's new installations were 44,500, remaining the world's second-largest market.

Japan's automation has not stopped. China is simply advancing automation with a much larger absolute volume and speed.

If you discuss rankings without looking at this structure, you will lose sight of the true competitive conditions in the era of physical AI.

Chapter 8: The Essence of Declining Production Share—What Japan Lacks Is Deployment Speed, Not Technology

There is another warning figure presented by McKinsey. It states that Japan's share of global industrial robot production has fallen from 54% in 2015 to 29% in 2024. This means it has nearly halved in less than a decade. However, it would be a mistake to interpret this figure as 'Japanese companies have collapsed.' Japan remains a massive robot-producing nation. The problem is that the growth rate of competing countries, led by China, is outpacing Japan. The robot industry is not a zero-sum game. The market itself is expanding. Therefore, it is not enough for 'Japanese companies to be growing' as well.

Can we grow faster than the global market?

This is what is important.

If McKinsey's analysis can be summarized in one phrase, it comes down to this.

Japan's problem is not technical capability. It is the speed of deployment.

No matter how excellent the research results are, if there are only 100 units in a laboratory, the data that can be acquired is limited. On the other hand, even if the initial performance is somewhat low, if you can deploy 1,000, 10,000, or 100,000 units into the field, you can accumulate vast amounts of experience and failure data. In the AI era, rather than the 'best robot from the start,'

'the robot that continues to improve the fastest'

has the potential to win in the end.

Chapter 9: China's Hardware-first and the US's AI-centric Approaches

McKinsey characterizes the representative strategy of Chinese companies as

Hardware-first

First, build the machine. Send it to the field. Collect data. Improve it. Mass-produce it. Further lower the price. China has a massive manufacturing supply chain that can rapidly cycle through motors, reducers, batteries, electronic components, sensors, processing, and assembly domestically. Therefore, the speed of moving from prototypes to mass production is fast. Price reductions through mass production increase sales volume, and increased sales volume increases data. This is a

mass production flywheel + data flywheel

system.

The opposite direction is prominent among US companies. McKinsey calls this

AI-centric

US companies are investing huge amounts of capital not only in the differentiation of the hardware itself, but also in AI, autonomous control, foundation models, simulation, and software stacks. The US has a powerful AI ecosystem that connects AI researchers, GPUs, the cloud, giant tech companies, VCs, and startups. On the other hand, not all robot parts can be procured within the US. Therefore, they utilize the global supply chain. To put it extremely, China's approach is 'from body to intelligence,' while the US's approach is 'from intelligence to body.'

Chapter 10: Japan's Third Way—Reliable Physical AI and New 'Suriawase' (Coordination)

It is not easy for Japan to compete on price using the same methods as China. It is also not simple to invest funds into massive AI models on the same scale as the United States. This is where a third path becomes important. That is,

Reliable Physical AI

a direction that could also be called. The third path Japan should aim for is a direction that can be called "Reliable Physical AI," which combines high quality, high precision, safety, long service life, maintainability, on-site integration, and edge AI. Especially in markets where the cost of failure is high, such as medical care, nursing, infrastructure, nuclear power plants, defense, and disaster response, the reliability and quality control that Japanese companies have cultivated hold great value.

Japan's traditional ability to refine and integrate is also effective in the AI era. However, the target needs to change. Conventional: Machinery x Electronics x Control Future:

AI x Sensors x Control x Machinery x Safety x Data x Cloud

AI gives the instruction "pick up the cup." The perception system recognizes the position of the cup. The arm moves. The tactile sensor detects contact. The force sensor adjusts the gripping force. It detects slippage. The motor control makes corrections. These work together in units of tens of milliseconds. This cannot be achieved with just a strong LLM. This is where there is room for Japanese mechatronics companies to enter.

Chapter 11: From "Manufacturing" to "Data Companies" for Robot Manufacturers

Until now, the sales of robot manufacturers have centered on the sale of the robot itself, maintenance, and replacement parts. However, in the physical AI era, Software Subscription, AI Model, Fleet Management, Remote Monitoring, Predictive Maintenance, OTA Update, and RaaS are added. In other words, it does not end with selling one robot. It generates revenue continuously for the 10 years that the robot is in operation. It can be called the SaaS-ification of the manufacturing industry.

The most important thing in physical AI is the

Data Flywheel

Deploy robots. ↓ Acquire on-site data. ↓ Collect failure examples. ↓ Re-train AI. ↓ Performance improves. ↓ Customers increase. ↓ Number of deployed units increases. ↓ Data increases further. Once this cycle begins, it becomes difficult for latecomers to catch up. What is important is not just success data. Dropped an object. Could not grip it. A human cut in. The floor was wet. The lighting changed. The shape of the box was different. This "long tail of failure" becomes an extremely valuable learning resource.

Chapter 12: Robot OS, OTA, and Fleet Management—Software Infrastructure That Grows After Shipment

McKinsey also points out an interesting point. It is not necessary to process everything with one huge general-purpose model. For example, in a logistics warehouse, create small models such as a pallet unloading model, a box gripping model, and a shelf stocking model. In a hospital, specialize in corridor movement, meal delivery, cleaning, and item transport. If you run small models distilled from large foundation models at the edge, you can also gain the benefits of low latency, low power consumption, communication failure resistance, and privacy. This is highly compatible with Japan's semiconductor, embedded, and control technologies.

The era where robot manufacturers have separate software for each model is also approaching its limit. What is needed is an integrated platform. Standardize API, Middleware, Simulation, Data Pipeline, Safety, Application, and Fleet Management. McKinsey cites mechanisms like the Open Robotics Middleware Framework (Open-RMF) as a concrete example. The idea is that standardized APIs and middleware lower the barriers to entry while simultaneously contributing to the expansion of the entire ecosystem. If third parties can develop applications, it could also develop into an ecosystem like an app store on a smartphone. In addition, McKinsey mentions robust OTA functions equipped with testing, phased deployment, and rollback, skill models optimized for specific uses, operation management and visualization tools for robot groups, and architectures with security built in from the design stage.

OTA, which is becoming standard in automobiles, will become even more important for robots. Make tasks that could not be done yesterday possible today through software updates. Fix security issues. Distribute new AI models. Improve safety. The time of shipment will no longer be the completion of the robot.

The day it is shipped becomes the day learning begins.

If there are a few robots, humans can manage them. However, as the number of robots increases to 100, 1,000, or 10,000, it becomes impossible to operate them with individual management. What is needed is Fleet Management. Centrally manage operating rates, failures, batteries, work assignments, positions, AI model versions, and security maintenance history. In the future robot industry, this management platform may generate higher profit margins than the robots themselves.

Chapter 13: Redefining RaaS and SI—From One-Time Sales to Continuous Service

Another model that McKinsey emphasizes is

Robotics as a Service (RaaS)

This is it. Customers do not buy expensive robots. They pay a monthly fee. Manufacturers provide not just the robot itself, but a package including software, AI, maintenance, and monitoring. For customers, this lowers the initial investment, making adoption easier. For manufacturers, it creates recurring revenue. This has the potential to significantly accelerate the speed of robot adoption. However, for RaaS to be viable, simply "selling" robots is not enough. McKinsey points out the need to establish regionally distributed service centers in addition to predictive maintenance, remote diagnostics, and continuous performance improvement. Failure rates and operational uncertainties are highest during the initial deployment phase. If a robot that stops on the factory floor cannot be restored in a short time, the customer will not order the next 100 units. Therefore, future competitiveness will be a product of:

AI performance × Hardware reliability × Maintenance network

For Japanese companies with nationwide service networks and quality control, this can be a powerful area of differentiation against US and Chinese startups.

Currently, industrial robot installation can require weeks to months of SI work. However, as AI robots become able to understand their environment and adapt on their own, installation time could be reduced to days, or in some cases, hours. If that happens, it will be difficult to maintain profits with only "SI that sets up robots one by one." Instead, SI companies need to expand their business into areas that optimize multiple robots and the entire site, such as Fleet Orchestration, Energy Management, Remote Operation, and Data Management.

Chapter 14: From Parts to Modules—The Winning Strategy of Smart Joints

Reducers. Motors. Sensors. Even if the performance of individual products is high, they will be subject to commoditization pressure in the long term. Therefore, what McKinsey proposes is a transition from:

Parts → Module → Platform

as the transition.

A representative example is

Smart Joint

Conventionally, motors, reducers, encoders, and force sensors were often designed and procured as separate parts. In a Smart Joint, high-precision motors, reducers, position sensors, force sensors, and computing units are integrated. Furthermore, internal software constantly monitors temperature, vibration, and load to detect signs of abnormality. Abnormalities are detected by AI, and parts are replaced before failure occurs. This is not just a joint; it is a

robot part with intelligence

.

Chapter 15: Safety, Action Stack, and Edge AI—Making Japanese Hardware Intelligent

Another very interesting concept is

Safety in a Box

If an AI robot is to work alongside humans, safety certification is essential. However, it is a huge burden for robot manufacturers to individually integrate safety MCUs, redundant power supplies, stop systems, verification software, and certification documents. Therefore, parts manufacturers provide everything integrated. Parts manufacturers provide a single package that includes FMEA, verification reports, and certification materials. Robot manufacturers can speed up safety design just by incorporating the "box." This is an extremely interesting market for Japanese companies that are strong in safety and quality control.

Converting AI instructions into reliable actions in the real world will also be a huge market. For example, a high-level AI commands, "Hold this egg gently." You cannot send that directly to the motor. What is needed is low-level control such as contact detection, force control, surface alignment, and slip detection. McKinsey lists the

Action Stack

as a new product candidate for parts manufacturers.

Physical control is difficult with cloud AI alone. There are applications where even a 100-millisecond communication delay can be dangerous. Therefore, it is necessary to run AI inside the robot. This is Edge AI. Elements such as small models, low power consumption, real-time inference, and safety control are highly compatible with the embedded and control technologies that Japanese companies have cultivated for many years. Even if we cannot build the "world's largest AI model,"

a strategy to build the world's most reliable Physical AI Edge Stack

is entirely viable.

McKinsey points out that in the future, the robotics industry will also need a manufacturing system equivalent to the "Toyota Production System." This would integrate Lights-out Factory AI quality control, predictive maintenance, reconfigurable lines, and digital twin real-time logistics management. As we compete with China, not only performance but also manufacturing costs will become critical. For Japan to make a comeback in physical AI, we need to advance to the point of

using AI to build robots, and then using those robots to build even more robots

as well.

Chapter 16: Why a National Strategy Became Necessary in 2026—The Challenge of a 20 Trillion Yen Market

In 2026, the Japanese government upgraded AI robotics to a full-scale national industrial policy. The Ministry of Economy, Trade and Industry's "AI Robotics Strategy Review Council" was held on January 21, February 13, and March 12, and the first government policy was compiled at a liaison meeting of relevant ministries and agencies on March 26. However, the timeline does not end there. The current page of the Ministry of Economy, Trade and Industry clearly states that it was compiled on March 26 and revised on May 27. Furthermore, at the Minister of Economy, Trade and Industry's press conference on June 30, the introduction target of approximately 10 million units by 2040 and the social implementation in a total of 18 fields, including food service, food manufacturing, and medical care, were explained as the revised content. Therefore, it is accurate to read the 2026 policy as following the flow of March formulation → May revision → June concretization of revised content and policy package. The industrial sector is also moving during the same period. On March 17, 2026, Keidanren published "Recommendations on the Future of Japan's Robot (AI+) Strategy." It places Japan's winning edge in "on-site capabilities x safety x quality" and advocates for global expansion as a robot (AI+) system. Politics, administration, and the industrial sector turned in the same direction almost simultaneously. This is rare in Japan. This is not just a robot policy. It has become a national strategy that cuts across AI policy, industrial policy, population policy, economic security, and regional policy.

The reason the government feels a sense of crisis is clear. In the era of physical AI, where AI intelligence enters the real world, the competitiveness of the robotics industry will be directly linked to the nation's manufacturing power itself.

In the government's AI robotics strategy, the global market in 2040 is estimated to be around 60 trillion yen. The term the government uses here is "multi-purpose robot." In the Ministry of Economy, Trade and Industry's materials, it is defined as a concept that keeps in mind forms such as humanoids, quadrupedal types, and mobile manipulators. It can be considered to refer to a range that almost overlaps with McKinsey's "general-purpose robotics." The government's view is this: The global market for multi-purpose robots, centered on humanoids, will expand rapidly around 2030 and reach a scale of approximately 60 trillion yen by 2040. And Japan aims to

capture over 30% of the global market

and aim for a market of

20 trillion yen in scale

. The United States. China. And Japan. It is a vision to make Japan the "third pole" of AI robotics. This is an extremely ambitious goal. The reason it is ambitious can be understood by reading the premises presented by the government itself. The Ministry of Economy, Trade and Industry's AI Robotics Review Council materials state the following:

Assuming that current market trends and the policy trends of each country continue, it is expected that China will capture more than half of the market size.

In other words, the goal of "Japan capturing over 30%" is placed on the premise that "if we do nothing, China will take more than half." The difference is the amount of policy change. The same document describes Japan's situation even more concretely. In the US and China, automakers and semiconductor manufacturers are also entering the robotics field, and research and development and capital investment exceeding 1 trillion yen are underway while utilizing existing AI infrastructure. Regarding startups, while the US and China have market capitalizations reaching hundreds of billions to trillions of yen and fundraising reaching hundreds of billions of yen, Japan remains at a maximum of tens of billions of yen. The order of magnitude is one to two digits different. This is the reality on the Japanese side of the recommendation that McKinsey makes to "rethink capital allocation." However, when considered together with McKinsey's $370 billion forecast, the direction in which a huge market will be formed is common. It is interesting to compare the numbers here. McKinsey's $370 billion general-purpose robot market is approximately 59.2 trillion yen if converted at 1 dollar = 160 yen. The robot market in 2040 envisioned by the government is approximately 60 trillion yen.

They are almost the same scale. However, just because the numbers are close, one should not conclude that they are the same market. The Ministry of Economy, Trade and Industry's approximately 60 trillion yen is a citation in policy documents of private forecasts for the "multi-purpose robot" market, which keeps in mind humanoids, quadrupedal types, mobile manipulators, etc. On the other hand, McKinsey's approximately $370 billion is a scenario for "general-purpose robotics." Although the two overlap significantly in concept, it cannot be confirmed that the target airframes, treatment of services and software, and estimation premises are completely identical. Therefore, in this article, we will organize it as **"approximately 60 trillion yen and approximately $370 billion show a similar scale, but they are not figures measured by the same method for the same market."** Also, $111 billion is not a figure that McKinsey guarantees or predicts for the future sales of Japanese companies. It is a conditional opportunity that applies the policy goal of "over 30% global share" put forward by the Japanese side to the $370 billion market.

Chapter 17: 10 Million Domestic Units, 18 Fields—Making Social Implementation a National Project

In the revised AI robotics strategy,

approximately 10 million units domestically by 2040

An AI robot adoption target was set. At the June 30th press conference of the Minister of Economy, Trade and Industry, this target and the social implementation in 18 fields, including food and beverage manufacturing and medical care, were explained as the pillars of the revised strategy. The true meaning of this number is not just the number of units deployed. If 10 million units enter the field, 10 million sensors will be in motion. 10 million units will experience failure. Field data will be generated from 10 million units. In other words,

the domestic market itself can be transformed into a massive physical AI learning device

This is where McKinsey's Data Flywheel strategy and the government's strategy connect.

The strategy set the target at 18 fields. It is designed to formulate an "AI Robotics Implementation Roadmap" for each field and proceed with social implementation based on it. Representative examples include manufacturing, shipbuilding, logistics, construction and civil engineering, architecture, infrastructure maintenance, retail, accommodation, food and beverage manufacturing, medical care, nursing care, security, agriculture, forestry, waste disposal, disaster response, police activities, and defense. This is an extremely important policy shift. Robots are not thought of as "factory machines."

They are thought of as social infrastructure.

Chapter 18: Domestic Foundation Models and Field Data—Building a Japanese Version of the Data Flywheel

A robot's body alone is not enough. What is needed is a brain. Therefore, the Ministry of Economy, Trade and Industry and NEDO launched the "Development Project for Multimodal Foundation Models Aimed at AI Robots and Physical AI in 2026. What is being aimed for here is an AI that can handle not only text but also images, audio, video, and sensor data in an integrated manner. In Physical AI, an "AI that understands language" is not enough. This is because an AI that understands the real world is necessary.

On June 30, 2026, NEDO announced the results of the public offering, and Noetra and the National Institute of Advanced Industrial Science and Technology were selected as the implementing organizations. The project period is from fiscal year 2026 to fiscal year 2030. What is being aimed for here is not just a Japanese version of an LLM. It is a

domestic multimodal foundation model

that also looks ahead to future robot control. It is also significant from the perspective of economic security. There is a risk in relying entirely on overseas AI companies for data acquired from manufacturing sites, hospitals, defense, and critical infrastructure. A domestic foundation is also necessary to ensure model sovereignty and data sovereignty.

In his press conference on June 30, 2026, Minister of Economy, Trade and Industry Akazawa cited data accumulated in elderly healthcare, disaster response, manufacturing sites, and the decommissioning of the Fukushima Daiichi Nuclear Power Plant as Japan's winning strategy. This is a very important statement. In the AI of the internet era, web data was important. In Physical AI,

real-world behavioral data

becomes important. And Japan possesses real-world environments that are not easily replicated in other countries, such as world-leading factories, an aging society, advanced medical care, disaster response experience, massive infrastructure, and nuclear decommissioning. If Japan can convert these into AI-Ready Data, the weakness of a shrinking population could conversely become a massive AI testing ground. However, data does not mean "you win if you collect it." McKinsey suggests that Japanese companies should consider building marketplaces and common models where they can share and use data. At the same time, clear governance is required for patient information and factory confidential information from the perspectives of privacy, trade secrets, and cybersecurity. There is not just one option for foundation models either. Develop a massive domestic model from scratch to increase sovereignty and control. Distill existing large models to make them smaller and lower cost. Fine-tune open models with Japan-specific manufacturing, nursing care, and disaster response data. The important thing is not the binary choice of "domestic or foreign," butstrategically deciding which layers to keep in-house and which layers to connect to the global ecosystem.

Policy Implementation Part 1: 300,000 Unit Deployment and 1 Trillion Yen Data Infrastructure

In connecting McKinsey's corporate recommendations to Japanese policy, the Liberal Democratic Party's "Robot Parliamentary League" is indispensable. The league was established in March 2025 and began its activities against the backdrop of a sense of crisis that the national-level robot strategy had not been sufficiently updated for about 10 years since the "New Robot Strategy" of 2015. In response to its efforts, the formulation of a new robot strategy was included in the 2025 Basic Policies, and the government's "AI Robotics Strategy" was officially formulated on March 26, 2026.

Five Recommendations for Social Implementation

On May 26, 2026, at a meeting in the National Diet, the Robot Parliamentary League compiled "Recommendations for AI Robotics Social Implementation: Realizing the Winning Strategy of Japan, a Country Facing Advanced Challenges, with the Full Strength of the Public and Private Sectors" and handed it to the Chief Cabinet Secretary in June (based on reports from the Sankei Shimbun and others). By this day, they had conducted a total of 14 hearings from companies such as Kawasaki Heavy Industries and Yaskawa Electric, as well as central government ministries. It is reported that the chairman, House of Representatives member Daishiro Yamagiwa, stated at the beginning of the meeting that robots are Japan's winning strategy, and that they have set a numerical target of introducing a cumulative total of 300,000 or more units and a cumulative total of 500 billion yen by 2030. What is important here is that they did not just stop at "R&D support," but stepped intomass adoption, data acquisition, government demand, domestic production, and budget with numerical targets. The main goals set by the league are as follows: 2026–2030: Introduce a cumulative total of 300,000 or more AI robotics units, with an annual average of 60,000 or more in 7 priority fields. The scale of introduction is approximately 500 billion yen in total. The 7 priority fields designated are manufacturing, logistics, construction and civil engineering, architecture, retail, the security industry, and nursing care. It is designed to concentrate resources on 7 fields where labor shortages are particularly serious, rather than the entire 18-field implementation roadmap.

June 2027: Release a beta version of the domestic robot foundation model as open source. 2030: Advance social implementation of common tasks such as "patrolling" and "moving objects" in the fields covered by the implementation roadmap. 2040: Japan will capture over 30% of the global 60 trillion yen market, approximately 20 trillion yen, and realize the introduction of a cumulative total of 10 million units domestically.

1 Trillion Yen Scale "Data Foundry"

The point in the recommendations that most overlaps with McKinsey's argument is the concept of apublic-private joint data foundry. In physical AI, a "data dilemma" occurs: robots don't sell ↓ real-world data isn't collected ↓ models don't improve ↓ performance doesn't increase ↓ robots don't sell even more. The league proposed that the government break this cycle through initial demand and a common data infrastructure. The scale of building the domestic foundation model and data infrastructure is said to be in the 1 trillion yen class, and the idea of increasing robot deployment from the demand side is shown, including support for monthly RaaS usage fees. Note that the total amount requested by the recommendations as the fiscal year 2027 budget is said to be on the scale of 1.5 trillion yen. 1 trillion yen is the portion allocated to the foundation model and data infrastructure. The text of the recommendations lists many concrete arguments that support this idea. If a robot that can reduce labor costs by 50,000 yen per month can be used for 10,000 yen per month, adoption will proceed explosively. Singapore achieved the creation of 300 robot companies through 70% subsidies to the demand side. Government support on the scale of 3 to 4 billion yen for 2,000 convenience stores will become the spark for mass production scale. In all cases, they are thinking in the order of "if you lower the price to an affordable level, the technology will grow," rather than "if you create the technology, it will sell."

The proposal positions the replacement cycle for service robots arriving between 2027 and 2028 as the "inflection point for physical AI dominance," stating that this is the last chance for Japan. This is nothing less than the idea of implementing the data flywheel that McKinsey emphasizes—

Deployment → Data → Model Improvement → Deployment

—as Japan's industrial policy.

Making government demand the "first customer"

The proposal also sets a direction for the government and local municipalities to take the lead in introducing AI robotics unless there is a specific reason not to. Targeted areas include disaster prevention infrastructure inspection, police activities, defense, academic and research, and administrative facilities. It goes further to include priority procurement of domestically produced robots, safety certification, a shift from specification-based ordering to performance-based ordering, and even triple-use across "industry, disaster prevention, and defense." Here, too, McKinsey's recommendation that "the government should be the First Customer" overlaps with the policy discussions on the Japanese side.

July 8, 2026—300,000 units and a 1 trillion yen initiative move to LDP PT proposal

Following the parliamentary league's proposal, the Robotics Strategy Project Team of the Liberal Democratic Party's Policy Research Council (chaired by House of Representatives member Fumiaki Kobayashi, with House of Representatives member Taishiro Yamagiwa as chair) compiled the "Proposal (Draft) for Social Implementation of AI Robotics" on July 8, 2026. The proposal consists of five pillars: 1. Countermeasures for labor shortages, 2. Large-scale introduction and practical application of robot foundation models, 3. Thorough utilization of government demand, 4. Phased domestic production, and 5. Steady promotion of the AI robotics strategy and implementation roadmap. In particular, it specifies investing multi-year budgets into seven priority fields such as manufacturing and nursing care, aiming for the large-scale introduction of over 300,000 units by the end of fiscal year 2030, and further advancing the construction of a 1 trillion yen scale domestic foundation model and data infrastructure. At the same time, it advocates for priority introduction by the national and local governments in principle, phased replacement of foreign-made units with domestic ones, and integrated promotion of hardware/AI development and human resource development, aiming to ensure these are reflected in the fiscal year 2027 budget compilation. The main battlefield of policy has already shifted from "whether to create a strategy" to "whether it can be put into the budget." It is moving from a policy that ends with providing research funds to a "policy that creates demand, produces units, generates data, and strengthens models."

Japan's AI robotics policy, at least in terms of design philosophy, is beginning to move significantly in the direction McKinsey is calling for. However, the true evaluation will not be determined by the budget amount.

How many units actually entered the field, how many hours did they operate, and how many pieces of valid learning data were returned to improve the model?

This is the KPI that should ultimately be watched.

Policy Implementation Part 2: AI Robotics Core Hubs and CoE

In the original McKinsey document, one policy item appears that is not often covered in Japanese articles: **AI robotics hubs**. McKinsey points out that as the Japanese government reviews its robotics industry development strategy for the first time in over a decade, it is advancing the development of hubs responsible for research and development and industrial support, and that OEMs and parts manufacturers can deepen their understanding of trends in robot introduction and future development directions in the Japanese market through collaboration with these hubs. The source cited is an article from The Japan News by the Yomiuri Shimbun dated March 26, 2026. In Japanese materials, this is positioned as a Center of Excellence (CoE)**. The government has indicated a policy to develop a world-class AI robotics core hub that integrates research and development, data collection, demonstration, safety evaluation, and human resource development. The background to this is the situation overseas. Robot companies and AI companies are setting up large-scale data collection hubs within their own facilities, collecting learning data while having robots repeat tasks. The government is preparing hubs that companies, universities, research institutions, and field operators can use, sharing demonstration environments and computational resources that are difficult for individual companies to maintain on their own. That is the CoE concept.

It is the idea of running the data flywheel mentioned in Chapter 21 not at the individual company level, but at the national level. Since this is a point easily missed in Japanese articles, I have intentionally devoted a section to it. When McKinsey wrote "leverage collaboration with the government," what they had in mind was not subsidies, but participation in this type of shared infrastructure.

Policy Implementation Part 3: July 2026, Japanese companies begin coordination

I have followed McKinsey's recommendations this far. Build an integrated software foundation. Share operational data with other companies and jointly develop AI models. Bring the strength of 'suriawase' (coordination) into the AI era. These are reasonable recommendations. However, recommendations usually end with "easier said than done." Moreover, about seven weeks after the publication of the original McKinsey document, concrete collaborations have been announced one after another on the Japanese side.

Fujitsu + FANUC + Yaskawa Electric + Kawasaki Heavy Industries + NVIDIA

On July 16, 2026, Fujitsu announced that it would begin business discussions in the field of physical AI with three companies: FANUC, Yaskawa Electric, and Kawasaki Heavy Industries (based on the announcement/Fujitsu press release). What Fujitsu is responsible for is the development of a cooperative control foundation. Utilizing the company's software foundation for physical AI, "Fujitsu Kozuchi Physical OS," it will develop software and hardware that serve as a common foundation. The goal is to facilitate coordination between "AI and robots," "robots and robots," and "robots and equipment." There are three points to note. First, the developed cooperative control foundation is planned to be provided to companies and research institutions as an open platform. Second, it is stated that sovereignty—data sovereignty—will be ensured in terms of security. Third, three companies that are originally direct competitors—FANUC, Yaskawa Electric, and Kawasaki Heavy Industries—are in the same framework. According to reports, the cooperative control foundation is scheduled to be provided to the three companies within 2026, and the plan is to implement it at Fujitsu's factory in Kahoku City, Ishikawa Prefecture, starting at the end of September, provide version 1 to each company based on that, and release version 2 in 2027 after receiving feedback. This is exactly what McKinsey wrote: "Break away from software fragmented by model and build a highly scalable integrated software foundation."

GENIAC adoption—three competitors share tactile data

There is another, even more advanced move. On July 2, 2026, it was announced that a joint project with Kawasaki Heavy Industries as the representative company, in collaboration with Osaka University, FANUC, FingerVision Inc., and Yaskawa Electric, titled "Construction of a Dataset for VTLA Foundation Models through Data Collection of Tactile Senses at Manufacturing Sites," was adopted for a public offering project by the Ministry of Economy, Trade and Industry and NEDO (based on the announcement). The official project name is "Research and Development Project for Strengthening Post-5G Information and Communication System Infrastructure / Research and Development on Building Data Ecosystems (GENIAC)." The target period is one year from August 2026 to July 2027. Based on reports, it will receive up to 2 billion yen in support from NEDO. In the first year, the plan is to collect 5,000 hours worth of video, tactile, and motion data from manufacturing sites and design and accumulate a dataset optimized for VTLA models. The targets are a wide range of manufacturing sites such as automobiles, robots, and food, and it will focus on collecting data on delicate tasks that conventional robots have been poor at, such as assembling small parts and handling soft cables that easily change shape.

The core is the VTLA (Vision-Tactile-Language-Action) model and the construction of a dataset and data ecosystem suitable for its learning. VTLA is a model that processes Vision, Tactile, Language, and Action in an integrated manner. The three robot manufacturers will standardize data specifications and collection infrastructure to build a dataset that can be used by various robots and devices. Furthermore, in collaboration with Nagoya University and the National Institute of Advanced Industrial Science and Technology (AIST), ABEJA will be in charge of advancing AI foundation technology, commissioned by Osaka University. FingerVision's tactile sensor has a vision-based structure that captures the deformation of an elastic body with a camera, and can detect the force distribution and slippage of the contact surface with high spatial resolution. By incorporating this into the VTLA model, the AI will be able to learn the subtle force control of "not gripping too hard, and not letting it slip." One can see the idea of the Action Stack mentioned in Chapter 31 beginning to move in the form of a national project. What is critically important here is the point that the three major competing industrial robot companies are standardizing data specifications.

McKinsey writes in its report that sharing operational data with other robotics companies and jointly developing AI models is also an effective way to overcome scale constraints. The three Japanese companies had begun moving in almost the same direction as the issues identified by McKinsey, in the form of concrete joint projects.

Participation of Japanese Companies in the Cosmos Coalition

On July 16, 2026, NVIDIA announced in Tokyo that a group of companies leading Japan's physical AI sector are leveraging the physical AI stack, including NVIDIA Cosmos, Isaac, Metropolis, and Jetson, to accelerate the deployment of intelligent machines in the fields of manufacturing, mobility, infrastructure, and robotics (based on the announcement/NVIDIA official press release). Companies and organizations that have expressed their intention to participate include AIRoA, classmethod, Enactic, FANUC, Fujitsu, GROOVE X, Hitachi, Honda R&D, Kawasaki Heavy Industries, Kubota, Mitsui & Co., Mitsubishi Corporation, Mujin, NEC, Preferred Networks, SoftBank, Sony Group, Telexistence, TIER IV, TRON K.K., Turing, and Yaskawa Electric. At the same time, NVIDIA announced a new model in the open world model group Cosmos 3,Cosmos 3 Edge. This is a model that enables embodied systems to "see," "reason," and "predict robot actions locally" in real-time on NVIDIA Jetson.

We should also note the specifications. Cosmos 3 Edge is a4-billion parameter model built on the foundation of NVIDIA Nemotron. The "Super" configuration, which is the largest in the Cosmos 3 family, is said to have approximately 65 billion parameters, making Edge about one-sixteenth of that size. The Cosmos 3 family itself was announced by NVIDIA at GTC Taipei on June 1, 2026. It adopts a Mixture-of-Transformers (MoT) architecture and features a dual-tower structure that combines an autoregressive inference path with a diffusion-based generative path that generates video and action sequences while considering physical laws. It is characterized by integrating world generation, physical reasoning, and robot action generation into a single model. Edge is a version of that architecture compressed for Jetson, and according to reports, it is designed for real-time control on Jetson Thor. Why 4 billion parameters? It is to complete the "see, think, and act" cycle within a single robot without having to travel back and forth to the cloud. This is directly linked to the "Edge AI" mentioned in Chapter 32 of this report. It can be seen that a concrete industrial foundation is beginning to form, connecting world models and edge inference to Japan's strengths in embedded systems, control, and robot hardware.

Challenges That Still Remain

When laid out like this, it may seem as if Japan is ahead of McKinsey's recommendations. However, there are things that should be viewed calmly. All of these are at the stage of "business feasibility study," "foundation building," and "dataset design." Dataset construction is by no means a small step. However, China's AgiBot officially announced on June 28, 2026, that it hadproduced its 15,000th robot in total. It should be noted here that cumulative production volume is not the same as actual shipment volume or operational volume. Even so, the fact that mass production capacity itself is expanding rapidly indicates the magnitude of the "implementation speed" required from the Japanese side. The version 1 provision of the cooperative control foundation is within 2026, and version 2 is in 2027. In the meantime, China's mass production lines will continue to run. What McKinsey emphasized at the end was not capability, but speed. There is no doubt that Japan has started walking in the right direction. The problem is the stride length.

Chapter 19: Manufacturing, Logistics, and Construction—The Three Fields Where Huge Markets Will Emerge First

The first huge market is, of course, manufacturing. However, it is not just the conventional automobile welding line. High-mix low-volume production. Parts supply. Inspection. Assembly. Inter-machine transport. AI robots will enter processes where "humans were cheaper" until now. Japan has a diversity of manufacturing sites that is rare in the world. This can be used as a demonstration field.

Logistics will become one of the most important markets for Physical AI. Palletizing. Depalletizing. Picking. Sorting. Shelf replenishment. Truck unloading. While logistics centers have a large amount of repetitive work, product shapes are irregular. It is precisely because this is an area where conventional automation was not good that AI robots have great value.

Construction sites are extremely difficult for robots. The floor is not flat. The weather changes. The location of materials changes. Humans move around. In other words, it is an Unstructured Environment. However, the more serious the labor shortage in the construction industry becomes, the higher the economic value will be. It is highly likely that introduction will proceed in stages, starting with inspection, transport, welding, and painting.

Chapter 20: Nursing and Medical Care—What Role Will Robots Play in a Society with Labor Shortages?

Nursing care could become a huge market unique to Japan. However, this is not about "replacing all human caregivers with humanoids." Cleaning. Trash transport. Meal delivery. Linen transport. Supply replenishment. Night patrols. Entrusting these tasks to robots and allowing humans to focus on interpersonal care has significant meaning in itself. McKinsey also highlights nursing and medical care as fields where Japanese companies can use quality, safety, and reliability as weapons. There is also backing for the market size. According to McKinsey, the cost of public nursing care services in Japan reached approximately 12 trillion yen in fiscal year 2024. According to the government's long-term projections, nursing care benefit expenses could expand to approximately 25 trillion yen by fiscal year 2040. The original English text shows this with a different indicator. It is a figure stating that the elderly care market will expand from approximately $650 billion in 2023 to approximately $780 billion by 2040. Although a simple comparison cannot be made because the scope of coverage differs between the Japanese and English versions, the conclusion that "the shortage of nursing care personnel will become even more serious" is common to both. It is an expenditure item that will double in 15 years. Even if only a few percent of that is directed toward robots, the market is sufficiently large.

As an example, McKinsey cites the "HSR (Human Support Robot)" developed by Toyota. Development is progressing for the purpose of supporting the elderly and people with disabilities. However, there is a clear condition here. To fully introduce robots that directly interact with the elderly and patients in hospitals and nursing facilities, issues regarding regulations, safety, and liability must be resolved first. McKinsey also cautiously reserves judgment on this point. It is an area where systems are more of a limiting factor than technology.

Chapter 21: Agriculture, Food, Disaster, and Defense—Turning Japan's Unique Real-World Environments into Competitive Advantages

In agriculture, harvesting, sorting, transport, weeding, and monitoring are targets for automation. In food manufacturing, processes such as plating, cooking assistance, cleaning, transport, and packaging are targets. Food is not uniform in shape, is soft, and is fragile. In other words, it is difficult for robot AI. However, precisely because it is difficult, there is a global market for successful technology.

An area where Japan is highly likely to create a global advantage is disaster response. Earthquakes. Tsunamis. Heavy rain. Volcanoes. Nuclear accidents. Robots are deployed to places where humans cannot enter. In particular, the decommissioning of the Fukushima Daiichi Nuclear Power Plant is an extremely unique robotics environment in the world. The data acquired there on remote operation, autonomous movement, and recognition could potentially be applied to nuclear power, mining, space, disaster relief, and defense in the future.

The importance of Physical AI is also rapidly increasing in the security domain. Explosive ordnance disposal. Hazardous material handling. Reconnaissance. Material transport. Facility inspection. McKinsey points out that the reputation as a "trusted partner" that Japanese companies have built over many years could be a strength in applications important for security. Here, not only performance, but Supply Chain Security, Cybersecurity, and Data Sovereignty become important.

Chapter 22: Management Reform—Mass Deployment, Business Sandbox, M&A, and Talent Acquisition

Perhaps what Japanese companies should change first is their KPIs. It is not just about the number of papers, the number of patents, or prototype performance. How many units were deployed? How many hours did they operate? How many types of tasks did they experience? How many failure data points were acquired? How many days did it take to update the AI model? These need to be added to management indicators. In the Physical AI era,

Deployment itself is R&D

That is why.

An interesting mechanism that McKinsey proposes to Japanese companies is

Business Sandbox

It involves creating small business entities separated from the normal organization of large corporations. They are given their own KPIs, salary systems, and decision-making budgets. They do not wait months for head office approval. They develop at the speed of a startup. Japanese companies have the capital and the technology. What they lack is speed. The Sandbox is an organizational design to buy that speed.

While you are training AI talent from scratch, your competitors will move ahead. That is where M&A becomes necessary. AI startups. Robot software companies. Sensor companies. Smart Joint companies. Perception companies. Through acquisitions, you gain not just technology but also talent. This is what is known as Acqui-hire. Whether Japanese companies can actively incorporate overseas startups will be extremely important.

McKinsey also points out that Japan has fewer engineers and Physical AI talent compared to China and the United States, which is a competitive constraint. It is difficult to acquire world-class AI researchers using only the conventional salary tables of Japanese companies. If necessary, you must also introduce stock options, equity, performance bonuses, and so on. The problem is not just technology. The talent market itself is becoming globalized. For Japanese companies to become Global Native, their human resources systems must also become Global Native.

Chapter 23: Becoming a Global Native Company—Assuming a Global Market from the Start

One of the most important recommendations McKinsey makes is

Global Native

In the past, Japanese companies expanded into the world in the order of success in the Japanese market, then Asia, then the West. However, today, the share of the global market held by Japanese manufacturing is significantly smaller than in the past. McKinsey cites specific figures. At the time these companies built their global status, Japan was the world's second-largest manufacturing power, accounting for about a quarter of global production. Domestic manufacturers were also active in automation investment. Today, the share of global production held by Japanese manufacturing is less than 5%.

From one-quarter to less than 5%.

The strategy of building a huge scale in the domestic market alone before expanding overseas is becoming difficult to sustain for this reason alone. Therefore, you must design prices, specifications, safety standards, APIs, languages, and supply chains for the global market from the beginning. It is fine to sell in Japan first. However,

you must not create "Japan-only products."

McKinsey cites Japanese laptop manufacturers as a warning. While they succeeded in the domestic market, they lagged behind competitors in China and the United States in the global market. The point is not to repeat the same mistake. Conversely, Hitachi, Ltd. is cited as a good example. It is evaluated for building a broad global business foundation and operating its business while giving each region a certain degree of discretion. Being Global Native is not a matter of export ratios. It is a matter of where you place decision-making authority.

Chapter 24: Can the Government Be the "First Customer"?—Government Demand and Regulatory Sandboxes

The government has another important role. It is not just about providing subsidies.

First Customer

It is about becoming the. Local governments. Hospitals. Public facilities. Disaster response. Infrastructure inspection. Defense. If the government creates initial demand, companies can deploy. If they deploy, data is created. Data makes AI stronger. Furthermore, it can be sold overseas. What McKinsey also emphasizes is the **Regulatory Sandbox**. For general-purpose robots that operate in the same space as humans, it takes time to proceed with verification using only normal regulatory processes when safety standards and liability boundaries are at an immature stage. Therefore, real-environment testing is permitted in limited areas, for specific uses, and for set periods, allowing companies and the government to design the safety standards themselves while collecting performance data. McKinsey points out that the available labor force in Japan will decrease by approximately 15 million people over the next 20 years. That is precisely why it is important to see whether fields with serious labor shortages, such as nursing care, logistics, construction, and infrastructure, can be turned intodemonstration fields that accelerate social implementation while verifying safetyrather than "experimental grounds for loosening regulations." Considering that the U.S. aerospace and defense industry grew through government procurement, public procurement can also be an important industrial policy for Japan's Physical AI policy.

Chapter 25: Conditions for Making the $100 Billion Market a Reality—Japan's True Path to Victory

$111 billion is not a guaranteed future. To repeat, this is not a McKinsey estimate, but a policy goal set by the Ministry of Economy, Trade and Industry. It is merely a figure based on capturing 30% of a $370 billion market. Therefore, it is a mistake to understand that "a $100 billion market is promised to Japan." Rather,

“If we can change the industrial structure starting now, we can target a market of that size.”

This is an opportunity with conditions. There is not just one condition. We must simultaneously advance AI, data, software, mass production, human resources, capital, M&A, regulatory reform, and global expansion, and connect all of these to deployment speed. This is not a competition where you can win by improving just one of these areas.

Looking at it this far, Japan's path to victory becomes clear. We should not compete with China solely on price. We should not compete with the U.S. solely on the number of parameters in Foundation Models. What Japan should create is,

“The world's most reliable Physical AI system that can be deployed to the field.”

To achieve this, it is necessary to integrate hardware, AI, Smart Joints, Edge AI, safety design, data, maintenance, and manufacturing into a single system. In this paper, we call this integration capability Reliable Physical Intelligence and call it that.

Chapter 26: Structural Changes in the Robot Industry in the 2030s

In the 2030s, the robot industry has the potential to become a massive ecosystem similar to the current automotive or smartphone industries. The robot industry of the 2030s may not just be a vertically integrated model where one company makes everything, but a massive ecosystem where specialized companies in AI Foundation Models, Robot Foundation Models, Simulation, GPU/SoC, sensors, Smart Joints, Robot OEMs, RaaS, Fleet Management, SI, Data, and Maintenance are interconnected. Therefore, it is not strictly necessary for a Japanese company to build a 'Japanese version of Tesla Optimus' on its own. We could supply Smart Joints to robots around the world. We could adopt a world-standard Safety Stack. We could capture the Foundation Model for nursing care robots. We could handle logistics Fleet Management. The important thing is,

which part of the value chain Japan will dominate

is the key.

Chapter 27: Conclusion—It is 'Learning Speed' That Will Determine the Next Robot Superpower

When interpreting McKinsey's '$100 Billion Market in Japan,' the biggest message is not the market size. It is not $370 billion. It is not $111 billion. The real message is,

the competitive principle of the robot industry itself has changed

that is the point. In the 20th-century robot competition, precision, speed, durability, and price were important. In the 21st-century Physical AI competition, AI, data, software, deployment, and learning speed also become competitive advantages. Japan still has powerful cards left. World-class precision machinery, robot manufacturers, the automotive industry, sensors, motors, reduction gears, machine tools, semiconductor materials, manufacturing sites, quality control, and 'suriawase' (craftsmanship-based integration). However, just possessing these will not lead to victory. We must get robots out into the field, let them fail, collect data, feed it back to the AI, improve them, and get them out again. We must increase that speed. China's greatest strength is mass production speed. The U.S.'s greatest strength is AI and capital. If Japan is to become the third pole, we must combine

“Field Capability × Monozukuri (Manufacturing) × Reliability × Physical AI”

into a single industrial system. And here, Japan's greatest weakness—a shrinking population—has the potential to turn into its greatest opportunity. In Japan, the number of working people will decrease significantly in the future. Nursing care, logistics, construction, agriculture, manufacturing, food service, infrastructure—it will become difficult to maintain society with humans alone. That is why Japan has a reason to use AI robots earlier than other countries. If so, we should turn all of Japan into a massive implementation field. Learn in factories, learn in warehouses, learn in hospitals, learn in nursing facilities, learn on farms, learn at construction sites, learn at disaster sites, and learn at the decommissioning site of the Fukushima Daiichi Nuclear Power Plant. Export the Physical AI born there to the world. Japan once supported the global manufacturing industry through automobiles, machine tools, and industrial robots. In the next era, we have the potential to become a country that supplies

“labor machines with intelligence”

to the world. But the time left for that is not long. China is already moving toward mass production. The U.S. is investing huge capital into AI. While Japan continues R&D, thinking, 'Let's wait until we improve the level of perfection a little more,' competitors' robots are walking in the field, grabbing things, failing, and learning. In the Physical AI era, failure is not just failure.

Failure itself is data, and data itself becomes the next competitive advantage.

That is why the next robot superpower will not be the country that makes the perfect robot first. It will be the country that sends the most robots into the real world, lets them experience the most, learns the most, and continues to improve the fastest. What the Japanese robot industry is being asked for is not technical capability.

It is speed.

And the "$100 billion market" presented by McKinsey is not merely a market forecast. It is an industrial competition through 2040 that asks whether Japan will end as a "robot superpower of the past" or be able to evolve once again into a "robot superpower of the Physical AI era."

Reference Materials and Related Links

The following is a compilation centered on primary and official materials used to verify the article's content and for additional research.

• McKinsey & Company, "Japan's $100 billion opportunity in
  general-purpose robotics," May 26, 2026

• Cabinet Secretariat, "Liaison Conference of Relevant Ministries and Agencies on AI Robotics"

• Ministry of Economy, Trade and Industry (METI), "Robot Policy"

• Ministry of Economy, Trade and Industry (METI), "AI Robotics Strategy Review Committee"

• Ministry of Economy, Trade and Industry (METI), "AI Robotics Strategy Review Committee" 1st Secretariat Materials (January 2026)

• Ministry of Economy, Trade and Industry (METI), "AI Robotics Review Committee" Reference Materials (October 8, 2025)

• NEDO, "Development Project for Multimodal Foundation Models Aimed at AI Robots and Physical AI"

• Ministry of Economy, Trade and Industry (METI), "Announcement of Selection for Development Project for Multimodal Foundation Models Aimed at AI Robots and Physical AI" (June 30, 2026)

• International Federation of Robotics, "Automation is a Cornerstone
  of Modern Manufacturing"

• IFR, "Robot Density Surges in Europe, Asia, and the
  Americas" (April 8, 2026)

• IFR, "Robot Density Surges in Europe, Asia, and the Americas" Japanese Press Release (April 8, 2026)

• IFR, "Global Robot Demand in Factories Doubles over 10 Years," World
  Robotics 2025 (September 25, 2025)

• McKinsey, "Japan's $100 Billion Market Opened Up by General-Purpose Robots" Japanese Edition (August 4, 2026)

• McKinsey, "Turning humanoid supply chain constraints into
  billion-dollar wins"

• McKinsey, "Humanoid robots: Crossing the chasm from concept to
  commercial reality"

• McKinsey, "Will embodied AI create robotic
  coworkers?"

• Keidanren, "Recommendations on the Future of Japan's Robot (AI+) Strategy" (March 17, 2026)

• FingerVision, "'Construction of a Dataset for VTLA Foundation Models through Visual-Tactile Data Collection at Manufacturing Sites' Selected for METI/NEDO Public Offering Project" (July 2, 2026)

• Robot Parliamentary League, "Five Recommendations for the Social Implementation of AI Robotics" (June 2, 2026)

• Liberal Democratic Party, "Towards the Social Implementation of AI Robotics:
  Recommendations from the Robotics Strategy Project Team of the Industry and Trade Division" (July 8, 2026)

• Fujitsu, "Commencement of Business Discussions with FANUC, Yaskawa Electric, and Kawasaki Heavy Industries for the Social Implementation of Physical AI" (July 16, 2026)

• NVIDIA, "Japanese Robotics and Manufacturing Leaders Pioneer the Frontier of Physical AI Based on NVIDIA
  Cosmos" (July 16, 2026, Japanese Press Release)

• "Construction of a Dataset for VTLA Foundation Models through Visual-Tactile Data Collection at Manufacturing Sites" by FingerVision Selected for METI/NEDO Public Offering Project (July 2, 2026)

• AGIBOT, "15,000th Robot Rolls Off the Production
  Line" (June 2026)

• Ministry of Economy, Trade and Industry 'Robot Policy' — Compiled March 2026, Revised May 27

• Ministry of Economy, Trade and Industry 'Summary of Minister Akazawa's Press Conference After Cabinet Meeting' (June 30, 2026)

• IFR 'Automation is a Cornerstone of Modern
  Manufacturing'

• NVIDIA Newsroom 'Japan's Robotics and Manufacturing Leaders Build
  on NVIDIA Cosmos to Advance Physical AI Frontier' (July 15, 2026)

Hashtags

#AI #ArtificialIntelligence #Robot #Robotics #AIRobotics #PhysicalAI
#PhysicalAI #GeneralPurposeRobot #GeneralPurposeRobotics #Humanoid
#HumanoidRobot #EmbodiedAI #GenerativeAI #RobotFoundationModel #VLA
#VTLA #MultimodalAI #McKinsey #METI #NEDO #AIST #Noetra
#GENIAC #FANUC #YaskawaElectric #KawasakiHeavyIndustries #MitsubishiElectric #Toyota #Fujitsu #NVIDIA
#Cosmos #CosmosCoalition #AgiBot #Unitree #FigureAI #Manufacturing #Logistics #NursingCare
#Construction #SmartFactory #DataFlywheel #DataFoundry
#RaaS #SmartJoint #SafetyInABox #EdgeAI #RegulatorySandbox #GovernmentDemand
#RobotParliamentaryLeague #IndustrialPolicy #EconomicSecurity #FutureOfJapan #JapanRevival #AIStrategy
#IFR #RobotDensity #WorldRobotics #CoE #AIRoboticsHub #FingerVision
#ABEJA #OsakaUniversity #Keidanren #DaishiroYamagiwa #FumiakiKobayashi