Japan Bets on Physical AI Robots to Fill a Labor Gap

What people on the ground actually want from robots when there aren’t enough workers

I was reading an article about Japan’s robot industry, and what caught my eye was the stories from workplaces struggling to keep up current levels of production and service because they simply can’t find enough people. Two problems are tangled together here: automation changing what human workers do, and the plain inability to hire anyone at all. It struck me that Korea has something to think about here too.

In March 2026, reports emerged that Japan’s Ministry of Economy, Trade and Industry is aiming to grow the AI robot industry—Physical AI¹—to capture roughly 30% of the global market by 2040. The goal is to address the labor shortage while cultivating a new industry at the same time. Future Forecast 2040 (Mirai Yosoku 2040), published by the Recruit Works Institute, estimated in its baseline scenario that the labor supply-demand gap will reach around 11,000,000 people by 2040. To be clear, this isn’t today’s shortfall, nor a locked-in future figure.

In this issue, I want to look at Japan’s labor demand, the strengths of its robot industry, and what it actually takes to bring robots into real workplaces.

The pressure to keep production and services running

There are several reasons behind robot adoption: cost cutting, productivity gains, safety, and securing workers, among others. It’s not that each country has a single, fixed motive. But in TechCrunch’s reporting from Japan, labor shortages keep coming up.

According to Japan’s Statistics Bureau, the country’s total population stood at roughly 123.8 million as of October 1, 2024, down 550,000 from the previous year. That marked 14 consecutive years of population decline, with the 15-to-64 age group making up 59.6% of the total. It’s worth distinguishing between figures that count only Japanese nationals and total population figures that include foreign residents.

Demand in caregiving is also rising. Japan’s Ministry of Health, Labour and Welfare projects that the country will need approximately 2.72 million caregiving workers by fiscal 2040 — 570,000 more than the roughly 2.15 million in 2022. This isn’t a forecast of a nursing shortage; it’s the projected increase in workers needed for future caregiving services.

Hogil Do, a partner at Global Brain whom TechCrunch interviewed, explained that there’s strong demand for keeping factories, logistics centers, and services running with fewer people. I read this as a reminder that the case for robot adoption shouldn’t come only from vendors’ technology pitches — it should be grounded in the operational problems customers actually face.

In these settings, finishing the work already on the table on time matters just as much as boosting output further. That said, robots can’t solve labor shortages on their own. Work practices, working conditions, and workforce recruitment all need to be addressed together.

Making Manufacturing Experience Work for New Tasks

Japan is home to companies like FANUC, Yaskawa, and Kawasaki, which have been building industrial robots for decades. According to preliminary figures released by the IFR (International Federation of Robotics) in July 2025, roughly 13,000 new industrial robots were installed in Japan’s automotive industry in 2024 alone. The manufacturing and operational experience gained from automating repetitive processes forms a crucial foundation here.

For a robot to move precisely, actuators², sensors, motors, and control technology all have to work in concert. This requires not just precision components, but real factory-floor experience in handling malfunctions and errors. The challenge is connecting the capabilities Japanese firms have accumulated over the years with new AI models.

What matters most is how much of the strength built into existing industrial robots carries over to new tasks. Repeatedly moving an object from a fixed location is a fundamentally different problem—in terms of perception and control—from picking up an object placed in a different spot each time.

When comparing competitive positions, it also helps to break things down into hardware, perception models, control, data, and on-site operations. If you assume one country handles software and another handles hardware, you’re likely to miss the different strengths and collaborations that exist at the level of individual companies.

Issei Takino, CEO of Mujin, made a similar point in a TechCrunch interview, emphasizing the need for a deep understanding of hardware’s physical characteristics. Because instructions issued by software ultimately translate into the movement of real equipment, he argued, you have to account for precision control and the cost of failure.

I think this is precisely where Japan’s opportunity lies. The question is whether Japan can build systems that respond to new objects and working conditions while drawing on its deep understanding of existing robots and processes.

Investment Announcements and Actual Deployment Are Two Different Things

On top of the Japanese government’s AI and robotics development plans, private-sector investment is being announced too. On April 3, 2026, Microsoft said it would invest $10 billion in Japan between 2026 and 2029. The plan covers AI infrastructure, cybersecurity, and workforce training — it doesn’t mean the entire sum goes toward buying robots, or that it’s already been spent.

Evaluating real-world deployment requires a separate set of criteria. We need to look at whether customers actually pay for and keep using the systems, whether they run stably through an entire shift, and what the uptime rate³ and the number of human interventions look like. That’s the distinction between a successful trade-show demo and results from long-term operation.

The cases of SoftBank and Yaskawa illustrate this gap. On March 25, 2026, SoftBank announced it had verified pick-and-place operations at its own logistics warehouse. The setup paired a vision-language model (VLM)⁴, which interpreted objects and layout information, with a separate action model on the robot’s side that generated the actual movements. It was a verification using real warehouse tasks, but it’s not an announcement that proves the profitability of full commercial deployment or long-term unmanned operation.

Getting multiple pieces of equipment to move together also requires software to coordinate task sequencing and integration with existing logistics systems. Digital twins⁵ or simulations can be used to test layouts and movements in advance. Excelling at either hardware or software alone doesn’t solve all the problems that show up on the ground.

Equipment Makers and Software Companies, Working Together

There’s room in this industry for hardware makers and the companies that connect and operate their systems to collaborate. That said, we can’t assume in advance that the market will stay concentrated around any particular company going forward.

Mujin, which appears in the article, builds a platform that automates equipment from multiple manufacturers at once. WHILL combines powered mobility devices with sensors, autonomous driving, and operations management. Terra Drone also works on field applications for autonomous systems. None of these are companies making the same robot—they’re targeting different tasks and different customers.

Sho Yamanaka of Salesforce Ventures explains that large companies’ manufacturing base and customer relationships can complement startups’ software and systems development. For this kind of collaboration to actually work, though, you also need to settle who fixes things when they break, and how data gets shared to improve performance.

Partner Ho-gil Do emphasizes the role companies play in installation, integration, and continuous improvement. It’s a point worth keeping in mind, given that customer working conditions need to keep being factored in even after the robot is sold.

Oswarld’s Lens

I think Japan’s labor shortage can create concrete demand for robotics companies. When the work that needs solving is clear, it becomes easier to discuss where to apply the technology and what needs improving.

In working on go-to-market strategy, I’ve seen how adoption discussions become more concrete when customers face a problem they absolutely must solve. But urgent need doesn’t translate into immediate purchase. Price, installation timelines, integration with existing equipment, and safety and maintenance conditions all have to line up. Japan’s Physical AI push will need to clear these same hurdles one by one.

What worries me is a focus on equipment performance alone, at the expense of on-site operations. A robot that’s excellent at picking up objects won’t deliver the expected results if it isn’t connected to the production schedule, or if a human has to hover nearby every time it errors out.

So rather than growth forecasts for market size, I want to see why customers keep paying after adoption. Changes like whether actual throughput increased, whether workers’ exposure to dangerous tasks decreased, whether downtime shrank.

I think it could work to Japanese companies’ advantage to connect their equipment and process experience to software integration. Knowing hardware well can help, but it doesn’t by itself guarantee a more stable system. You also need the capability to test in the field and fix the root causes of failures.

Korea, too, is dealing with low birthrates, an aging population, and labor shortages across industries. I think we need to weigh both the impact robots will have on employment and their potential to ease labor shortages. A good starting point would be identifying which tasks are being delayed, and how automation would change the work of the people currently responsible for them.

Closing

Reading through this material, I kept coming back to one idea: Japan’s robot strategy is fundamentally about connecting manufacturing capacity to new on-site demand. Labor shortage forecasts make the case for why this is necessary, but they don’t answer which robots to deploy, or how many, to actually solve the problem.

So it’s worth keeping these things separate: government market-share targets, corporate investment plans, field validation, and actual commercial operation. Each one matters, but none of them can substitute for the others.

What I want to look at isn’t just what a single robot can do, but how consistently the organization that deployed it can keep the work running. That operational experience will matter just as much when we think about how to apply any of this in Korea.

References & Further Reading

• Recruit Works Institute, Future Forecast 2040, 2023. : A report containing a labor supply-demand simulation.

• Japan’s Statistics Bureau, Population Estimates as of October 1, 2024.

• Japan’s Ministry of Health, Labour and Welfare, Workforce Requirements under the 9th Long-Term Care Insurance Business Plan, 2024. : It explains the increase in required workforce compared to 2022.

• IFR, Robot Installations in Japan’s Automotive Industry, July 15, 2025. : These are preliminary figures for 2024.

• SoftBank, Validating Physical AI Operations in Warehouses, March 25, 2026.

• TechCrunch, “In Japan, the robot isn’t coming for your job; it’s filling the one nobody wants”, April 5, 2026. : This TechCrunch piece covers Japan’s robotics industry, offering an on-the-ground analysis based on interviews with VCs and corporate CEOs.

• News On Japan, “Japan Moves to Develop Domestic Physical AI, Targets 30% Global Share by 2040”, March 18, 2026. : This reports on METI’s draft Physical AI strategy announcement.

• Prism News, “Japan Accelerates Physical AI Deployment to Combat Persistent Labor Shortages”, April 6, 2026. : It rounds up recent deployment cases, including the FANUC-NVIDIA collaboration.

• The Diplomat, “Japan’s Grim Demographic Reality”, December 2025. : It offers an in-depth look at the structural causes of Japan’s population decline and the policy dilemmas it creates.

• CNN, “Japan’s population decline keeps getting worse”, August 7, 2025. : This report covers the decline in Japanese-national residents. Note that its scope and timing differ from the Statistics Bureau’s total population estimates, which include foreign residents.

• Microsoft Source Asia, “Microsoft deepens its commitment to Japan with $10 billion investment”, April 3, 2026. : Microsoft’s announcement of a $10 billion investment in Japan, detailing its AI infrastructure and workforce development strategy.

• Asia Tech Daily, “Japan’s AI Reset: What the Government’s First National Plan Means for Startups”, December 23, 2025. : This analyzes Japan’s ¥1 trillion national AI strategy from a startup’s perspective.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

Footnotes

1. Physical AI refers to AI that interprets information from sensors, cameras, and other inputs to guide the movements of robots or other physical devices. Not every device learns on its own or operates fully autonomously while running. ↩

2. Actuator: a device that converts energy into motion, such as an electric motor or a hydraulic cylinder. ↩

3. Operating rate (utilization): the share of a defined observation period during which equipment ran normally. Since “uptime” can also refer to operating time, it’s worth checking the baseline period and the definition of downtime before comparing figures. ↩

4. VLM (Vision-Language Model): a model that processes images and text together. To translate this into actual robot movement, it still needs to be connected to an action-generation and control system. ↩

5. Digital twin: a method of representing real equipment or systems as data and digital models for analysis and testing. It doesn’t perfectly replicate reality, so the model’s accuracy and data freshness matter. ↩