Four widely repeated claims about Japan’s eldercare crisis in 2026, where each one came from, and what survives after the arithmetic.
AsiaPresswire – Japan has been buying machines for its care homes for more than a decade, and it has been short of care workers for longer than that. Both facts are true at the same time, which is inconvenient for the story everyone prefers. The story is that robots arrived, the workers stopped being needed, and the arithmetic closed. In 2026 the arithmetic has not closed. The machines are real, they are working, and they have not removed a single worker from the shortage.
What follows is not a claim that the technology failed. It is a claim that four things said about it are wrong, that each wrong thing has a traceable origin, and that the origin explains the persistence better than any failure of reporting does. Every one of the four has the same shape: a true number was published, then attached to a conclusion it cannot support.
The four claims, and where each one came from
Table 1: Four claims about Japanese eldercare in 2026, their published origin, and what the number actually supports
| Claim | Origin | What the source number actually says |
| Robots will close the care worker gap | Government care robot subsidies and adoption programmes | Adoption measured by facilities with at least one robot, not by workers displaced |
| Foreign care workers will close the gap | Care visa categories created from 2017 onward | A stock figure reported against an annual increment requirement |
| The insurance fund is running out | Annual total care cost reaching record levels | A total without a per-recipient or GDP denominator |
| Robots pay for themselves because labour dominates costs | Standard industry cost structure | Labour being 60 to 70 per cent of cost is why the margin is 2.4 per cent |
Source: Ministry of Health, Labour and Welfare 🔗, ninth-term care insurance plan; Caijing 🔗, 2026
Myth one: the robots were bought to replace care workers
This is the oldest of the four and the easiest to source, because the government said something close to it. Care robot development subsidies from the trade ministry and deployment subsidies from the health ministry have run for years, and the programmes are named for what they buy. When a subsidy programme is called a robot programme, success gets counted in units purchased.
The fact core is that adoption is real but does not do what the name implies. A 2026 study by a Keio University team looked at 4,688 long-term care facilities and found that 1,250 of them, 26.7 per cent, had adopted at least one care robot. Larger facilities with better digital infrastructure were significantly more likely to have adopted. That last clause is the part that matters: adoption tracks organisational capacity, not labour scarcity. Facilities that can absorb a robot are the ones that buy one.
Then there is what happened to staffing. A 2024 National Bureau of Economic Research study using real deployment data from Japanese care facilities found that robot adoption was associated with higher staff numbers and higher retention, alongside improved productivity and declines in some care quality indicators, specifically physical restraints and pressure ulcers. Higher. Not lower. The machines did not remove workers; they were associated with facilities that could keep them.
Why is the myth believed? Three mechanisms, all concrete. Adoption is measured as “at least one robot”, so a facility with a single therapeutic seal robot counts the same as one running a full transfer system. Subsidy reporting counts procurement, not outcomes, and nobody reports displacement because displacement is not the programme’s stated objective. And humanoid photographs travel further than bed sensors; the transfer robot that can be filmed is one percent of the installed base, and the motion sensor that is actually everywhere is not photogenic.
What is left after the myth is removed is more useful than the myth. Robots in Japanese care homes do monitoring, lifting and transfer assistance, and documentation. Monitoring is the one with scale: bed sensors, floor sensors and privacy-preserving silhouette cameras let one caregiver oversee eight to twelve residents instead of three or four. That is a change in what the worker does with a shift, not a change in how many workers a shift needs.
Myth two: foreign care workers will close the gap
Japan created a care worker visa category in 2017, added a specified skilled worker route in 2019, and expanded permitted immigration with Type II status from 2024. The policy is real and the numbers have risen. This is the myth that a true statistic most reliably produces.
The Ministry of Health, Labour and Welfare published its ninth-term care insurance plan estimate of required care workers. Aggregating prefectural projections, the requirement is about 2.4 million in fiscal 2026, an increase of roughly 250,000 over the 2.15 million recorded in fiscal 2022, and about 2.72 million in fiscal 2040, an increase of roughly 570,000. The pace implied is 63,000 net additional workers per year through fiscal 2026, easing to 32,000 per year after that.
Against that, the number of foreign care workers in Japan under the specified skilled route reached about 65,500 in 2025, a record high. Government plans reportedly contemplate expansion to at most 160,700 across current and new schemes.
Set the two side by side and the mismatch is a category error rather than a slow start. 65,500 is a stock. The requirement is an annual increment of 63,000. Even if every one of those 65,500 workers had arrived in a single year, and they did not, one year’s increment would absorb the entire stock. The most generous reading of the 160,700 target is about two and a half years of requirement.
There is also a genuine absorption constraint, and it is not visas. Japanese-language training, certification and housing are what the sector reports as the binding limits. The Type II expansion in 2024 widened who could come; it did not widen how many the sector can onboard per year. A record-high stock figure reported without the annual increment beside it will always look like progress.
Myth three: the insurance system is running out of money
This one has the largest number attached to it, which is exactly why it travels.
Total long-term care cost in Japan reached 12.22 trillion yen in fiscal 2025, according to health ministry figures, up from roughly 3.6 trillion yen at the system’s start. Monthly figures for the April 2026 review cohort put care service cost at 1.0089 trillion yen for the month, with 4,841,200 recipients and 7.61 million people certified as needing care or support. Separate estimates put fiscal 2025 cost at 12.8 trillion yen, and the two are close enough that the level is not in dispute.
Table 2: Japan’s long-term care numbers, 2023 to 2026
| Measure | Figure | Period |
| Category one insured persons | 35.89 million | 2023 |
| Late-stage elderly insured, 75 and over | 20.18 million | 2023 |
| Certified as needing care or support | 7.08 million | 2023 |
| Certified as needing care or support | 7.61 million | April 2026 |
| Annual care service recipients | 5.78 million | fiscal 2025 |
| Monthly care service cost | 1.0089 trillion yen | April 2026 |
| Total annual care cost | 12.22 trillion yen | fiscal 2025 |
Source: Ministry of Health, Labour and Welfare 🔗; health ministry annual care cost figures
The number is large and rising, and it is not a solvency claim. Divide 12.22 trillion yen across 5.78 million annual recipients and the result is roughly 2.11 million yen per recipient per year. Divide it across 7.61 million certified people and it is about 1.61 million yen. Those are the denominators that turn a frightening total into a discussion about unit cost and demographic composition, and they are rarely printed next to the headline.
What the policy debate is actually about is narrower and more specific: whether to expand the share of users paying 20 per cent of costs rather than 10, with a decision on the criteria due within fiscal 2026 and the social security council examining options. That is a real distributional argument about who pays, which is a different thing from a fund that cannot pay.
The mechanism for this myth is simple and near universal. A total in the trillions is reported because it is the number the ministry publishes, and a per-recipient figure requires a division nobody in a newsroom has time to do. The claim of insolvency arrives afterwards, attached to a number that supports only the claim of size.
Myth four: robots pay for themselves because labour dominates costs
This is the most interesting of the four because the premise is true and the conclusion is backwards.
Labour accounts for 60 to 70 per cent of costs at a Japanese care facility. That is the standard industry figure, and it is why automation looks so obviously attractive: attack the largest line item. The problem is the other number. Average operating margin in the sector is about 2.4 per cent.
A business that spends two thirds of revenue on staff and keeps 2.4 per cent at the bottom does not have capital lying about for equipment. This produces the sector’s quiet paradox, which is that the more acute the staffing shortage at a facility, the more it would benefit from robots and the less able it is to buy them. Subsidies exist precisely because of this: government support for care robot and ICT equipment rose from 5.39 billion yen in fiscal 2021 to 12.24 billion yen in fiscal 2023, an increase of 127 per cent. Overall care technology adoption is still only about 30 per cent.
So the mechanism is an inference from a true cost structure to a false affordability conclusion. Anyone who has seen a cost breakdown knows labour is the big line. Fewer people have seen a margin line, and the two facts together invert the argument.
What is actually installed, and what is still a pilot
One reason the robot myth is hard to dislodge is that “care robot” covers at least five different things, and the public imagination keeps reaching for the one that barely exists.
Installed at operational scale in Japanese facilities by 2026:
- Lifting and transfer assistance, including the lumbar support variant of the HAL exoskeleton from Cyberdyne and the RIBA II lifting robot developed at RIKEN, alongside the PARO therapeutic seal on the companionship side.
- Movement and activity monitoring: bed sensors that detect a resident leaving bed, floor sensors for falls, and privacy-preserving silhouette-only cameras. Vendors include Konica Minolta, Aeolus Robotics and Z-Works. This is the category that changes staffing ratios rather than headcount.
- Communication and engagement, mostly tablets with activity software and video calling, which have proved more durable than the humanoid platforms that briefly occupied the same role.
- Health monitoring through wearables and ambient sensors from Hitachi, Omron and Toshiba, feeding facility care management systems.
- Care planning and documentation, where generative drafting of care notes, shift handovers and regulator-required paperwork has been the fastest-growing category, deployed by operators including Benesse Style Care and Nichii Gakkan across hundreds of facilities.
Still in pilot, and consistently described as such by the people running them: humanoid robots for fetch and carry, where cost remains prohibitive; autonomous mobility for residents with disabilities, at a similar distance from production as automotive autonomy; and predictive monitoring that would anticipate falls or deterioration days ahead, where the signal is present but the false positive rate is not yet acceptable in a care setting.
The market reflects the gap between the two lists. Japan’s AI-driven eldercare robotics market was around US$210 million in 2025 and is projected to reach about US$506 million by 2031, with physical assistance the largest revenue pool and software and services approaching 40 per cent of value by the end of the forecast. That is a real market and it is small. A sector with 2.4 million workers and a 12 trillion yen annual cost base is not going to be reshaped by a US$210 million equipment line, which is another way of stating the same finding from the other direction.
Why all four spread, in one mechanism
Each myth above has its own origin, but they share a reporting structure. In every case a number was published by a credible body, in a form designed for a different purpose, and then used for a purpose it was not designed for.
- Subsidy programmes count procurement, so procurement is reported as adoption.
- Visa programmes count people present, so a stock is reported against a flow.
- Insurance programmes count total cost, so a total is reported as a solvency signal.
- Cost structures show labour share, so labour share is reported as an automation case.
None of these is a fabrication, and none is a reporting error in the narrow sense. Each is a metric used outside its design. The correction is not to find better numbers. It is to ask, for every number, what denominator would make it mean what the sentence says.
What is left when the four are removed
Strip out the four claims and the situation is less dramatic and more difficult, which is probably why the claims survive.
The demand side is unambiguous and worsening. Japan’s population aged 65 and over was around 29 per cent in 2026 and is projected to cross 35 per cent by 2040. The requirement for care workers rises to 2.72 million by fiscal 2040. The care workforce is itself ageing, with 41.6 per cent of care workers aged 50 and over and 7.7 per cent aged 65 and over. In households providing care at home, 64.5 per cent have both the carer and the person receiving care aged 65 or over, and close to 30 per cent involve someone aged 75 or over caring for someone else aged 75 or over.
The supply side is being pulled in three directions at once, and none of them closes the gap alone:
- Retention and productivity, where robots demonstrably help, since adoption correlates with higher staffing and fewer restraints and pressure ulcers.
- Immigration, where the ceiling appears to be measured in tens of thousands against an annual requirement of 63,000.
- Cost structure, where a 2.4 per cent margin means the sector cannot self-fund capital, and subsidies rose 127 per cent between fiscal 2021 and fiscal 2023 without lifting adoption past about 30 per cent.
There is a fourth pressure that rarely appears in the same article as the robot statistics, and it is the one that sets the pace. The sector is failing financially at record rates while the demographic clock runs. Between January and November 2025, 85 home-visit care providers closed, and closures across elderly welfare facilities reached 172 for the year, a record high. Facilities closing is not a staffing story in the usual sense; it is capacity leaving a system that is simultaneously being told it will need 2.72 million workers by 2040.
The honest conclusion is that Japan is not solving this. It is distributing a fixed shortage across more monitoring equipment, more foreign workers than before, and a thinner margin, and it is doing so while the number of people over 75 rises. The machines are working. They are working on the wrong variable, and everyone involved can see it.
If you want one number to watch, take the ratio of foreign care workers entering per year to the 63,000 annual increment. Everything else in this story is slower to move. Certification numbers move on demographics, care cost moves on certification, and robot procurement moves on a subsidy line. Only migration is a policy dial that can be turned within a single fiscal year, and it is the one currently set at a fraction of what the arithmetic needs.
That is the whole picture, and it is less exciting than the robot version. Japan has bought machines, admitted workers, and kept a system running at a 2.4 per cent margin while a quarter of the population passed 65. What it has not done is find a substitute for a person who lifts, washes and talks to someone at three in the morning. The search continues, and the machines are helping with the parts that can be helped.
Data basis
Data as of October 2026. Care worker requirement figures are the health ministry’s ninth-term care insurance plan aggregation of prefectural projections, and are projections, not counts of employed staff; the 2.15 million fiscal 2022 base is a recorded figure. Care cost figures are annual totals and monthly service cost as published by the health ministry, with fiscal 2025 total at 12.22 trillion yen and an alternative published estimate at 12.8 trillion yen reported here as a range rather than a choice. Per-recipient figures are this piece’s own division of published totals by published recipient or certification counts, and are stated as approximations. Robot adoption and staffing correlations are from a 2026 Keio University facilities study and a 2024 National Bureau of Economic Research study, both observational, and association is not causation. Margin, labour share and subsidy figures are sector figures as reported. No figure in this piece is a projection presented as an actual.
Limitations
- Care worker requirement is a projection from prefectural plans, not a measured headcount, and the fiscal 2040 figure carries two decades of assumption.
- The two robot studies are observational; the association between adoption and higher staffing may reflect that better-resourced facilities adopt, rather than that adoption raises staffing.
- Published fiscal 2025 care cost ranges from 12.22 to 12.8 trillion yen depending on source, so per-recipient figures carry that spread.
Sources
- Ministry of Health, Labour and Welfare 🔗, required care workers under the ninth-term care insurance plan, 2026
- Caijing 🔗, Keio facilities study, NBER findings, sector margins and subsidy figures, 2026
- China Times 🔗, foreign care worker numbers and government expansion target, September 2026
- PDP Spectra 🔗, categories of eldercare technology deployed at scale in Japan, 2026
- Japan Ministry of Health, Labour and Welfare, annual long-term care cost and monthly service figures, fiscal 2025 and April 2026 (no public link)