Japan promised its Government Cloud would cut local IT costs by 30%. Its own estimate says 1.8x

"Moving city-hall systems to the cloud should make them cheaper, right?"

That was the first question DATA TIMES editor-in-chief Kaito Ogasawara asked when he started digging into this story. For those of us who work with AI and data, the cloud is how you run things cheaper and faster: you pay for what you use.

Japan's government thought so too. It set out to rebuild the core IT systems that every city, town and village runs — the resident registry, local taxes, long-term care and more — to one national specification, and to move them onto a national "Government Cloud." Once that was done, running costs would be at least 30% lower than in fiscal 2018. That target sits in a policy the Cabinet adopted in October 2022.

The Digital Agency's own estimate now points the other way: 1.8 times.

For the systems that will have migrated by the end of fiscal 2026 (March 2027), running costs will rise from about ¥140 billion before the move to about ¥250 billion after it, the Digital Agency estimates. That is roughly ¥110 billion more — about $730 million at ¥150 to the dollar.

At the level of a single city, the numbers get starker. In one example the agency has shared, a "core city" of about 270,000 people expects the annual cost of running its resident-registry, tax and other systems to go from ¥208 million to ¥784 million. That is 3.8 times as much.

So why did the cloud make things more expensive? Who pays for the difference? And is the whole plan a failure? We went through the government's own documents, local governments' estimates and examples from abroad — and then counted every municipality in Japan ourselves.

WHAT IS

Local system standardizationStandardization of Local Government Information Systems

Japan's plan to put 20 core services at all 1,788 local governments onto one national standard and move them to the cloud

1. 2021.05Law passedA new law required local governments to bring 20 core services, from resident records to taxes, onto one national standard
2. 2022.1030% cut promisedThe Cabinet's basic policy set a goal: running costs at least 30% below fiscal 2018 after migration
3. 2025.06Cost measures adoptedAfter local governments warned of soaring costs, the Digital Agency adopted a package of countermeasures
4. NOW29% missed deadlineBy the March 2026 deadline, 10,013 systems, or 29.1% of the total, had not migrated

Officially called the "standardization of local government information systems," this is a national plan run by the Digital Agency and the Ministry of Internal Affairs and Communications. A 2021 law requires all 1,741 municipalities and 47 prefectures to bring the systems for 20 services — resident registration, family registers, local income and property taxes, national health insurance, long-term care, child allowances and more — into line with national specifications.

That covers 34,366 systems across the country. Each local government had bought its own from vendors and customized it over the years. The plan was to replace them all by the end of fiscal 2025 (March 2026), and the destination the government prepared was the Government Cloud, which the Digital Agency contracts for.

Getting a copy of your resident record, receiving a tax notice, having a child allowance paid into your account: behind almost every routine dealing with a Japanese city hall, it is these 20 systems doing the work.

There are three reasons to check the numbers now.

First, the deadline has passed. The government's migration support period ended in March 2026, and in June the Digital Agency published which systems made it and which did not. Second, the costs have reached the national budget: a subsidy for the extra running costs started in fiscal 2026, and a larger amount has been requested for fiscal 2027. Third, the law turned five in September 2026, and the government plans to review how it has worked.

A cloud bill can go up or down depending on how you use it. As people who read our own cloud invoices every month, we know moving a system is not the same as making it cheaper. Here is the report card: the government's goals, the gap between plan and reality, the money, the comparison with other countries, and our own count.

For decades, Japanese local government IT was made to order.

Two towns could run the same resident-registry software and still have different screens and different forms, because each bought a vendor's package and then had it adapted to its own way of working. Customization makes life easier for the people using the system. But every time a national rule changes, every local government has to pay to update its own version.

When the child allowance changes, for example, all 1,741 municipalities each ask their vendor for a fix and each pay for it. The interior ministry's own material says this customization has left local governments with a heavy burden of one-off work on maintenance and legal changes.

So in 2021 Japan passed the standardization law. The 20 services must follow national specifications. If there is one specification for the whole country, a rule change can be built once, by vendors, on the standard. One shape for every town, and one fix for every change: that is the core idea.

The second pillar was where the systems live.

Until now, local systems sat in server rooms at city hall or in vendors' and regional data centers. Groups of towns sharing one system in a joint "municipal cloud" had spread to 713 local governments by April 2021.

The government wants to bring all of this onto the Government Cloud. Spreading data across several regions of Japan keeps systems running through a major disaster, and one set of security settings designed by the Digital Agency can be applied to every system. A Cabinet Secretariat paper from April 2025 lists disaster resilience, more efficient administration and "advanced, uniform security" as the reasons to move.

And then there was the money. With one standard and shared cloud infrastructure, the government reasoned, the cost per town should fall. That expectation is why the 2022 basic policy wrote in a goal of cutting running costs by "at least 30%" from fiscal 2018.

Japan's population is shrinking, and so is the supply of both local civil servants and vendor engineers. The Digital Agency's countermeasures paper says it has reached the point where local governments and vendors maintaining their own systems separately is unsustainable "in terms of both people and finances." Standardization was designed with that limit in mind.

So what happened to the money? Here is everything we could verify, starting with the 1.8 times figure.

One caveat first. The 30% cut and the 1.8 times are not measured the same way: the target compares with fiscal 2018, while the estimate compares with costs just before migration. If costs moved between 2018 and the eve of migration, the comparison shifts. Even so, the direction is clear from the government's own estimate: instead of falling, costs are set to nearly double.

Core cities: 2.3 times on average, and only 2 of 59 expect savings

Local governments' own figures look even worse.

Japan's "core cities" are municipalities of 200,000 people or more that take on extra powers. In January 2025, their association put the results of its own survey into an urgent request to the national government. Before migration, running costs averaged ¥338 million per city. After migration, they averaged ¥684 million. The average city-by-city multiple was 2.3 times; more than half of the cities expected costs to at least double, and the highest multiple was 5.7.

A breakdown of the 59 cities that answered appears in a national meeting paper from April 2025.

Tokyo reached a similar conclusion. In May 2025 the governor and the associations of Tokyo's wards, cities, towns and villages sent a joint request to the national government. Even with the cloud discounts the Digital Agency cites applied in full (20% for volume and 33% for long-term commitments), costs across Tokyo would rise about 1.6 times: about 1.5 times in the 23 wards, 1.7 times in the other cities and 2.0 times in towns and villages. The smaller the government, the steeper the rise. The National Association of Towns and Villages told the government in April 2025 that in many towns costs would be "several times" what they were before.

The government has run its own trials, too. Since fiscal 2021 the Digital Agency has worked with "early migration" governments such as Kobe, Morioka and Suzaka to estimate the cost of moving to the Government Cloud. Its March 2026 report compared five years of running costs with simply renewing the existing systems at the same scale.

Source: compiled by DATA TIMES from the Digital Agency's "Verification Projects on the Government Cloud in FY2024" (March 2026), p. 18. The estimates cover the parts of each government's systems that moved to the cloud early, and assume measures such as reviewing network lines and trimming cloud use.

Even among the governments that prepared most closely with the national agency, not one reached a 30% cut. Only Morioka and Sakura saw costs fall. The biggest rise, 51%, came at Misato and Kawajima, two towns that had shared their systems. The Digital Agency's analysis is that governments that had already kept costs down by sharing hardware and software in a joint municipal cloud have less to save from the move, and are more likely to see costs rise.

Five reasons the bill grew

Why does moving to the cloud raise costs? The Digital Agency's countermeasures paper sorts the reasons into structural factors, service improvements and external factors, with more than ten items in all. In plain terms, they come down to five.

1. New costs that did not exist before

A dedicated network line from city hall to the Government Cloud. A contract with an "operations support" provider to manage cloud settings and split the bill. Neither existed before the move. In City A, these two alone added ¥64 million a year. And when some systems move to the cloud while others stay behind, a city ends up paying to maintain both the old home and the new one.

2. Systems were moved as they were, to beat the deadline

Clouds get cheaper when you use them well: switch servers off at night and on weekends, rent only what you need, sign up for long-term discounts. But to meet the deadline at the end of fiscal 2025, many systems were moved onto the cloud without being rebuilt. The Digital Agency itself lists among the causes of higher costs that packages and operations "have not been fully optimized for the Government Cloud".

3. Some of it buys better service

The Government Cloud stores data in a second region in case of disaster and applies the Digital Agency's high security standards to every system. Matching the national specifications also added features to the systems themselves. In one example, "Town C," a town of about 10,000 people, the work of folding and sealing forms in the new standard layouts increased, pushing up even the cost of outsourced printing. Better service shows up on the bill.

4. A weak yen and rising prices

Amazon Web Services, the most widely used provider on the Government Cloud, prices in dollars, and local governments pay in yen. According to the Digital Agency, the exchange rate went from about ¥110 to the dollar in 2018 to about ¥150 in 2025. Corporate prices have risen 20% to 25% since 2020, and engineers' wages have gone up too.

5. Too little competition

The surest way to meet a deadline is to hire the vendor that already knows your system. With no realistic way to switch, local governments often had to accept the quote they were given. At a national-local meeting in March 2026, local representatives said some vendors refuse to break down their costs, citing "trade secrets," so efforts to check estimates stall.

Blaming the vendors alone would miss the point. With every local government in Japan migrating in fiscal 2025, vendors discovered that the move and the period right after it needed far more engineers than planned, and they ran short. Of the systems that missed the deadline, 95.8% were delayed because vendors lacked the people. Replacing everything, everywhere, at the same time, with the same pool of engineers: the plan itself was built in a way that invites both higher costs and delays.

The money splits into two kinds: the one-off cost of moving, and the running costs that come every year after.

The national government pays the full cost of moving. It set up a dedicated fund and, by the fiscal 2025 supplementary budget, had put ¥774.2 billion into it, according to the interior ministry. To cover the systems that are running late, the fund was also extended to the end of fiscal 2030.

The harder question is running costs. The national position is that these are "basically borne by each local government," since they replace what towns already spend on their current systems.

Still, 1.8 times is more than local budgets can absorb on their own. For fiscal 2026, the government split the roughly ¥110 billion increase into two parts:

• About ¥40 billion: the permanent increase driven by outside factors such as wages and prices, covered by adding to local allocation tax grants according to each government's migration progress
• ¥70 billion: the temporary increase right after migration. The national government pays half (¥35 billion) as a subsidy, and the local share is also covered by an increase in fiscal 2025 grants

The word to watch is "temporary." The subsidy was set up for a single year, and nothing was decided for fiscal 2027 onward. Then in August 2026, the Digital Agency requested ¥47.5 billion in national funds for the same subsidy in fiscal 2027, covering ¥95 billion of eligible spending.

A subsidy for a temporary increase is being asked to grow 1.36 times in its second year. The same document says the subsidy will be reviewed every year and tied to how well each government meets its operations optimization plan. A government spending review also flagged the need to "end dependence on subsidies" and build a system that helps local governments and vendors stand on their own.

National funds, grants or local budgets — in the end, it all comes from taxes. So the answer to "who pays?" is: the national and local governments split it, and taxpayers foot the bill.

The government still lacks a clear view of local costs

There are efforts to bring costs down. The Digital Agency has started helping local governments check their quotes, and 331 had signed up by the end of February 2026. Among those whose checks were finished, the agency found room to cut Government Cloud fees by about 30% on average (16% to 60%). Switch off test servers at night and on weekends, shrink oversized servers, use long-term discounts — the same moves any company makes to cut its cloud bill.

But in the same paper, only 41 checks were complete, and 166 governments had yet to send their documents. At a March 2026 meeting, a Digital Agency official said that "at present, the Digital Agency does not have a full grasp of local governments' cost information."

A local representative at the meeting put it bluntly:

Our system running costs are expected to more than double compared with before migration. For local governments with small populations and weak finances, it is extremely difficult to keep paying for this increase out of general revenue indefinitely. ── Summary of proceedings, National-Local Digital Common Infrastructure Council working team (March 3, 2026)

How other countries have handled the cost of public-sector IT

Other countries have tried to consolidate government systems or move them to the cloud. Three offer useful lessons on cost.

• United States: The federal government launched "Cloud First" in 2010 and replaced it with "Cloud Smart" in 2019. The new strategy tells agencies to weigh the long-term inefficiency of moving applications "as-is" against the cost of modernizing them first, and says cloud plans should not be built around anticipated savings. Shortly before, the Government Accountability Office had found that agencies were not consistently tracking cloud savings.
• Denmark: To break one vendor's (KMD's) long-standing hold on many key municipal systems, KOMBIT, a company set up by the association of local governments, put the systems out to tender on behalf of every municipality. In a 2012 letter to mayors, the association, KL, expected the joint tender to cost at least 25% less than staying with the monopoly. All 98 municipalities joined. But a 2019 Norwegian review still described the 25% as a target.
• South Korea: On September 26, 2025, a lithium-ion battery caught fire at the Daejeon data center of the National Information Resources Service, which runs the government's IT. According to The Korea Herald, 647 systems went offline at once and 96 were destroyed. A cloud center in Gongju meant to keep services running elsewhere in a disaster had been finished in 2023, but its launch was held up for lack of budget.

The US wrote the inefficiency of moving systems unchanged into policy. Denmark bet that municipalities buying together would pay less. South Korea paid the price for skimping on disaster backup. Part of what Japan pays for the Government Cloud is insurance for keeping systems running somewhere else — exactly what Korea saved on. What Japan has not yet harnessed is the Danish lever: local governments banding together to push prices down.

After cost comes the migration itself. How many towns had finished by the March 2026 deadline?

According to the Digital Agency, 10,013 of the 34,366 systems (29.1%) missed the deadline. The 1,015 local governments with at least one such system make up 56.8% of all 1,788. At the end of January 2025, the count had been 2,989 systems (8.6%); it kept climbing as the deadline approached.

Those are the official figures. We took the list of late systems by government that the agency released at the same time (1,015 governments × 21 services) and matched it, municipality by municipality, with the interior ministry's resident-register population data as of January 1, 2026, to see how many people the delays touch.

82% of Japan's people live where migration isn't finished

Of Japan's 1,741 municipalities, 978 (56.2%) had at least one system that missed the deadline. That is just over half by count. By population, it is 101.44 million of Japan's 123.77 million people. As of March 2026, 82.0% of the population lived in a municipality whose migration was not finished.

The gap between the share of municipalities and the share of people comes from one thing: the delays are concentrated in big cities.

The bigger the city, the further behind

Counting by type of local government, here is the share with at least one late system.

All 20 of Japan's ordinance-designated major cities had late systems, along with 22 of Tokyo's 23 wards and 58 of 62 core cities. The share falls as places get smaller: 65% of other cities, 46% of towns and 37% of villages.

Big cities also have more services left to move. Of the 21 services on the list (the 20 services plus shared functions), the median major city still had 15 to migrate — 70% of the total.

Even Hamamatsu, a major city that moved 16 of its 20 services on time, expects to finish the remaining four, including child allowances, only by the end of fiscal 2027. Its costs are heavy too: according to an opinion by the city's audit commissioners in September 2026, the city's own optimization plan puts running costs at ¥2.11 billion after migration, against ¥910 million before.

Big-city systems have been tailored to those cities for years. The Digital Agency notes that many major cities run systems built from scratch or heavily customized big-city packages. The more customization, the heavier the work of matching the standard.

By service, the most delays were in long-term care and child allowances (622 governments each) and child and childcare support (621). The Digital Agency says these welfare services were hit by delays at the main vendors that specialize in welfare systems. Family registers, which were already largely standardized and have few vendors, had far fewer delays (135).

1.8 times is still an interim number

Our count shows one more thing.

The 1.8 times estimate covers systems that will have migrated by the end of fiscal 2026. The Digital Agency expects about 90% of all systems to be done by then. Much of the remaining 10% belongs to big cities with deeply customized systems. As of October 2025, according to the interior ministry, only 38 of the 265 late systems in major cities were scheduled to move during fiscal 2026. Back in May 2025, the Digital Agency also wrote that many major cities might not yet have estimates for post-migration running costs.

Somewhere in the towns where 80% of Japan's people live, migration is still under way. As of November 2026, no one yet knows the true total of post-migration running costs.

By the deadline, 70.9% of systems had migrated. For an unprecedented project to replace more than 30,000 systems across the country at once, I think that is a respectable result. The share is expected to reach about 90% by the end of fiscal 2026.

The 30% cut, on the other hand, is a long way off. The national estimate is 1.8 times, the core-city average is 2.3 times, and not one of the early-migration governments that prepared with the agency reached a 30% cut. But there is also no yardstick yet to judge whether the plan has paid off. The target is measured against 2018, and the government itself does not have a full picture of local costs. My verdict: it hasn't gotten cheaper, and there is no way yet to measure whether it will.

Calling it all waste would be wrong. Keeping data in a second region and giving every town the same level of security matters, and Korea's fire shows how much that insurance is worth. Standardization should also reduce the work of updating systems town by town every time a rule changes.

What matters is splitting the extra ¥110 billion into "what we pay for better service" and "what we pay because we rushed." The first can be justified. The second is money that can be cut after the move. In the government's own breakdown, ¥70 billion of the ¥110 billion is classed as a "temporary increase." We think the room to cut is far from small.

The tools to cut a cloud bill already exist: switch off at night, right-size servers, use discounts. The Digital Agency's checks found room to cut cloud fees by about 30% on average. What is missing, I think, is the ability for towns to compare prices with each other.

The Digital Agency is preparing to share data so local governments can compare their cloud fees with those of governments of similar size. Danish municipalities tried to break a monopoly by buying together. The moment Japanese local governments can really push prices down is the next time they replace these systems, now built to one national standard. If the specification is the same everywhere, you can switch and you can compare. Whether standardization pays off, I believe, will be decided in that second round of buying.