Japan spent ¥16 billion putting its cities in 3D. Did it pay off? PLATEAU's report card

Japan has spent six years and well over ¥16 billion turning its cities into 3D. Has it been worth it?

PLATEAU (pronounced "pla-toh") is the Japanese government's project to build a free, open 3D model of the country's cities. Since it launched in fiscal 2020 (Japan's fiscal year runs April to March), the national budget for it has topped ¥16 billion, roughly $107 million at ¥150 to the dollar. By the end of fiscal 2025, 329 cities had been modeled. Every building comes with its use, its height, even how deep it would flood, and anyone can download it for free.

So why does the government's own spending watchdog keep marking it down? Every national program in Japan files a public scorecard each year under the Administrative Project Review. In both 2025 and 2026, the reviewers told PLATEAU it needed a "fundamental improvement of the entire project." The reason: nobody can say, in numbers, what it has actually done for society (Administrative Project Review findings).

The map keeps getting bigger. But when we counted the public data city by city, we found far more places still sitting on their very first model, never updated, than we expected.

Here's PLATEAU's report card: what the government was after, which projects delivered and which fizzled, where the money went, how Japan stacks up against Europe and Singapore, and what our own count says about 3D cities that quietly go stale.

MLIT's City Bureau started PLATEAU in fiscal 2020. In year one it modeled 56 cities and released them in March 2021, together with PLATEAU VIEW, a viewer that runs in the browser (MLIT's launch announcement).

The seed money, about ¥2 billion, came from a fiscal 2020 supplementary budget passed as part of Japan's COVID-19 stimulus (FY2021 Administrative Project Review sheet). A supplementary budget is extra spending approved mid-year on top of the regular budget. The pandemic had given "digitize city planning" sudden momentum.

On paper, the goal was to recreate cities as data and unlock new value for urban planning. MLIT's original 3D City Model Introduction Guidance lists three ambitions: "optimized, sustainable urban development," "human-centered development with citizen participation," and "agile, responsive urban development."

That sounds abstract, but the idea is simple. Japan's city planning maps were flat. Add height and building-level detail, and you can test things in 3D before they happen: how deep a flood would get, where the sun falls, what a skyline looks like, what a redevelopment would change.

The name comes from A Thousand Plateaus, the book by French philosophers Gilles Deleuze and Félix Guattari, with the plateau standing for a place where different kinds of knowledge meet (PLATEAU's official About page). The data is free to use, even commercially, as long as you credit the source (PLATEAU's site policy). The government builds it, and anyone can use it. That's PLATEAU's defining feature.

Coverage grew from 56 cities to about 130 by the end of fiscal 2022, about 200 by fiscal 2023, about 250 by fiscal 2024, and 329 by fiscal 2025. The government's Digital Garden City Nation strategy commits to 500 cities by fiscal 2027.

The 3D map itself isn't the end goal. In April 2026, Japan began reviewing building permits using BIM, the digital design files architects work in. In its long-term Building and City DX vision, MLIT has laid out a plan to feed those design files back into PLATEAU, and to link everything through a "real estate ID" that tags each property. The real ambition is a shared data backbone that ties together buildings, land and government paperwork.

So what has six years actually produced? The government has run about 100 pilot projects, which it calls use cases (MLIT's PLATEAU Vision 2026). Here are the ones with hard numbers attached.

Nine in ten permit questions handled, 80% of inspection work gone. Where PLATEAU is used to take tedious work off people's plates, the results are clear.

It also turns up in places you wouldn't expect. The music video for "Kizuna no Kiseki" by MAN WITH A MISSION and milet, the opening theme of Demon Slayer: Swordsmith Village Arc, credits PLATEAU in its description. The ministry says it used the data for Tokyo's Minato Ward (CGWORLD article). As of September 29, 2026, the video had passed 150 million views. And when the Saitama Shintoshin district was rebuilt inside Fortnite for an event, 694 people showed up (PLATEAU use-case report).

The misses are on record, too.

• Self-driving car positioning (Numazu, from fiscal 2022): A pilot tried to locate cars by matching their camera footage against the 3D model. Differences in brightness and contrast between the CG and real video meant some scenes simply wouldn't match (PLATEAU pilot report).
• Estimating disaster damage (Kanazawa): A pilot combined satellite radar images with the 3D model to estimate damaged buildings. The imagery covered only 22% of buildings, short of the 30% target, and results took a week after the disaster to arrive (PLATEAU pilot report).
• Rooftop solar estimates (Kaga and Osaka): Estimates landed within 4–12% of actual output, yet only 40% of city staff said the tool was easy to use (PLATEAU pilot report).
• A commercial spin-off: In March 2022, insurer Tokio Marine & Nichido and partners announced a "3D virtual city flood simulation model," aiming to offer it to local governments nationwide during fiscal 2023 (Tokio Marine & Nichido's joint announcement). We could find no announcement that it launched in that form.

Local governments are candid about why. A report from a September 2026 meeting of municipal staff, published by MLIT, lists their frustrations: it's hard to explain to the finance department why they need 3D at all, and when the person in charge rotates to another post, nobody picks it up. Who keeps using the data, and for what job? That, more than the technology, is what separates the successes from the stalls.

PLATEAU's budget has two halves: research the national government runs itself, and subsidies that help cities build their own models. Here's the national funding by year, from MLIT budget documents (in ¥100 million, so 20.05 means about ¥2 billion).

Sources: MLIT City Bureau budget documents (FY2023 budget summary, FY2025 budget summary, FY2026 budget summary and FY2027 budget request) and the Administrative Project Review sheet, compiled by DATA WORLD (all in Japanese). The FY2021 regular budget and the FY2023 supplementary budget are left out of the total because PLATEAU's share can't be separated out.

Add up fiscal 2020 through 2026, and PLATEAU's own budget comes to about ¥16.3 billion. Count the part of the 2023 supplementary budget we couldn't separate, and it could be as much as ¥18.1 billion.

About ¥50 million (roughly $330,000) per city is only a rough average. The budget also pays for tools like PLATEAU VIEW, about 100 pilots and training. It leaves out the other half that cities pay themselves when they take a subsidy. MLIT's own sample cost estimates for building a model alone run from about ¥5.5 million to ¥56 million per city, depending on its size.

So what's the return? PLATEAU Vision 2023, published in November 2023 by MLIT's Urban Policy Division, says PLATEAU is "expected to generate about ¥114 billion a year in economic impact" by 2027. That figure comes with two big caveats.

First, it's a forecast of future market size, built by taking a global digital-twin market projection and multiplying it by Japan's share and the share for cities. It doesn't measure any benefit that actually happened. Second, it was calculated by Accenture, PLATEAU's main research contractor. In fiscal 2025, Accenture was the top recipient of PLATEAU research contracts, at about ¥330 million (Administrative Project Review spending records). The ¥114 billion figure doesn't appear in the 2026 roadmap.

Which brings us back to the spending review. For a program that bundles PLATEAU with other pilot urban-development research, the review team handed down "fundamental improvement of the entire project" in 2025 and again in 2026. The 2026 finding reads:

Set outcome measures that can concretely and quantitatively capture what effects and benefits the program has brought to society. ── Administrative Project Review team (2026 review sheet)

Sure enough, every target on the scorecard counts outputs: cities modeled, municipalities subsidized, people trained. Not one measures how often the data is downloaded, or which companies use it for what. In its response, the ministry itself concedes that business use "requires further effort."

Look abroad, and PLATEAU's position comes into focus.

• UK: For NUAR, a national map of underground pipes and cables, the government projects £480 million a year in benefits against about £230 million in costs over ten years (UK government impact assessment). But that's an up-front estimate resting on assumptions such as preventing 30% of accidental strikes on buried pipes, and access is limited to utilities and contractors.
• Netherlands: A research group and the national mapping agency publish 3D data for more than 10 million buildings for free, refreshed once a year (3DBAG documentation, Kadaster's 3D data page).
• Germany: 3D data for about 58 million buildings is updated every year. The national dataset costs money, but states such as Bavaria give theirs away (Bavarian Surveying Administration's data page).
• Singapore: Virtual Singapore, often called the first national digital twin, was a five-year, S$73 million research program (National Research Foundation Singapore's program page).

The UK narrowed the use and ran the numbers before building. The Netherlands and Germany refresh every building every year. What sets PLATEAU apart is that it offers city-scale data, down to flood scenarios, free to anyone. What it hasn't caught up on is measuring its impact and keeping the data current.

Is the data being kept fresh? We pulled PLATEAU's public data catalog through its AI gateway, the PLATEAU MCP server, and counted when each area's building data was first created and when, if ever, a newer edition came out.

There are 448 areas with building data. Folding the wards of Japan's largest cities into their parent cities leaves 306 municipalities. Of those, 79 got their first building data in fiscal 2020–2022. 58 of the 79 (73%) have never published a newer edition.

Narrow it to the 38 municipalities modeled in fiscal 2020, PLATEAU's first year, and 31 are still on that 2020 version. They include Sapporo, Kitakyushu, Naha, Matsuyama, Fukuyama, Yokosuka, Kashiwa, Koriyama, Iwaki, Okazaki, Toyonaka, Takatsuki and Kurume.

On a separate public portal, the G-Spatial Information Center, Kashiwa's data was likewise a single 2020 edition. MLIT's own figures, in its Building and City DX vision, show the same pattern: compared with the number of newly modeled cities, updates are rare, 88 in total from fiscal 2022 through 2025. That lines up closely with our count of municipalities with two or more editions (90).

An old edition isn't automatically useless. Some neighborhoods barely change. But building uses and similar details come from the City Planning Basic Survey, which the law calls for roughly every five years (MLIT's Building and City DX vision). Would you pick a store site or plan an evacuation on data older than that? It takes some nerve.

The count turned up one more pattern. Look at the companies credited for each municipality's latest edition, and 285 of the 306 (93%) list at least one of Japan's four big aerial-survey firms: Kokusai Kogyo, PASCO, Aero Asahi and Asia Air Survey. Survey from a plane, finish by hand. That way of building is expensive, and it's one reason updates stall.

The government says as much. PLATEAU Vision 2026 sets a goal of a "100% update rate" by the end of fiscal 2032, meaning every city whose source data is five or more years old gets refreshed.

The ace up its sleeve is AI. A new method builds 3D models automatically from satellite imagery. For 10 square kilometers with roof shapes included, it's expected to cut the time from about three to six months to about one to two hours, and the cost from about ¥30–40 million to about ¥3–4 million (MLIT's paper on improving the update cycle). Accuracy drops a little, but the ministry has formally adopted it as "Model B," going so far as to write that "there are cases where a high-precision model isn't needed." The plan is to roll it out in 10 cities in fiscal 2026 and nationwide from 2027. The government has even published the training data for the AI that does the generating (MLIT's dataset on the G-Spatial Information Center).

PLATEAU set out to model 300 cities by fiscal 2025 and reached 329. On building, it beat its target. And there are real wins on the ground: a system answering nine in ten permit questions, inspection work cut by 80%.

Whether the program as a whole has paid off, though, can't be judged today. There are no usage metrics, and the ¥114 billion figure is a contractor's forecast. My verdict: it hasn't paid off yet, and there's no way to prove it has. For a ¥16 billion program, the first step is to publish download counts and what the data is used for. That alone would change the debate.

What surprised me most in our count: eight in ten of the cities modeled in year one were still on their original version six years later. A map starts going out of date the moment it's made. The Netherlands and Germany refresh every building every year. In Japan, each city rebuilds its model on a subsidy, by hiring one of the big survey firms. That puts every update in the tens of millions of yen.

That's why the plan to cut costs to a tenth with satellites and AI matters well beyond the tech. Make updates cheap, and PLATEAU's whole profit-and-loss sheet changes. With the fiscal 2027 request doubling, what the ministry really needs to show isn't the number of cities. It's the number of updates, and how often the data gets used.

Counting update status across 306 municipalities, we never once opened a map. An AI gateway handed us the catalog and a script did the counting. In its Building and City DX vision, MLIT has said that from fiscal 2026 it will prototype an AI gateway of its own, serving 3D city models, land registry maps and more.

As a map for people to look at, PLATEAU was heavy and hard to handle. To an AI, it's a beautifully readable record of a city, every building tagged with its use and its disaster scenarios. If AI brings into everyday work the data that people never quite managed to use, PLATEAU may pay for itself sooner than anyone expects.