The 2026-2030s DC pipeline — 15 projects verified against primary sources

Start by fixing the project list. Only those whose location and scale could be confirmed from an operator’s official release or a municipality’s siting-agreement announcement or primary reporting are included. Plans still described as “somewhere in the Kanto region” and concepts whose operator cannot be identified are excluded. A survey conducted by Nikkei xTECH in April and May 2026 (approximately 80 companies contacted, 42 valid responses) reported that 24 companies plan 38 new builds or expansions between 2026 and 2030, and that of the 35 with a confirmed location, 13 are in neither Greater Tokyo nor Greater Osaka. The 15 projects here are a subset of that picture, not an exhaustive market list.

The share of regional projects is the thing to hold on to. Even on the market-survey basis, 13 of the 35 location-confirmed projects sit outside Greater Tokyo and Greater Osaka, and within the range this article could trace via primary sources, 9 of 15 are regional. Regional projects also differ in character from Tokyo-area ones. In areas such as Inzai and Shiroi, where data centers have already clustered, construction traffic and lodging demand are absorbed into existing flows. But where a large box goes up in a place almost nobody lives — the Ishikari Bay New Port area, the former Sendai thermal power plant site — the lodging of construction crews lands entirely on the nearest town.

Location assessment — nodes split between “nobody lives here” and “regional hub city”

Seen through a lodging lens, DC node locations split cleanly into two types. One is the coastal industrial site or former power plant site with almost no resident population or businesses around it. The other is the station-front area of a regional hub city that already has commercial and employment density. The first depends entirely on outside capacity to house construction crews; the second competes for slots with existing business demand. That difference changes the product design of anyone entering the market from the ground up.

Fifty-four people live within a 1.5km radius of the Ishikari Bay New Port area. There are 30 establishments and 503 employees. In other words, this is a place to work but neither a place to stay nor a place to live. While a 15MW DC is being built, crews will lodge in central Ishikari or northern Sapporo. Wakkanai has a similar structure: 351 establishments and 2,969 employees within 1km of the station, while the resident population has shrunk to 3,241 and is projected to fall a further 36.9% by 2040. The Soya Green Data Center site is a 9,900 sqm plot adjacent to the Kabaoka Wind Farm, away from the town center.

Satsumasendai is the contrast. Within 1km of Sendai Station there are 698 establishments and 5,948 employees, and within 1.5km a resident population of 17,514. Its 9.4% office-sector share puts Satsumasendai and Yuzawa at the top of the six nodes here, and the functions of a regional hub city are already in place. The projected population decline to 2040, at −12.2%, is also gentler than at the other regional nodes. A 350MW-class DC arriving here in phases makes this the node with the widest gap between demand scale and existing base.

Inzai and Shiroi are best treated as a third type. The resident population within 1.5km of Chiba New Town Chuo Station is 17,569 and is projected to be essentially flat at +0.1% in 2040. Population holds up as a commuter-belt new town, yet establishments within 1km of the station number only 48 with 302 employees, and the office-sector share stays at 2.0%. The DCs sit on industrial land away from the station, so demand arises on a different layer from the station-front commercial and employment cluster. In MLIT’s “2026 Official Land Price” survey (as of 1 January 2026), Chiba Prefecture reported that industrial land in Inzai rose more than 10% year on year. Hokkaido was also reported to be up 1.3% year on year across all uses, a tenth consecutive annual rise. Land pricing has already begun to reflect this demand.

Measuring room stock — 23,357 rooms across 13 municipalities, but few of them next to a DC

Next, count the capacity. The scope is each DC node’s host municipality plus the neighbouring municipalities crews could realistically commute from. Listed rooms are the sum of rooms at properties for which a selling price could be confirmed on an OTA for a check-in date within the last 60 days. Properties whose listings cannot be confirmed (temporarily closed, direct-booking only, small simple lodgings) are excluded, so the figure runs below the true total of rooms in existence. Registered counts from the property master are shown alongside for reference.

The spread in listing rates is what catches the eye. At 97.1%, Inzai sells almost every registered room on OTAs. Its stock is mostly urban business hotels with a year-round product mix. Yuzawa, by contrast, sits at 49.2% and Minamiuonuma at 39.9%. Both have many ski-season pensions, rental villas and small ryokan, which suggests a considerable number of properties stop selling altogether in the off season. Dormant capacity is waiting there to be brought back if a year-round source of demand arrives.

The 53.4% and 56.4% listing rates in Oga and Satsumasendai are also worth noting. Both cities hold stock that includes onsen ryokan and tourist inns, yet close to half of it is not on sale at any given time. DC construction demand — weekday-centric, long-stay and year-round — could be exactly the route that puts this dormant stock into operation.

Twelve months of estimated settled ADR — a 2.38x swing in Yuzawa against 1.06x in Inzai

Now the price level and seasonality. The chart lines up each municipality’s estimated settled ADR (area median) for the twelve months from August 2025 to July 2026. What matters here is less the absolute level than the annual swing. The wider the swing, the greater the levelling effect when year-round demand arrives.

Inzai’s 1.06x swing shows what a DC cluster arrives at. Rates barely move across the year. This is a state in which rooms are filled purely by business demand, with no seasonal tourism wave and little distinction between weekdays and weekends. What Inzai looked like before the DCs gathered is outside this article’s data range, but its present shape is at least underpinned by demand that can be read year-round.

Read the other way, Yuzawa’s 2.38x and Wakkanai’s 2.05x are a measure of headroom. Yuzawa’s June estimated settled ADR is ¥9,700, only just over 40% of its ¥23,200 January level. If weekday-centric, long-staying construction crews land in that trough, the annual RevPAR structure changes. Wakkanai’s trough is ¥6,800 in March, and the Soya Green Data Center schedule — construction from April 2026, operation during 2027 — straddles that shoulder season twice.

Rooms needed per DC project — a coefficient of 7 to 15 rooms per MW

From here it is an estimate. How many workers are on site at the peak of DC construction, and how many of them need a room? No DC project discloses crew numbers directly, so the approach here back-calculates from published total construction costs and construction labour unit rates. Every assumption is stated.

The coefficient obtained from the two real projects is 7.3 to 14.5 rooms per MW of contracted power. The difference in unit investment per MW (¥1.30bn versus ¥0.65bn) shows up directly as the width of that range. The denser and more liquid-cooled the AI specification, the larger the unit investment and the more rooms required. Applying the coefficient to each node gives the following.

Sensitivity analysis — rooms needed when the assumptions are flexed (50MW, ¥65.0bn case)

Even with the assumptions flexed widely, the room requirement for one 50MW-class phase stays roughly within a 400 to 1,100 range. Where it lands within that band depends heavily on how many workers local contractors can mobilise (that is, the outside-area crew share). At nodes where population decline is advanced, such as Wakkanai and Oga, dependence on outside crews is likely to be higher, so it is natural to read those toward the upper end.

And the demand after commissioning should not be forgotten. DCs are described as close to unmanned, but in practice maintenance, network and electrical staff rotate through continuously. Add scheduled maintenance of liquid cooling systems and server expansion or replacement works every few years. Once the construction peak has passed, a mid-sized weekday-centric stay demand remains as a base load. That is the decisive difference from a typical construction project, where demand vanishes at completion.

The gap in category mix — kitchen-equipped inventory tops out in the 8% range

What kind of room matters as much as the room-count gap, and arguably more. Stays tied to DC construction run from several weeks to several months. What suits that segment is apartment-style, long-stay rooms with a kitchen and washing machine, not a one-night business hotel with breakfast. So for the registered properties in each municipality, the depth of inventory that explicitly promotes medium- to long-term stays was tallied from published room names and plan names.

Properties that explicitly state self-catering facilities stay in the single-digit percentages at every node. Satsumasendai and Inzai have none at all, and Yuzawa has six out of 344 properties (1.7%). Yuzawa should hold a substantial amount of rental-villa inventory converted from resort condominiums, yet few of them mention a kitchen in the published room or plan names. Put the other way, this is an area where demand can be captured simply by changing how the offer is described.

Properties with consecutive-night or weekly plans are more common on the Tokyo side — 50.0% in Inzai and 42.6% in Narita — and less common regionally, at 13.0% in Ishikari, 16.7% in Ichikikushikino and 17.4% in Minamiuonuma. Where DCs and logistics have already clustered, in Inzai and Narita, long stays have been productized. That difference is exactly the headroom at regional nodes. On how many nights a consecutive-night plan should start from and what discount rate to design, our article Japan Hotel Long-Stay Discounts: 15% at 7 Nights, 20% at 30 Nights sets out the measured market levels.

Properties already coming on stream

Openings aimed at industrial infrastructure demand have in fact already begun. In Oga, Akita, “Hotel Kinoshita Akita Oga Ekimae” (ホテル木下 秋田男鹿駅前) opened on 5 March 2026. Of its 161 rooms, 10 are long-stay double rooms (19 sqm, equipped with a washing machine and microwave), and the location has been reported as positioned for the start-up of offshore wind power. That single property accounts for well over a quarter of Oga’s 566 listed rooms.

Hokkaido is moving quickly too. Ishikari added a 175-room property in October 2025, and Tomakomai saw a 50-room property explicitly aimed at workers open in December 2025. Yuzawa, Niigata, saw a 220-room property in February 2026 and an 83-room property in September 2025. None of these are scaled on tourism demand alone. The movement covered in our earlier article Chitose & Tomakomai Worker Lodging — 7 New Hotels, 738 Rooms in 2026 can be read as spreading across a wider geography.

Looking at planning data on a building-confirmation basis from MLIT’s “Building Construction Statistics Survey”, however, no hotel-use applications can be confirmed in Akita, Kagoshima or Niigata, and the applications in Chiba and Hokkaido are concentrated in places away from the DC nodes — Funabashi, Urayasu, central Sapporo, Kutchan. Because building confirmation is normally filed one to two years before opening, this should be read as a lower bound on the confirmed pipeline at this point in time, with the count and room numbers expected to rise as further applications come in. Read the other way, for operators targeting openings from 2027 onward, there are still seats free. That said, new supply in that period is itself set to thin out on construction-cost grounds.

Conclusion — how to go after year-round demand

To put it in order, lodging demand originating from DCs has three layers. First, the peak demand during construction — 7 to 15 rooms per MW of contracted power, 364 to 725 rooms for a 50MW-class phase. Second, a year-round base load from maintenance and expansion staff after commissioning. Third, business travel from the industries a DC cluster draws in: power, telecoms, cooling equipment, IT vendors. Judging the opportunity as “one-off” by looking only at the first layer means missing the second and third. Inzai’s 1.06x annual ADR swing is what those three layers look like once stacked.

For a regional node, this demand does not compete with tourism. Weekday-centric, long and year-round, it sits exactly on the other side of tourism demand, which concentrates on weekends and high season. June in Yuzawa, March in Wakkanai, February in Oga — the months where rates sink lowest are the ones with slots left to fill. DC construction demand overlays those troughs neatly.

For developers, the significant point is that there is no need to build the business case on visitor numbers in municipalities with thin existing lodging demand. For a feasibility assessment, take the operator’s published commissioning year and contracted power, derive a range of rooms needed, and lay it over existing listed rooms. The coefficient here should serve for that first-pass screening. From there, the specifics come down to local contractors’ mobilisation capacity (the outside-area crew share) and how the construction schedules overlap.

Frequently asked questions

Q. How is lodging demand from data center construction different from semiconductor plant construction demand?

The dispersion of locations and the time horizon differ. Semiconductor clusters concentrate in specific regions in search of talent and supplier density, whereas DCs seek electricity, renewable energy and a cool climate, so they locate at scattered nodes such as Ishikari, Wakkanai, Oga, Satsumasendai and Yuzawa. Demand also does not disappear at completion: maintenance, network and electrical staff rotate through continuously once operations begin, and server expansion and replacement works run every few years.

Q. How should the “7 to 15 rooms per MW of contracted power” coefficient be used?

As a rule of thumb for first-pass screening. Multiply the contracted power an operator publishes by the coefficient and compare it with the municipality’s listed rooms, and you get a rough read on how tight capacity is. Note, though, that the coefficient here is back-calculated from the two projects with published total costs, and the difference in unit investment per MW (¥0.65bn and ¥1.30bn) translates directly into the width of the range. Assume the denser and more liquid-cooled the AI specification, the closer to the upper end.

Q. Why do listed rooms differ from the room count in the property master?

Because listed rooms count only properties for which a selling price could be confirmed on an OTA for a check-in date within the last 60 days. Temporarily closed properties, those selling direct only, and those that stop selling in the off season are not included. Yuzawa’s low 49.2% listing rate reflects the large number of properties there that operate mainly around the ski season.

Q. What should an existing property do first to go after this demand?

Start by stating self-catering facilities and length-of-stay conditions in room names and plan names. In this tally, properties that state a kitchen or similar in public information stayed in the single-digit percentages at every node. A considerable number are likely to have the facilities without describing them. A permanent seven-night-plus plan, carving out rooms with a washing machine, and a room-only price tier are all moves that can be made without building anything.