Hokkaido's Data Centres Run on Power That Cannot Leave

Japan’s grid planner publishes a ten-year demand forecast every January. In the January 2025 edition, comparing FY2024 with FY2034, OCCTO wrote that new data centres and semiconductor plants make demand rise most in three areas: Hokkaido, Tokyo and Chugoku (OCCTO).

Hokkaido’s grid reaches the rest of the country through one ageing interconnector. OCCTO’s note on commercial demand growth names Sapporo redevelopment and resorts. The compute projects arrived separately after 2023.

Nationally, OCCTO’s separate line item for new data centres and semiconductor plants runs from 560,000 kW of peak demand and 3.6 TWh of energy in FY2025 to 7.15 million kW and 51.4 TWh in FY2034, of which data centres are 6.16 million kW and 44 TWh (same document). Hokkaido takes a real share of that against a base that is small to begin with.

What is being built

Rapidus selected Chitose as the site for its advanced logic fab on 28 February 2023 (北海道). The fab produced a working 2nm gate-all-around transistor in July 2025 and targets mass production in 2027 (Rapidus). We covered the chip programme itself in Japan’s bet on chip sovereignty.

SoftBank broke ground on an AI data centre in Tomakomai in April 2025, targeting opening in fiscal 2026. The company describes a site growing to 700,000 square metres and a power intake above 300 MW, running on 100% Hokkaido renewable electricity as a local-generation, local-consumption facility (SoftBank).

In Ishikari, Sakura Internet has run a data centre since November 2011 (石狩市). In April 2024 it announced an expansion of its generative-AI cloud to a combined 18.9 EFLOPS of half-precision compute by the end of 2027, adding roughly 8,000 GPUs on top of about 2,000 already installed, at a scale of ¥100 billion, certified by METI under the Economic Security Promotion Act’s cloud programme (Impress).

Sakura’s Ishikari site uses outside-air cooling, drawing on Hokkaido’s cold air to cool the server rooms (石狩市). That was the original reason to build in Ishikari in 2011, and it is still the cheapest operating advantage the island has. A degree of ambient temperature is worth more per year to a 300 MW campus than most efficiency projects.

OCCTO builds this forecast from connection applications, not signed leases. Its footnote says growth slows after FY2031 and may change with future applications (OCCTO). Some share of the 7.15 million kW is capacity that developers may never build.

The cable south

Japan decided in principle to connect Hokkaido to Honshu with a new subsea HVDC link at the GX Implementation Council on 22 December 2022, with a stated target of construction by FY2030. Hokkaido’s governor recorded the grid planner’s investment estimate at the same time: ¥2.5 trillion to ¥3.4 trillion for the new Hokkaido–Honshu route, plus ¥1.1 trillion for reinforcement inside Hokkaido (北海道). The Japan Sea route is planned at 2 million kW (Team Sapporo-Hokkaido).

Four transmission companies were designated as qualified operators for the Japan Sea route on 26 February 2025. Hokkaido Electric Power Network calls it long-distance subsea cable work with no domestic precedent (北海道電力ネットワーク). The implementation-plan deadline moved from the end of 2025 to the end of 2026 because financing was difficult. The national transmission master plan is costed at ¥6–7 trillion (Nagasawa & Associates).

So the 2030 target rests on a plan not yet submitted for a cable nobody in Japan has laid at this length. Treat FY2030 as an aspiration with a financing question attached.

The 2 GW cable is late

The Ishikari Bay New Port offshore wind farm began commercial operation on 1 January 2024. Its fourteen turbines provide 112 MW and feed through a substation with 180 MWh of battery capacity (JERA). A data centre sits a short distance from it.

More is designated. On 12 May 2023 the government named five Hokkaido sea areas as promising for offshore wind: Ishikari, Ganu-Minami Shiribeshi, Shimamaki, Hiyama and Matsumae (北海道). Two of those, Matsumae and Hiyama, were designated promotion areas on 30 July 2025, at an expected 250,000–320,000 kW and 910,000–1,140,000 kW respectively, both fixed-bottom (Wind Journal).

The two zones hold roughly 1.2 to 1.5 GW of auctionable capacity. The grid must use that power locally or send it about 1,000 kilometres south. Japan’s offshore wind programme has had a hard two years, which we covered in the auction reset. The economics of these zones are still uncertain.

What the 7.15 GW forecast includes

OCCTO’s 7.15 GW national figure is not completed capacity. Regional utilities counted connection applications and the information available when they prepared their estimates. OCCTO’s footnote says growth slows after FY2031 and may change with future applications.

A grid application can come before the developer has every customer, permit and financing condition needed to build. It records demand for a connection. It does not show that the whole campus will open on schedule.

Some of the work is already physical. SoftBank has broken ground in Tomakomai. Rapidus has built its Chitose pilot line. Sakura Internet operates in Ishikari and is adding computing equipment.

The final load will therefore differ from the application total. Grid planners cannot wait for every campus to open before they act. A long subsea transmission line takes years to finance, permit and build. The forecast is best read as the capacity the grid may need to serve, with uncertainty around the timing.

Tomari Unit 3: 912 MW

A 300 MW data centre campus needs power that is firm, not merely clean. Wind on the Japan Sea coast in winter is strong and variable, and 180 MWh of batteries is minutes of a campus that size, not hours.

Hokkaido’s firm capacity question runs through Tomari. Unit 3 is a 912 MW pressurised water reactor that entered commercial operation in December 2009 (北海道電力). The Nuclear Regulation Authority granted the reactor installation change permit on 30 July 2025, the governor stated consent on 10 December 2025, and reported that consent to the government on 18 December 2025 (北海道).

That single unit is comparable in size to the entire Matsumae and Hiyama offshore programme, and it runs at night. Firm power comes from Tomari and imports over the old interconnector. Renewable contracts sit alongside them.

The same shortage of things that can absorb or shift power appears on the storage side, which we covered in Japan’s battery storage interconnection queue.

What changes now

A campus accepts more latency to Tokyo in return for reliable power and cheaper cooling. It also needs long-duration storage, private wires and a contract that connects local wind to local demand.

The planned cable costs ¥2.5–3.4 trillion. Every large load built in Hokkaido reduces the amount of electricity that must travel south. That will change the cable’s use before anyone lays it.

Tech for Impact Summit 2027 is in Tokyo on 18–19 May 2027, co-hosted with the United Nations University. It is invitation-only. Join the waitlist at tech4impactsummit.com/apply.