Japan's Data-Centre Energy Crunch Concentrates in Two Cities
Japan's data-centre site selection has become an energy problem. Power, land and cooling capacity around Tokyo and Osaka are increasingly difficult to secure, Japan NRG's weekly analysis published Monday (2026-07-06) reported, even as those two cities still account for roughly 90% of the country's total data-centre inventory.3
The concentration matters because the investment pipeline has no intention of slowing. Oracle, Google and Microsoft have been designated official cloud providers by the Japanese government, with hyperscalers committing US$28 billion, or about 4 trillion yen, to the country's digital infrastructure, according to Wood Mackenzie.2 Their preference is deployment within five years. The power grid has other ideas.
Combined-cycle gas turbine projects — the most likely source of dedicated generation for large data-centre loads — require seven to ten years from planning to completion, Wood Mackenzie noted.2 That gap between hyperscaler ambition and generation lead times is the underlying constraint that Tokyo and Osaka are discovering in real time.
Wood Mackenzie projects Japan's data-centre electricity consumption rising from 19 TWh in 2024 to between 57 TWh and 66 TWh by 2034.2 That threefold increase would represent 60% of Japan's entire incremental power demand growth over the decade, with peak load from facilities reaching 6.6 GW to 7.7 GW — roughly 4% of Japan's total peak load, and equivalent to the consumption of 15 million to 18 million Japanese households.2
Japan's current generation mix makes the arithmetic tight. Natural gas accounts for around 32% of power output, coal for 28%, nuclear for 9% and oil-fired generation for 7%, according to data from OilPrice.com.1 The power sector already absorbs 55 to 65% of total domestic gas consumption.1 Any significant incremental load from data centres feeding on the existing grid will first reach gas turbines — putting upward pressure on LNG term demand at a time when JKM, the Asian LNG benchmark, was trading at $16.07 per MMBtu on Monday (2026-07-06).
Japan's LNG position leaves limited headroom. The country imported 66.3 million tonnes in 2025, down 1.5% year-on-year, retaining its rank as the world's second-largest buyer after China.1 Around 98% of domestic gas demand is met by LNG imports, with Australia providing roughly 26 Mt annually, Malaysia 10 Mt, and Russia's Sakhalin-2 at 5.8 Mt.1 There are no pipeline gas alternatives; every cubic metre burned in a new data-centre-serving gas plant must arrive as a cargo.
The hyperscaler rush partly reflects geography. Tokyo and Osaka offer fibre density, financial infrastructure and market access that justify premium siting. Secondary cities where grid capacity may be available require operators to trade off latency and connectivity, a compromise the industry has consistently resisted in the US and Europe.
The structural issue Japan faces differs from either of those markets. The government has addressed near-term supply risk with other tools — releasing around 80 million barrels from strategic petroleum reserves following Middle East disruptions, equivalent to roughly 26 days of domestic oil demand — but gas reserve capacity is structurally thinner and cannot be drawn down the same way.1
What moves the calculus is whether grid operators accelerate interconnection approvals under explicit government pressure, and whether secondary prefectures around Osaka or in Kyushu can attract developer interest on competitive latency terms. The hyperscaler commitments are real. The generation capacity to serve them at the sites operators prefer is not yet approved, let alone built.2