Japan Scrambles to Grow AI Data Centers Amid Power Crunch

Executives at SusHi Tech Tokyo 2026 said Japan’s effort to expand AI data center capacity is being slowed by limited electricity supply, competition with households for power and regulatory hurdles.

Speakers at the event described rising demand for low-latency services as a reason companies want centers close to Tokyo and Osaka, increasing strain on local grids. Rocky Lee of Taiwan-based Zettabyte said latency drives high electricity use for GPUs. “If you ask an AI a question and get a response 40 seconds later, that’s not an ideal customer or enterprise experience. Power has to be transferred to GPUs, which is where we see the shortage,” Lee warned. He added that AI operators could outbid residential users for supply, and that wholesale prices have risen in U.S. regions with dense data center clusters.

The national government announced an expansion of its GX strategy on April 24 to encourage industrial clusters around renewable energy in regional areas. Officials have not named specific sites, but industry participants expect candidates such as Hokkaido, Tohoku and Kyushu. The plan aims to place large, continuous loads where new clean power generation can meet demand without overtaxing city grids.

GMI Cloud is planning a gigawatt-scale complex in Kagoshima with an estimated cost of $12 billion and a target completion date in 2030. The company’s founder and CEO, Alex Yeh, cited local access to nuclear power and Japan’s role as a fiber-optic hub connecting the U.S. and Asian markets as reasons for the site choice. “Data is sensitive. There’s government data, military data, and enterprise data. You don’t want data situated in geopolitically sensitive areas such as the U.S. and Korea. That’s why Japan matters,” Yeh said.

Legacy firms are moving into data center and power infrastructure. NTT, which operates more than 160 data center sites across all 47 prefectures, announced an AI x OWN initiative on April 27 to redesign networks for real-time AI. NTT President Akira Shimada said the company plans to shift from conventional ICT to infrastructure designed for AI and to raise domestic power capacity from roughly 300 megawatts to about 1 gigawatt by fiscal 2033.

Industry executives criticized Japan’s approval processes and construction norms for slowing delivery. Alex Yeh urged regulators to allow modular data centers-prebuilt 40-foot container units that arrive wired and can be deployed quickly-so projects can finish in six to eight months instead of the 18 to 24 months typical for concrete facilities. He pointed to modular builds used in the U.S. and Taiwan as examples.

Japan currently has an estimated 256 operational data centers, compared with roughly 5,400 in the United States, about 520 in Germany, 500 in the U.K. and roughly 450 in China. Executives said where capacity is sited will affect which industries get AI services first: urban locations reduce latency but strain city grids, while regional clusters offer access to cleaner power but require longer fiber runs and incentives to attract investment.

Speakers at SusHi Tech said regulators’ choices on permitting, grid upgrades and data policies will affect the pace and location of AI deployment across Japan.