Japan’s Liquid‑Cooling Push Targets AI Data‑Center Power Bills

Business news

Japan’s leading industrial players – including Fuji Electric, Nidec, Daikin and Mitsubishi Corp. – are accelerating development of liquid‑cooling systems for artificial‑intelligence (AI) data centres. The move follows a series of pilot projects launched in 2024 that demonstrated up to 40 % reduction in power‑usage effectiveness (PUE) compared with conventional air‑cooling. Fuji Electric announced a partnership with a Tokyo‑based cloud provider to retrofit 12 MW of AI‑training racks with its new two‑phase coolant loop, targeting a PUE of 1.15 by 2027. Nidec, leveraging its motor‑technology expertise, unveiled a high‑density immersion‑cooling chassis that can sustain 500 kW per rack while keeping coolant temperatures below 30 °C.

Market context

The global AI‑compute market is projected to consume 200 TWh of electricity annually by 2030, according to a recent International Energy Agency (IEA) forecast. In Japan, AI workloads now account for roughly 15 % of total data‑centre demand – a share that is expected to double as enterprises adopt large language models and generative AI services. Conventional air‑cooling architectures typically operate at a PUE of 1.5‑1.7, meaning that for every megawatt of compute power, an additional 0.5‑0.7 MW is spent on cooling and ancillary systems.

Liquid cooling offers a two‑fold advantage: it removes heat more efficiently, allowing higher chip power densities, and it reduces the need for energy‑intensive chillers and fans. The technology also aligns with Japan’s broader energy‑policy goals. After the 2023 power‑price surge, the Ministry of Economy, Trade and Industry (METI) set a target to cut data‑centre electricity consumption by 30 % by 2030. The government’s “Green AI” subsidies, worth ¥120 billion (US$820 million) for 2025‑2028, are earmarked for firms that demonstrate ≥25 % PUE improvement.

What it means

1. Cost compression for AI operators – A typical AI training job that runs 24 hours on a 10 MW cluster can save roughly ¥1.2 billion (US$8.5 million) in electricity costs per year when PUE drops from 1.6 to 1.15. Those savings directly improve the economics of large‑scale model development, making Japan a more attractive location for multinational AI firms.
2. Supply‑chain ripple effects – Demand for high‑performance pumps, sealed coolant loops and corrosion‑resistant alloys is expected to rise 45 % YoY through 2028, creating new revenue streams for precision‑machining firms and specialty‑chemical producers.
3. Competitive positioning – While U.S. hyperscalers have largely relied on evaporative cooling in desert sites, Japanese firms can market a “low‑carbon, high‑density” proposition to European and Asian customers facing stricter emissions caps. This could translate into export contracts worth an estimated ¥300 billion (US$2 billion) over the next five years.
4. Regulatory alignment – By adopting liquid cooling, data‑centre operators can more easily meet METI’s upcoming “Zero‑Carbon Data‑Centre” certification, which will require PUE ≤ 1.2 and a verified renewable‑energy mix for at least 80 % of power consumption.
5. Technology convergence – The same coolant‑circulation infrastructure can be repurposed for next‑generation edge AI nodes and for thermal management of quantum‑computing modules, positioning Japan’s hardware ecosystem at the intersection of several emerging compute paradigms.

Overall, the liquid‑cooling wave reflects a pragmatic response to the twin pressures of rising AI compute demand and tightening energy costs. If the projected PUE improvements materialize at scale, Japan could shave more than 10 TWh of electricity off its AI data‑centre load by 2030 – a reduction comparable to the annual output of a mid‑size nuclear plant. The financial upside for early adopters, coupled with supportive policy incentives, suggests that liquid cooling will become a standard design choice rather than a niche experiment.