⚡ Is AI devouring the world's electricity? A single ChatGPT query consumes roughly 10 times more power than a Google search. If every Google search worldwide switched to AI, it would require the equivalent of Ireland's entire annual electricity output. Now, tech companies are no longer just buying power — they're investing billions to produce it. Here's how AI is triggering a global energy transformation.
The "Power Hunger" of AI: Just How Much Electricity Does It Consume?
The rapid spread of generative AI is fundamentally reshaping the global energy landscape. According to the International Energy Agency (IEA), combined electricity consumption from data centers, AI, and cryptocurrencies worldwide is projected to more than double from 460 terawatt-hours (TWh) in 2022 to as much as 1,050 TWh by 2026 — roughly equivalent to Japan's entire annual electricity consumption.
Driving this surge is an unprecedented boom in AI data center construction. Meta plans at least $60 billion in capital expenditure for 2025, aiming to operate 1.3 million GPUs by year-end. Microsoft announced $80 billion in AI infrastructure spending for fiscal 2025. The power draw of individual GPU chips keeps climbing too: NVIDIA's latest Blackwell chip consumes approximately seven times more electricity than the A100 chip used to train early generative AI models.
The energy gap between traditional web services and generative AI is staggering. A single AI query uses about 10 times the electricity of a conventional Google search. If all of Google's searches (roughly 9 billion per day) shifted to an AI-based system, it would require an additional 29 TWh annually.
When Tech Companies Become Power Companies
Alarmed by the looming power shortage, major tech companies are moving beyond simply purchasing electricity — they're investing directly in the energy business itself.
The most symbolic development is investment in restarting dormant nuclear power plants. Microsoft signed a 20-year power purchase agreement (PPA) with Constellation Energy to revive Unit 1 at Three Mile Island — famous for its 1979 accident — investing approximately $1.6 billion. Once restarted, the facility will supply 835 MW of carbon-free electricity directly to Microsoft's data centers.
Google has partnered with NextEra Energy to support the restart of the Duane Arnold Energy Center in Iowa, signing a 25-year PPA with commercial operation targeted for Q1 2029.
Amazon has taken an even bolder approach, acquiring data centers co-located with nuclear power plants for direct electricity access. The company is also investing in developers of small modular reactors (SMRs) — next-generation nuclear technology — and signed a memorandum of understanding with Dominion Energy for SMR development in Virginia.
The numbers are staggering. Capital expenditure on AI and data centers by the "Magnificent Seven" mega-cap tech companies is expected to reach approximately $414 billion in fiscal 2025 and exceed $430 billion in 2026. Across the entire tech industry, global data center capex is projected to reach $1.2 trillion by 2029.
Nuclear Energy's Comeback as "AI's Power Source"
Tech companies are turning to nuclear power because renewables alone cannot satisfy data centers' demand for uninterrupted, 24/7 electricity. While solar and wind capacity is expanding rapidly, these are intermittent sources that cannot yet reliably meet data centers' zero-downtime requirements. Natural gas turbines are expected to serve as the primary near-term power source, but reliance on fossil fuels contradicts the carbon-neutral commitments made by most tech companies.
Nuclear energy has emerged as the solution. The U.S. government has announced plans to quadruple nuclear capacity by 2050 and signed a strategic partnership worth at least $80 billion with Westinghouse Electric to build new reactors using the company's technology.
SMRs differ from traditional large-scale nuclear plants: they're compact, quicker to build, and can potentially be located near data centers. Hyperscalers like Google, Microsoft, and Amazon are leveraging their trillion-dollar balance sheets to secure every possible power source — nuclear, natural gas, solar, and beyond.
Impact on Japan and the "Watt-Bit Integration" Strategy
Japan is far from immune to this global power demand surge. According to Wood Mackenzie, Japan's data center electricity consumption is projected to triple from 19 TWh in 2024 to 57–66 TWh by 2034. Peak power demand from data centers could reach 4–5% of Japan's total peak electricity demand by 2034. Japan's Organization for Cross-regional Coordination of Transmission Operators (OCCTO) forecasts that peak demand from new data centers and semiconductor factories could increase roughly 13-fold by 2034 compared to 2025 levels.
However, Japan faces challenges distinct from those in the United States. Data centers have concentrated in specific areas — particularly the Inzai-Shiroi district in Chiba Prefecture — where applications for grid connection now far exceed available capacity. Lengthy grid-connection timelines are increasingly at odds with business plans.
To address these challenges, the Japanese government unveiled the concept of "Watt-Bit Integration" (ワット・ビット連携) as part of the "GX 2040 Vision" approved by cabinet in February 2025. This innovative framework aims to integrate energy infrastructure (Watts) with information and communications technology (Bits) holistically.
In practice, this means strategically distributing data centers across regions rich in renewable energy and dynamically shifting AI computational workloads between regions based on real-time power supply conditions — a concept called "workload shifting." For example, during daylight hours, computing tasks could be concentrated at data centers in Kyushu where solar power is abundant, then shifted to Hokkaido or Tohoku at night when wind generation peaks. The University of Tokyo and Fujitsu launched a proof-of-concept experiment for this technology in January 2026.
SoftBank Group is constructing a large-scale AI data center in Tomakomai, Hokkaido, leveraging the region's cold climate for energy-efficient cooling and locally produced renewable energy. The facility is scheduled to open in fiscal 2026.
The Japanese government also announced plans to invest approximately $1.4 billion over five years starting in fiscal 2026, subsidizing up to 50% of investments in semiconductor plants and data centers that utilize 100% decarbonized power sources such as renewables and nuclear energy.
"The ESG Paradox" and Japan's Structural Challenge
Japan faces a structural dilemma. Global tech companies demand clean energy for their data centers, but Japan's renewable energy supply remains insufficient. Data centers in the greater Tokyo area often have no choice but to rely on thermal power generated from coal and LNG (liquefied natural gas). This creates what some analysts call the "ESG Paradox" — an ironic situation where the world's supposedly greenest companies end up using some of the most carbon-intensive electricity in Japan.
Overcoming this challenge requires simultaneously advancing nuclear plant restarts for stable baseload power, accelerating renewable energy deployment, and upgrading the power transmission grid. AI is a critical technology that will determine Japan's industrial competitiveness, and the power infrastructure that underpins it is no longer just an energy policy issue — it has become a matter of national economic strategy.
In Japan, efforts to solve the AI era's power challenges through both technology and infrastructure are accelerating. Tech companies investing in power utilities is an unprecedented phenomenon happening worldwide. How is your country addressing the power demands of the AI age? We'd love to hear your perspective.
References
- https://globalxetfs.co.jp/research/ai-hinges-on-rapid-electrification/index.html
- https://www.jogmec.go.jp/publish/plus_vol27.html
- https://www.enecho.meti.go.jp/about/special/johoteikyo/gxseisaku2025.html
- https://www.woodmac.com/ja/press-releases/japan-data-centers-power-demand/
- https://www.iaea.org/bulletin/data-centres-artificial-intelligence-and-cryptocurrencies-eye-advanced-nuclear-to-meet-growing-power-needs
- https://www.nga.org/publications/nga-nuclear-dispatch-powering-a-new-era-of-innovation/
- https://www.u-tokyo.ac.jp/focus/ja/press/z1313_00027.html
Reactions in Japan
Tech companies investing in nuclear restarts — nobody would've believed this 10 years ago. But this basically proves stable supply isn't possible with renewables alone. A victory for realism, I'd say.
Microsoft restarting Three Mile Island... using the symbol of that nuclear accident for AI? I think we need more ethical discussion about whether technological progress justifies this.
The data center concentration problem in Inzai has been talked about for years. In the end, transmission line capacity shortages becoming a bottleneck for Japan's IT industry... we're so behind on infrastructure.
Watt-Bit Integration trials starting is good news. But workload shifting between Kyushu and Hokkaido hinges on acceptable network latency. Fine for inference, but real-time processing might be tough.
I was shocked to learn one ChatGPT query uses 10x more power than a Google search 😱 It's convenient, but the AI we casually use consumes this much electricity... We need to think about how we use it.
Seeing US utility stocks soar on AI demand, I've been buying Japanese power stocks. KEPCO and Kyushu Electric are my picks. If nuclear restarts proceed, data center demand will keep climbing. Looking at a 10-year horizon.
Cases where data center grid connection applications are delayed over 10 years is abnormal. The problem is operators who apply first without even securing land. Regulations against speculative capacity hoarding are needed.
The SoftBank DC in Tomakomai is exciting for locals — jobs will be created. Cold climate cooling makes total sense. But I do worry about heavy snow cutting off access in winter.
The trend of tech companies investing in utilities is still slow in Japan. METI talks about promoting GX but deregulation is too slow. Compared to US movements, Japan seems 2-3 years behind.
SMRs won't be practical until the 2030s. The biggest question is what bridges the gap until then. If we rely on natural gas, we should drop the carbon-neutral banner.
When they say AI causes power shortages, it honestly doesn't feel real to me. But rising electricity bills are a real problem. Building data centers is fine, but please don't let it drive up household electricity costs...
If energy-efficient high-performance AI like DeepSeek emerges, power issues could ease. Throwing money at hardware isn't the only solution. Software efficiency improvements need to happen simultaneously.
The forecast that Japan's solar and wind will only reach 17% by 2030 is grim. A future where tech companies bypass Japan because they can't get green power is becoming increasingly realistic.
My startup uses AWS Tokyo Region. If power constraints raise cloud costs, it's a matter of survival. Distributing to regional DCs means latency tradeoffs, which is a tough call.
It's such a waste that a volcanic nation doesn't leverage geothermal. All the focus is on nuclear restarts, but geothermal provides stable 24/7 baseload power. If coordination with the hot spring industry progresses, there's huge potential.
Singapore imposed a moratorium on new data center construction. As an island nation with limited power and land, Japan's 'Watt-Bit Integration' concept is inspiring. Unfortunately, we don't have a cold region like Hokkaido.
I live in Virginia. Data centers keep popping up in the neighborhood, completely changing the landscape. Our electricity bills have gone up too, probably from increased demand. It's not fair that local residents bear the burden for tech company profits.
I'm a data center cooling engineer in northern Sweden. SoftBank's Hokkaido DC plan makes perfect sense. Here we use free cooling with ambient air and cut annual cooling costs by over 60%.
India just experienced widespread blackouts from last year's heat wave. If AI demand further strains power supply, ordinary citizens suffer. I'm concerned that developed-world tech companies will monopolize clean energy while developing nations bear the fossil fuel burden.
In Ireland, data centers consume 21% of all electricity and are expected to hit 32% by 2026. We should have acted sooner. Japan thinking about regional distribution and power strategy at this stage is wise.
China's AI data center power demand is projected to double to 400 TWh by 2030. We're running a two-front strategy: relying on coal while rapidly expanding renewables. Different from Japan, but the power shortage challenge is shared.
The UAE is securing power through desert solar and Barakah nuclear while building a data center hub in Dubai. Stable power and cooling are key to attracting tech companies. Japan's challenge appears to be aging transmission grid infrastructure.
In Finland, we reuse data center waste heat for district heating. Could Japan do something similar in Hokkaido or Tohoku? It's important to be creative about the 'output' side of power consumption too.
Brazil's hydro provides over 60% of electricity, so we have plenty of clean power for data centers. But our transmission infrastructure is fragile with significant losses from the Amazon to urban centers. The infrastructure challenges are somewhat similar to Japan's.
Germany shut down all nuclear plants but now imports French nuclear power. In the end, going nuclear-free was idealism. Looking at tech companies' AI demand, Japan proceeding with nuclear restarts is a rational decision.
Australia has world-class solar resources but battery storage for data centers hasn't caught up. Japan's 'Kyushu by day, Hokkaido by night' workload shifting is a creative approach using time-zone-like differences.
As an engineer, GPU power consumption increasing 7x while generation infrastructure can't keep up is a clear design failure. Chip power efficiency and software optimization are the real solutions; building more power plants is just treating symptoms.
France generates about 70% of electricity from nuclear, giving us the advantage of stable, clean power for data centers. That's why tech companies are increasing investment here. I hope Japan won't be afraid to utilize nuclear power.
In Nigeria, half the population lacks stable electricity access. While developed nations consume massive power for AI, some countries can't even provide basic electricity. This imbalance needs more discussion.
South Korea is also facing tight power supply from rapid data center growth. The competition for land and power near Seoul mirrors Japan's Inzai situation. It would be interesting if there were plans for Japan-Korea grid interconnection.