Global Data Centers Now Consume 2 Percent of World Electricity as AI Demand Surges

--- headline: "Global Data Centers Now Consume 2 Percent of World Electricity as AI Demand Surges" slug: data-centers-2-percent-global-electricity category: research story_number: "12" date: 2026-05-18 ---

# Global Data Centers Now Consume 2 Percent of World Electricity as AI Demand Surges

The world's data centers have crossed a stark threshold: they now devour roughly 2 percent of global electricity, up from an estimated 1.7 percent just two years ago, as the insatiable power demands of artificial intelligence training and inference reshape the energy landscape at a pace that has caught grid operators and policymakers off guard.

The figure, drawn from a convergence of recent analyses by the International Energy Agency, Gartner, and independent research groups, underscores how quickly AI has moved from a marginal contributor to data center load to the single largest driver of growth. In 2024, global data center electricity consumption stood at approximately 415 terawatt-hours. By 2025, that number climbed to roughly 460 to 490 TWh. Current projections for 2026 place the total at around 545 TWh -- with AI-specific workloads accounting for an estimated 200 TWh, or 37 percent of the total data center draw.

The AI Accelerant

What makes this surge distinctive is its velocity. According to Gartner, electricity demand for data centers worldwide grew 16 percent in 2025 alone and is on track to double by 2030, reaching 980 TWh. The firm projects that AI-optimized servers, which represented 21 percent of total data center power usage in 2025, will command 44 percent by decade's end, with their electricity consumption rising nearly fivefold from 93 TWh to 432 TWh.

The IEA's Energy and AI report paints a similar picture. The agency projects data center electricity demand will reach roughly 945 TWh by 2030 in its base case -- slightly more than Japan's total electricity consumption today -- with AI-driven accelerated servers growing at 30 percent annually compared to 9 percent for conventional servers.

Bob Johnson, a senior research director at Gartner, has noted that AI-optimized servers are the primary catalyst. "The rapid rise of AI-optimized servers is fueling the increase in data center power consumption," he said, pointing to the fundamental shift in how computing infrastructure is being built and deployed.

Regional Fault Lines

The burden is not evenly distributed. The United States hosts roughly 45 percent of global AI data center capacity by power draw, with Northern Virginia, Texas, and Phoenix emerging as acute pressure points. China accounts for approximately 14 percent, while Europe -- excluding Ireland -- carries 12 percent and rising.

Ireland presents perhaps the most dramatic case study of what unchecked data center growth looks like at a national scale. Data centers there already consume more than 20 percent of the country's total electricity, with some estimates suggesting that share could climb to 32 percent by the end of 2026. In the United States, data centers now account for nearly half of all projected electricity demand growth through 2030, according to the IEA.

Europe faces its own reckoning. The IEA projects 45 TWh of additional data center electricity demand across the continent by 2030, while Gartner estimates European data center electricity consumption will rise from 2.7 percent to 5 percent of regional consumption over the same period -- a near doubling that will test grid infrastructure and renewable energy commitments.

The Capital Behind the Consumption

The spending required to feed this appetite is staggering. In 2025, the five largest hyperscalers -- Microsoft, Google, Meta, Amazon, and Oracle -- collectively disclosed more than 355 billion dollars in capital expenditure on AI-relevant infrastructure, representing the largest single-cycle infrastructure investment outside government spending in modern memory.

Microsoft alone committed approximately 80 billion dollars in fiscal year 2025 capex, the majority directed toward AI. Amazon's AWS division led all companies at roughly 110 billion dollars. These sums are not speculative; they are reflected in earnings disclosures and are already translating into physical facilities drawing power from grids around the world.

Water, Carbon, and the Sustainability Paradox

Electricity is only part of the environmental equation. Hyperscaler water consumption for evaporative cooling rose 15 to 25 percent year-over-year in 2024 and 2025 disclosures. Google reported 9.1 million cubic meters of water consumption in 2024; Microsoft disclosed 7.8 million cubic meters, a 22 percent increase. A typical 100-megawatt AI data center consumes between 1.5 and 3 million cubic meters of water annually.

Fatih Birol, Executive Director of the IEA, has framed the challenge in unmistakable terms, noting that AI is set to drive surging electricity demand from data centers while also offering the potential to transform how the energy sector operates. The tension between AI's energy cost and its potential to optimize energy systems is becoming a central policy question.

Most hyperscalers purchase renewable energy certificates against their operational electricity, but absolute Scope 2 emissions continue to climb as growth outpaces clean-energy procurement. Carbon intensity per unit of compute is generally declining, but total emissions are heading in the wrong direction.

What Comes Next

The industry is searching for answers. Liquid cooling technologies, which can reduce direct water consumption by 70 to 90 percent, are gaining traction as AI accelerator power densities exceed what air cooling can handle economically. Longer-term bets include green hydrogen, geothermal energy, and small modular nuclear reactors as on-site power alternatives, though none are expected to reach meaningful scale before the end of the decade.

For now, the math is unforgiving. If current growth trajectories hold, data centers will consume nearly 3 percent of global electricity by 2030 -- roughly 945 TWh -- with AI workloads responsible for close to half of that total. The question is no longer whether AI will reshape the global energy system. It is whether the energy system can adapt fast enough to keep the lights on for everything else.