While artificial intelligence uses only about half a percent of the world's electricity, the real challenge isn't global consumption,it's where that power is concentrated. In places like Ireland and Virginia, data centers are consuming a quarter or more of local electricity supplies, creating pressure on regional grids even as worldwide energy demand remains manageable.

How Much Electricity Does AI Actually Consume Globally?

The numbers might surprise you. According to the International Energy Agency (IEA), data centers worldwide consumed around 485 terawatt-hours (TWh) of electricity in 2025, equivalent to Germany's entire annual electricity generation. That sounds enormous until you put it in context: it represents just 1.5% of global electricity generation. For AI-focused data centers specifically, the figure is even smaller. Researchers estimate that AI consumed around 0.5% of the world's electricity in 2025, or roughly 0.1% to 0.2% of total primary energy consumption.

The distinction between AI and non-AI data centers matters here. Non-AI facilities, which power everything from email and Netflix to Google Maps and online banking, consumed about two-thirds of that total energy. AI-focused facilities used the remaining third. But this gap is closing fast. Projections suggest that by 2030, data centers could account for 3% of global electricity, with AI and non-AI facilities consuming roughly equal amounts.

Why Are Different Organizations Reporting Such Different Numbers?

If you've seen wildly different estimates of data center energy use, you're not imagining it. The IEA's figures differ significantly from those published by S&P Global, which are reported by the Energy Institute. For 2025, S&P's estimates are around 60% higher than the IEA's numbers.

The main culprit is cryptocurrency mining. S&P Global includes electricity consumption for crypto mining in its data center figures, while the IEA does not. Cryptocurrency mining likely consumes between 150 and 200 terawatt-hours annually, which accounts for nearly two-thirds of the difference between the two organizations' estimates. Even accounting for this gap, methodological differences remain. S&P models power demand based on installed data center capacity and assumptions about how often that capacity is used, while the IEA attempts to estimate electricity consumption for IT equipment, cooling, and infrastructure more directly.

Where Is Data Center Power Consumption Most Concentrated?

This is where the real story emerges. While 1.5% of global electricity might seem manageable, the geographic concentration of data center demand creates acute challenges for specific regions and communities. The United States uses about 5% of its electricity for data centers, with AI-focused facilities accounting for roughly 2% of that. But these figures mask dramatic regional disparities.

Consider the extremes. In Ireland, data centers account for more than 20% of electricity consumption, making them a dominant force in the country's energy landscape. In the United States, several states exceed 10% of total electricity demand for data centers, with Virginia reaching more than one-quarter of the state's electricity consumption. This concentration, combined with the rapid pace of AI growth, puts significant pressure on local grids, even if the worldwide impact appears modest.

Understanding the Energy Cost of Individual AI Queries

Many people wonder about the environmental impact of their personal AI use. If you ask ChatGPT, Gemini, or Claude a question, how much electricity does that single query consume? The answer is surprisingly small, though exact figures remain difficult to pin down because most technology companies don't publish detailed energy breakdowns for their models.

Google provided one of the first public estimates in 2025 when it released energy data for its Gemini model. The company found that a median text-based query consumed around 0.24 watt-hours (Wh) of electricity. To put that in perspective, that's roughly the amount of energy a microwave uses in less than one second, or a television in about ten seconds. While individual queries are negligible, the cumulative effect of billions of daily queries across millions of users creates the substantial aggregate demand we see in data centers.

How to Understand Data Center Energy Consumption Claims

• Check the Scope: Verify whether estimates include only AI data centers or all data centers, and whether cryptocurrency mining is included in the figures, as this can create 60% or larger differences between sources.
• Consider Geographic Context: Global averages mask critical regional disparities; data centers may consume 1.5% worldwide but over 20% in specific countries or states, creating localized grid pressure.
• Distinguish Training from Inference: Energy consumption is dominated by inference (running models) rather than training (building models), though most technology companies don't publish detailed breakdowns of either.
• Account for Infrastructure Overhead: Published energy estimates include cooling, lighting, and other operational costs beyond just the servers themselves, which can significantly increase total consumption.

The broader picture reveals a technology sector at an inflection point. While AI's global energy footprint remains modest relative to worldwide consumption, the concentration of that demand in specific regions creates real challenges for local electricity grids and communities. As AI adoption accelerates and data center construction continues, the geographic mismatch between where power is needed and where it's available will likely become an increasingly important constraint on AI expansion.