The United States now uses almost 40% of the world’s data center electricity

Data center electricity demand has nearly doubled in five years

According to new data, global data center electricity demand reached 787.8 terawatt-hours in 2025, up from 658.2 TWh in 2024. That represents an extraordinary increase of almost 20% in a single year. If we go back to 2020, when global demand was 410.8 TWh, the scale of the expansion becomes even more striking: electricity use in data centers has increased by around 92% in just five years, representing an average annual growth rate of almost 14%.

From some perspective, if the world’s data centers were treated like a country, their electricity consumption would place them among the largest energy markets in the world. This is no longer a niche burden that is absorbed at the margins of existing energy systems. It is becoming an important new source of demand that utilities, grid operators, regulators and energy producers must plan around.

Growth has also been highly concentrated geographically. The United States and China dominate global data center electricity use, while Europe is a distant third. That distribution is important because adding a large amount of electricity demand to a mature grid that has seen little growth for decades creates a very different challenge than adding it to an economy where power generation and transmission are already expanding rapidly.

The United States represents almost 40% of the world total

US data centers consumed approximately 312.6 TWh of electricity in 2025, representing 39.7% of the global total. China came in second with 205.7 TWh, or 26.1%, while Europe accounted for 144.6 TWh, or 18.4%. Other parts of the Asia-Pacific region consumed another 63.2 TWh, with the remaining demand spread across Latin America, the Middle East, Africa and the rest of North America.

That means the United States and China together accounted for nearly two-thirds of all data center electricity consumption worldwide. But the United States’ share of last year’s growth is even more notable. US consumption increased from 249.0 TWh in 2024 to 312.6 TWh in 2025, a gain of 63.5 TWh. As global data center demand increased by approximately 129.6 TWh, the United States alone accounted for approximately 49% of the global increase.

U.S. data center electricity consumption has increased approximately 81% since 2020, but the 25.5% increase in 2025 represents a significant acceleration. This helps explain why the discussion about artificial intelligence has quickly become an energy discussion. From a utility’s perspective, an AI data center is, first and foremost, a huge electrical load that must be reliably served every hour of the day.

The long era of flat electricity demand in the United States is coming to an end

That load comes after a long period in which American electricity demand barely changed. The Energy Information Administration reports that electricity demand grew just 0.1% annually between 2005 and 2019. However, since 2020, demand has increased at an average annual rate of about 1.7%, and the EIA specifically identifies electricity use by data centers as a major driver of change.

Virginia offers perhaps the clearest example of what this can mean for an individual energy market. Commercial electricity sales there increased by nearly 30 million megawatt-hours between 2019 and 2025, and the EIA attributes much of that increase to the state’s huge concentration of data centers. On the Dominion portion of the PJM network, peak summer demand in 2025 was 23% higher than in 2019, while peak winter demand in 2025-26 was 45% higher than six years earlier. PJM currently expects summer peak demand in that region to grow at an average annual rate of 5.4% over the next decade, largely due to continued data center development.

That growth has consequences far beyond the generating plant. Electricity has to pass through transmission lines, transformers, substations and local distribution equipment before reaching the servers that consume it. During the first half of 2026, transmission congestion costs in PJM increased 43% to $6 billion, and Northern Virginia is among the areas experiencing significant limitations. Data centers are not solely responsible for those costs, but rapidly increasing demand is adding pressure to a network where large transmission projects can take years to permit and build.

The forecasts are huge, but also uncertain

There is little disagreement that data center electricity demand will continue to increase, although there is considerable disagreement over exactly how fast. S&P Global Energy estimates that global demand could increase at an annual rate of 12% to 16% between 2025 and 2030. Its electricity sector scenario reaches around 1,550 TWh by 2030, almost double the 2025 level and roughly equivalent to projected electricity consumption for all of Latin America. In that scenario, data centers would consume around 6% of the world’s electricity.

The range of American forecasts is equally wide. An analysis supported by the Department of Energy estimated that data centers consumed about 4.4% of US electricity in 2023 and could consume between 6.7% and 12% by 2028, depending on the pace of deployment and improvements in computing efficiency.

This is where the AI ​​story begins to collide with physical reality. Someone has to build the generation needed to supply these facilities and someone has to build the transmission lines and substations that supply the electricity.

Natural gas generation can often be built faster than nuclear power, but turbine supply chains have become tighter. Renewable capacity can be added relatively quickly, but intermittency increases the need for storage, firm generation and transmission. Nuclear plants that once seemed destined for retirement have suddenly become attractive sources of 24-hour power for technology companies seeking large blocks of reliable electricity.

Energy can become the limitation of AI

The technology industry naturally focuses on computing power, but the limiting factor in AI development may increasingly lie outside the data center itself. A company can order advanced chips and design a huge computing campus, but those servers aren’t much use without access to hundreds of megawatts of reliable electricity. At the largest proposed campuses, the requirement can reach gigawatt scale, putting an individual data center complex in the same general energy demand category as a sizable city.

This also means that the economic impact of AI is spreading to sectors that have little to do with software. Utility companies are reviewing load forecasts. Energy developers are planning a new generation. Gas pipeline companies can benefit from the additional demand for natural gas, while manufacturers of turbines, transformers, switchgear, refrigeration equipment and electrical components are seeing a new source of orders. Battery storage and renewable energy will have a role to play, as will nuclear plants and existing fossil generation. The AI ​​revolution increasingly requires an all-of-the-above energy response because the scale and speed of demand growth leaves little room to rely on a single technology.

The big picture

For most of the past two decades, the U.S. energy sector was built around the assumption of relatively stagnant electricity demand. That assumption is quickly becoming obsolete. Data centers are not the only reason, but they are emerging as one of the most powerful forces behind the change. In just five years, global data center electricity consumption has nearly doubled, and by 2025 the United States accounted for nearly half of the global increase.

The debate over artificial intelligence often focuses on software, chips, productivity, employment and whether valuations of the technology have gotten too far ahead of reality. However, from an energy perspective, something very physical is happening behind that digital revolution.

The United States now consumes nearly 40% of the electricity used by data centers worldwide, and the infrastructure necessary to support further growth will have to be financed, licensed, manufactured and built. AI may live in the cloud, but the cloud increasingly relies on power plants, pipelines, transmission lines, transformers, cooling systems and a grid that was never designed for demand to grow so quickly.