In 2024, something happened in Northern Virginia that alarmed power grid operators. Dozens of data centers suddenly disconnected from the grid. In a single instant, approximately 1,500 MW of load disappeared. Grid operators managed the situation and prevented a widespread outage. But the warning was clear. Grids are not built to handle fluctuations like these. And this was just one incident. The overall picture is far more concerning.
Electricity consumption in the US rose by 2% in 2025. That may seem like a figure hardly worth noting, but the fact is that data centers alone accounted for approximately 50% of this entire increase. They surpassed households, industry, and transportation combined. This is according to a report by the International Energy Agency (IEA) published in April 2026. And the agency expects the same share of growth to continue until at least 2030.
Compare that with the global figures: worldwide, data centers accounted for 17% of the increase in electricity consumption. In the US, the figure is three times higher. America is the epicenter of the AI boom, and the grids are beginning to feel it. Last year, technology companies poured more than $61 billion into data centers around the world. The United States and Canada accounted for more than $47 billion of that. Construction took place on a massive scale. But hardly anyone gave much thought to how the grid would handle it all.
The grid is not designed for such a sharp increase in load
Traditional industrial facilities operate predictably. Machines start up, stabilize, and keep running. Grid operators know what to expect. AI data centers are different. Their consumption changes every second. Workloads shift between model training and operation, and electricity demand jumps along with them. Electricity consumption can fluctuate by hundreds of megawatts within seconds as large computing processes start or stop. Such fluctuations outpace the grid’s traditional ability to respond.
That is a problem. Grid protection systems are designed to respond to predictable load behavior. When these assumptions break down, the system may disconnect before it has time to respond correctly. And large blocks of load may suddenly disconnect in ways no one anticipated.
Before an electricity distributor allows a large consumer to connect to the grid, it must perform a series of detailed calculations. Power flow simulations, fault analyses, and models of load behavior under various conditions. With AI data centers, it is not enough simply to measure how many megawatts they require. They must model how that consumption behaves, how quickly it rises and falls, and how it responds to faults. And the number of data centers is growing faster than engineers can process these studies.
“We are adding a large new load and have to model the impact on the entire system. That work is incremental. There is a limited number of engineers who do it,” Khajuria explains. Distributors are facing growing workloads, and the queue of applicants waiting for grid connections is getting longer. This is slowing down development across the board. In a world where technology companies want to build data centers as quickly as possible, this is a serious obstacle.
Denmark stopped accepting applications
Here in Europe, the problem is even more acute. Denmark, which had long been considered an ideal location for data centers, stopped approving new grid connections this spring.
In March 2026, state-owned grid operator Energinet introduced a temporary suspension of new connection agreements. The reason? An explosive increase in capacity requests. Approximately 60 GW of projects are waiting in the queue. Yet peak electricity consumption across Denmark is around 7 GW. Data centers account for nearly a quarter of the requested capacity, or roughly 14 GW.
“We have to be realistic and look at what is actually available. We cannot simply issue connection agreements without limits, because the electricity just is not there,” says Henrik Hansen, head of the Danish Data Industry Association (DDI). The suspension was intended to last three months. But an extension cannot be ruled out, Hansen admits. Before political decisions can be made regarding connection priorities, Denmark needs a new government and a new legislative framework. Elections have taken place, but the new government is still being formed.
Bills are rising and people are angry
Technology companies are celebrating and stock markets are rising, but for millions of households the situation is different. Electricity prices are soaring.
In 2025, US electric utilities requested rate increases totaling more than $30 billion, affecting 81 million Americans. Since 2021, electricity bills have risen by 40% overall. Data centers are not responsible for everything, but they are bearing the brunt of the anger. Surveys show that most households associate the expansion of data centers with rising electricity prices.
Communities across the United States are pushing back. Last year, local opposition blocked or delayed at least 16 data centers with a combined value of more than $64 billion. In April 2026, Maine passed a proposal for a statewide ban on new data center construction. Governor Janet Mills ultimately vetoed it, but the pressure continues. Meanwhile, lawmakers in Congress have proposed tightening data center construction rules nationwide.
Sentiment toward AI is generally deteriorating. A Pew Research survey shows that Americans are more concerned than excited about artificial intelligence. More than half of respondents believe the technology will do more harm than good. This frustration is spilling into the streets as well, with someone throwing a Molotov cocktail at the home of OpenAI CEO Sam Altman last month.
Sources: cnn.com, fortune.com, datacenterknowledge.com and cnbc.com
