Surging electricity demand from data centers could more than double global power needs in the sector by 2030, but grid congestion is threatening to derail that growth.
A new report released today by the International Energy Agency finds that approximately 20% of planned data center projects risk significant delays, as power developers face interconnection wait times as long as a decade in key regions. This is injecting uncertainty into global demand forecasts.
IEA’s base case scenario, which is based on the most recent industry expectations for server shipments, predicts that global data center electricity consumption will skyrocket in the next decade, jumping from 415 terawatt-hours in 2024, to around 945 TWh in 2030, and then to 1,200 TWh in 2035.
But there are many challenges and uncertainties that complicate the forecast. The report’s “headwinds scenario” considers the possibility of slower than expected AI adoption due to challenges like grid congestion and lack of skilled labor, along with tight supply chains.
Under that scenario, data center electricity consumption rises to only around 670 TWh by 2030, and plateaus at around 700 TWh.
At the opposite end of the spectrum, a “lift-off case,” in which AI adoption goes faster than expected, and the challenges are minimal, predicts demand will exceed 1,700 TWh by 2035.
The headwinds
Because of strained electricity grids, the report estimates that “around 20% of planned data center projects could be at risk of delays.” For new data centers, getting power has become an increasingly complicated process, and operators in some regions have to wait as long as a decade to get it.
In Northern Virginia, for example, which is considered the largest and fastest-growing data center market in the world — and where data centers already consume 25% of the electricity supply — the average time in a queue is up to seven years. Data centers developers in Germany face similar waiting times, while in the Netherlands the average wait time for a load connection queue can be 10 years. In Dublin, Ireland, the grid operator got so overwhelmed it paused data center interconnection agreements in the area until 2028.
A congested grid is expensive. In the U.S. and Germany for example, the cost of managing congestion tripled between 2019 and 2022, while in the Netherlands it increased sixfold, before decreasing in 2023 in all regions due to cheaper natural gas prices.
Tight supply chains make grid congestion more complicated. The report notes that new transmission lines can take up to eight years to build in advanced economies, with wait times for transformers and cables doubling in the past three years.
Similarly, on the generation side, the backlogs for new gas turbines are growing, pushing up prices and potentially delaying construction schedules. Given those challenges, IEA estimates natural gas should expand by 175 TWh in the next ten years to meet growing data center demand.
Given that revised integrated resource plans coming from utilities in the U.S. call for an additional 84 GW of natural gas-fired capacity by 2035, such delays could be particularly disruptive.
And the supply chain challenges highlighted in the report presumably don’t yet take into account President Trump’s trade war, which, even with a 90-day delay on some tariffs, is already reshaping the global energy landscape and complicating energy supply chains.
Flexibility is key
Data center flexibility, both from a geographical and operational perspective, is “key” to mitigate some of these risks.
“Locating new data centers in areas of high power and grid availability, and operating either data center servers or their onsite power generation and storage assets more flexibly,” are still “underexplored” strategies, according to the report. (Listen to Shayle Kann’s conversation with Tyler Norris on the capacity potential for flexible data centers.)
Historically, data centers have developed in clusters. Globally, more than 15% of data center capacity under development is in the top ten largest data center markets, like Northern Virginia, the Netherlands, or Ireland. In the U.S., half of the data centers under development are in markets with over 1 gigawatt of capacity already installed. But this trend creates grid congestion, as illustrated by the fact that these places are also the ones with the longest load connection queue times.
From an operational point of view, the report finds that if data centers are flexible for 0.1-1% of the time, by adopting options like batchable workloads and load shifting, “there is enough room in current electricity systems to integrate all new data centre capacities to 2035.”
In the U.S., studies have found that the grid may have over 100 GW of load to spare with minimal curtailment. In China, “data center additions would need to be flexible for 0.2% of the time to fit within the existing system capacity, the equivalent of 20 hours per year on average,” according to the IEA report.
For data centers to achieve this degree of flexibility, however, solutions like on-site generation, additional batteries, and the management of computational workloads, which are still limited, would have to be developed at scale.
