For the first time in decades, U.S. electricity demand is rapidly increasing across nearly every region of the country.
EPSA’s Fact Sheet explores the factors driving rising demand and policy approaches to meet the moment reliably and cost-effectively.
Several factors are driving this growth, including transportation and building electrification, reshoring of manufacturing spurred in part by funding from incentives like the 2022 CHIPS and Science Act, and AI-driven data center proliferation.
To meet this rapidly growing power demand reliably, the U.S. will need to quickly deploy new generation assets of all types, including both dispatchable resources like natural gas, nuclear, and batteries, and intermittent resources like solar and wind. However, this increase in power demand is coinciding with a surge of premature retirements of existing dispatchable generation assets, like natural gas power plants, which is placing grid reliability at risk.
Data Centers’ Growth Is Driving New Technological Innovation
Over the past two decades, data centers have become critical infrastructure to the U.S. and global economy by enabling internet traffic, cloud-based computing services, and crypto-mining. However, recent breakthroughs in generative artificial intelligence (AI) have driven a significant increase in the demand for new data centers, with $150 billion in investments forecasted through 2028. According to data compiled by Statista, the U.S. had over 2,700 data centers at the end of 2022. By 2024, the number nearly doubled to over 5,300.
Estimated Electricity Demand From Data Centers in U.S. Power Markets (MW)
Data Center Power Consumption – By the Numbers
Data centers operate 24/7/365 and are power hungry.
U.S. data centers tax the power grid.
Even with the immense growth over the past few years, those projections still don’t yet account for radical adjustments due to adoption of AI, according to S&P Global.
Power Demand From Artificial Intelligence
Electricity usage from AI can be split into two categories: “narrow AI” and “interference usage.” The first is not energy intensive and is used for applications like forecasting and predictive maintenance. The second is used to run individualized user prompts and is much more energy intensive.
A recent peer-reviewed analysis published in Joule quantified the energy demand being driven by AI. If current demand trends continue, NVIDIA, a leading manufacturer of AI servers, could ship 1.5 million servers per year by 2027, which would consume 85.4 TWh of electricity annually at full operating capacity.
This is roughly equivalent to the energy consumption of Delaware (86.6 TWh) and more than Hawaii (79.4 TWh), Rhode Island (54.4 TWh), the District of Columbia (44.2 TWh), and Vermont (37.8 TWh).
For context, the three largest energy consuming states are Texas (475.4 TWh), California (251.9 TWh), and Florida (248.8 TWh).
Takeaways for Policymakers
Rapidly growing electricity demand driven by data center growth and AI will require vast amounts of new reliable power generation. The rapid expansion of data centers is happening at a time when grid operators are already warning of a looming reliability crisis. If power demand is to be met reliably, policymakers and regulators will need to implement policies that forestall premature retirements of critical dispatchable generation resources like natural gas and reduce barriers – for example, by implementing permitting reform – to constructing new generation assets.
EPSA’s Energy Expansion Principles outline how competitive wholesale power markets can be leveraged to deliver a reliable and cost-effective energy expansion – by encouraging private capital deployment and innovation at the lowest cost while appropriately shifting investment risk to resource developers and operators, not consumers.
