American electricity consumers and grid planners face a stark new forecast: data centers driven by artificial intelligence workloads could consume between 9 and 17 percent of all U.S. electricity by 2030, according to analysis released by the Electric Power Research Institute. That range represents a dramatic acceleration from the roughly 4 percent share data centers held in 2023, according to the U.S. Department of Energy, and it raises hard questions about who will pay for the infrastructure needed to keep the lights on everywhere else.
Why a Fourfold Jump in Data Center Power Demand Matters Right Now
The tension behind this forecast is simple: U.S. electricity demand had been essentially flat for more than a decade. Efficiency gains in appliances, lighting, and industrial processes offset modest growth. Data centers are reversing that trend at a speed few utilities planned for. When a single sector moves from 4 percent of national generation to a potential 17 percent in roughly seven years, the strain lands on transmission lines, generation capacity, and ratepayers simultaneously.
States that have fast-tracked data center permitting, particularly Virginia, are already feeling concentrated effects. The hypothesis that these fast-permitting states will see retail electricity rates climb at least 15 percent faster than the national average by 2028 is testable through public utility commission filings. Utilities in those regions must file rate cases to recover the cost of new substations, upgraded feeders, and generation contracts. If the load growth materializes at the scale EPRI projects, those filings will show the price signal well before 2030.
For households and businesses outside the tech sector, the consequence is direct. New power plants, whether gas-fired or renewable, require capital. Transmission upgrades require capital. Both get recovered through rates. A data center campus drawing hundreds of megawatts in a single county can force upgrades that residential and commercial customers in the same service territory help finance, even if they never run an AI query.
EPRI’s Range and the Federal Baseline That Frames It
The 9 to 17 percent range comes from EPRI’s analysis released on February 26, 2026. The spread reflects different assumptions about how quickly new facilities come online, how rapidly AI chip power density grows, and how much efficiency improvement offsets raw demand. At the low end, 9 percent would still represent more than doubling the 2023 share. At the high end, 17 percent would mean data centers alone consume more electricity than the entire residential air-conditioning load in many regions.
The baseline for these projections rests on historical consumption data compiled by Lawrence Berkeley National Laboratory. LBNL’s 2024 usage report (LBNL-2001637) tracks U.S. data center electricity use back to 2014 and projects growth through 2028. That dataset shows a steady upward curve that accelerated as hyperscale facilities expanded and AI training runs grew larger, even as operators improved cooling systems and server efficiency.
A notable gap exists between EPRI’s own figures depending on where they appear. The Department of Energy’s Office of Electricity, in its discussion of clean energy resources for data centers, cites EPRI’s estimate as facilities reaching up to approximately 9 percent of U.S. electricity generation annually by 2030, starting from roughly 4 percent in 2023. EPRI’s direct release, however, presents the wider 9 to 17 percent band. The discrepancy likely reflects different scenario assumptions or publication timing, but it means policymakers working from the DOE figure may be planning for a load increase less than half the size of EPRI’s upper bound.
Conflicting Forecasts and Unanswered Rate Questions
The gap between 9 percent and 17 percent is not a rounding error. It represents the difference between a manageable grid expansion and a generational infrastructure challenge. At 9 percent, existing utility build-out plans and interconnection queues might absorb the load with modest delays. At 17 percent, the country would need to add generation and transmission capacity equivalent to powering tens of millions of additional homes, all within a few years.
Several questions remain open. Neither EPRI’s public release nor the LBNL report provides granular state-by-state consumption forecasts beyond flagging Virginia and other high-impact states. Without that detail, regulators in fast-growing markets cannot easily model how much new load will land in their territory versus a neighboring state. The methodology behind the 9 to 17 percent range, including the specific scenario inputs and assumptions about chip efficiency trends, has not been fully published in the peer-reviewed literature as of this writing.
Utilities in high-growth corridors also lack updated meter-level data from 2025 and 2026 that would validate or challenge EPRI’s assumptions in real time. Data center operators often negotiate non-disclosure agreements around their power contracts, making it difficult for public utility commissions to share precise load profiles with outside analysts. That opacity slows feedback loops: planners cannot easily compare forecasted versus actual demand, and the public cannot see how much of a proposed rate increase is driven by a handful of large customers.
These information gaps feed into a broader question: who should bear the cost of building out a grid tailored to energy-intensive AI computing? Traditional ratemaking spreads fixed costs across all customers in a service territory. Large industrial users typically receive discounted rates in exchange for long-term contracts or flexible load. In the data center context, those discounts can collide with public perceptions that everyday ratepayers are subsidizing the infrastructure needed to power cloud services and generative AI tools.
Regulatory Levers and Emerging Responses
State regulators are not powerless in the face of these trends. Public utility commissions can require more detailed reporting on large new loads, including projected timing, location, and peak demand. They can also condition approvals for new substations or transmission lines on cost-allocation structures that assign a greater share of expenses to the customers that trigger them. Some commissions are already exploring separate rate classes for data centers, allowing them to tailor demand charges, time-of-use pricing, or interruptible service options to this specific sector.
On the planning side, integrated resource plans can be updated more frequently to reflect the rapid pace of AI-driven demand growth. Instead of assuming flat or gently rising load, utilities can model high-growth scenarios aligned with EPRI’s upper range and stress-test their portfolios. That approach does not guarantee perfect foresight, but it reduces the risk of being caught short on capacity and resorting to expensive stopgap measures that ultimately raise rates.
There is also a growing push to align data center development with local clean energy resources. Co-location near robust wind or solar generation, combined with long-duration storage, can reduce strain on congested transmission corridors. However, even fully renewable-powered campuses still depend on the broader grid for reliability, especially during periods of low output or equipment outages. The need for firm capacity and grid services does not disappear simply because a facility signs a green power purchase agreement.
What Policymakers Need Next
The immediate priority for policymakers is clarity. Reconciling the DOE-cited 9 percent figure with EPRI’s 9 to 17 percent band would help set a common planning baseline. That does not require choosing a single number, but it does mean publishing transparent scenario assumptions and clarifying which figures guide federal and state infrastructure programs.
More granular, public data on data center electricity use would also improve decision-making. Standardized reporting thresholds for large loads, combined with anonymized aggregation at the county or balancing-area level, could preserve commercial confidentiality while giving communities a clearer view of how much local infrastructure is being built to serve AI and cloud computing. With that information, debates over siting, tax incentives, and cost recovery can be grounded in measurable impacts rather than speculation.
Finally, regulators and legislators will have to decide how far to lean on cost-causation principles in ratemaking. If AI-driven data centers ultimately consume close to one-sixth of U.S. electricity, the balance between economic development goals and bill impacts for existing customers will become harder to ignore. The EPRI projections do not dictate those choices, but they sharply define the stakes: a world where data centers modestly reshape the grid, and a world where they drive its most significant expansion in decades.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
