What's Happening with AI and Electricity Usage in the USA?
AI is transforming US electricity demand. Data centres could account for nearly 12% of national consumption by 2030, while rising wholesale prices and investment in electricity networks raise questions about who will pay for the boom.
America's AI boom is no longer just a story about semiconductors and investment. It is becoming a major challenge for the electricity system, with important implications for household and business electricity bills.
A new study released by Lawrence Berkeley National Laboratory in June 2026 estimates that US data centres consumed 192 terawatt-hours (TWh) of electricity in 2024, representing 4.7% of national consumption.
By 2030, that figure could reach 649 TWh, or 11.8% of America's electricity consumption. The study's alternative scenarios range from 9.5% to 15.3%.
These figures cover all data centres, not just AI. However, increasingly powerful AI processors and their cooling requirements are major contributors to the growth.
What is happening to electricity prices?
The US Energy Information Administration's October forecast expects national electricity consumption to reach 4,288 TWh in 2026 and 4,356 TWh in 2027, both records.
The EIA forecasts average wholesale electricity prices of $52 per megawatt-hour in 2026, up 11% on 2025.
The regional differences are striking. In PJM, the electricity market covering much of the eastern United States, wholesale prices are forecast to increase 41% this year.
Average residential electricity prices nationally are expected to rise from 17.3 US cents per kilowatt-hour in 2025 to 18.2 cents in 2026.
Not all these increases are attributable to AI. Weather, natural gas prices and electricity network costs are important influences.
Who should pay?
The problem is that data centres can be built faster than the additional power stations and transmission infrastructure required to supply them.
In July, PJM's electricity capacity auction secured supply that was 6,831 megawatts below its future reliability requirement.
This creates an important economic question: should households and existing businesses contribute to the cost of infrastructure required by AI companies?
The answer will increasingly shape American energy policy.
|
BENCHMARK ANALYTICS | US ECONOMY
AI's Growing Demand for Electricity
How much electricity will data centres consume,
and what will it cost?
|
||||||||||||||||
|
||||||||||||||||
|
Data centre electricity consumption could
more than triple between 2024 and 2030.
One terawatt-hour (TWh) equals one billion
kilowatt-hours (kWh). Figures cover all
data centres, including non-AI uses.
|
||||||||||||||||
| Data centres' share of US electricity use | ||||||||||||||||
Bar scale: 0–15.3%. The 2030 forecast
range is 9.5%–15.3%, reflecting
uncertainty about equipment deployment,
processor power use and utilisation.
|
||||||||||||||||
| What is happening to electricity prices? | ||||||||||||||||
|
||||||||||||||||
|
Average residential electricity prices
US cents per kilowatt-hour
Wholesale and retail prices are different.
Household bills also include network
infrastructure and other costs. These
increases cannot be attributed solely to AI.
|
||||||||||||||||
|
THE ELECTRICITY NETWORK CHALLENGE
6,831 MW
Shortfall against PJM's future reliability
capacity requirement in its July 2026
auction for 2028–29.
This is a shortfall against a planning
standard, not a reported electricity outage.
|
||||||||||||||||
|
THE ECONOMIC QUESTION
Who should pay for the additional power
stations and electricity networks
required by the AI boom?
|
||||||||||||||||
|
Sources: Lawrence Berkeley National
Laboratory, United States Data Center
Energy Usage Report: 2025 Update
(June 2026); EIA Short-Term Energy
Outlook (6 October 2026);
PJM 2028/29 Capacity Auction
(14 July 2026).
Berkeley Lab report | EIA forecast | PJM announcement Benchmark Analytics | 12 October 2026 |
||||||||||||||||