U.S. Power Demand Is Set for New Records in 2026 and 2027 as AI Data Centres Lift Electricity Use

# U.S. Power Demand Is Set for New Records in 2026 and 2027 as AI Data Centres Lift Electricity Use

Finin2min 2-minute summary

The U.S. Energy Information Administration expects electricity demand to reach record levels in both 2026 and 2027 as data centres, AI computing and broader electrification increase load. EIA forecasts consumption of about 4,288 billion kWh in 2026 and 4,356 billion kWh in 2027, up from a record 4,195 billion kWh in 2025. The important finance implication is that AI growth is no longer only a semiconductor or software story: it increasingly affects generation, transmission, gas demand, wholesale electricity prices and capital expenditure across the power system.

**Last verified:** 2026-10-07 01:14 IST

Key verified facts

• EIA forecasts U.S. electricity demand of about 4,288 billion kWh in 2026 and 4,356 billion kWh in 2027.
• That compares with a record 4,195 billion kWh in 2025.
• EIA and Reuters link the demand increase partly to AI and data-centre growth as well as broader electrification.
• EIA expects average U.S. wholesale electricity prices around $52/MWh in 2026, about 11% above 2025.
• Regional effects differ sharply: Reuters cited a 41% rise in PJM wholesale prices and a 23% decline in Mid-Columbia prices.
• Renewables are projected to rise from 24% of generation in 2025 to 25% in 2026 and 27% in 2027.
• Coal’s share is projected to fall from 17% to 16% and then 15%, while natural gas remains around 40% in 2026 and 39% in 2027; nuclear is around 18%.

Why AI changes the electricity equation

Training and running large AI models requires dense computing infrastructure that operates for long periods and can consume substantial electricity. The power requirement is not only the servers themselves; cooling, networking and backup systems add load. When large data centres cluster in the same region, local grid capacity can become the binding constraint even if national generation is sufficient.

This is why utilities, grid operators and technology companies are signing long-duration power agreements and discussing new gas, renewable and nuclear capacity. Electricity availability can determine where a data centre is built just as much as land, fibre connectivity or tax incentives.

Simple capacity example

A 1-gigawatt data-centre campus running at an average 80% load would use about 7.0 billion kWh in a year before considering some ancillary losses. That is a meaningful amount of demand concentrated in one location. If several such campuses arrive faster than transmission can be built, the local market can see higher congestion and capacity costs even if national electricity supply is growing.

Why regional prices can move in opposite directions

The national average can hide very different regional conditions. PJM serves a large area with significant data-centre growth and has faced tighter capacity conditions, while the Pacific Northwest can have different hydro availability, transmission and demand dynamics. A 41% rise in one market and a 23% decline in another shows why investors should not treat “U.S. power prices” as one uniform benchmark.

For utilities and data-centre developers, location affects interconnection timing, power purchase prices and the cost of backup capacity.

Investment and credit implications

Higher demand can support regulated utility capital expenditure, generation investment and grid-equipment orders. But more capex also means more financing. Companies with large construction pipelines must fund transmission, transformers, substations and generation years before all customer revenue arrives.

Lenders therefore need to distinguish contracted data-centre demand from speculative projects. A signed long-term offtake agreement with a highly rated counterparty is different from building capacity based on an expected AI boom without firm commitments.

What not to misunderstand

Record electricity demand does not mean the U.S. is running out of power nationally, nor does it guarantee higher utility profits. Regulated utilities often recover investment through approved rates, while competitive generators face market-price risk. The generation mix forecasts are also projections that can change with fuel prices, weather, project delays and policy.

What to watch next

Watch grid interconnection queues, data-centre power contracts, gas and nuclear project announcements, transformer lead times, capacity-market prices and actual versus forecast demand. For AI investors, power availability is becoming a real capacity constraint that belongs in valuation models alongside chips and networking.

Finin2min bottom line

AI infrastructure is pulling electricity into the centre of the technology investment cycle. The next phase of AI spending will increasingly be judged not only by compute demand but by whether power, grids and financing can scale fast enough.

Source

• *Controlling source:** U.S. Energy Information Administration — Short-Term Energy Outlook
• *Source reference:** EIA STEO, 6 Oct 2026 — demand 4,288 BkWh in 2026 and 4,356 BkWh in 2027; changing generation mix.
• *Source URL:** https://www.eia.gov/outlooks/steo/report/elec_coal_renew.php

Disclaimer

Educational and informational content only. Not investment, tax, legal or accounting advice. Market conditions, company disclosures and regulatory positions can change; readers should verify the latest controlling source before acting.