Archive · 23 Sep 2026
AI Power Demand Drives America’s Electricity Infrastructure Race
Artificial intelligence needs very large computer buildings called data centres.
These buildings use a lot of electricity to train and operate AI systems.
The United States expects electricity use to reach new records in 2026 and 2027.
Solar panels and batteries can add power quickly, but wires and grid equipment are also needed.
Natural-gas plants can provide dependable power soon, but they produce fossil-fuel emissions.
Nuclear plants can provide steady power, but building them takes many years.
Some data centres may generate electricity themselves instead of waiting for a grid connection.
The country is therefore trying several approaches at the same time.
All of them require more investment in the electricity system.
US electricity consumption is projected to reach records in 2026 and 2027, driven partly by AI data centres and broader electrification.
Solar and battery projects are expanding quickly, but new transmission lines and grid connections are needed to deliver their power.
Natural gas is emerging as a near-term reliability solution, though greater fossil-fuel use could complicate emissions goals.
Nuclear power offers continuous electricity over the longer term, while new reactors and small modular reactors face lengthy development challenges.
Data-centre developers are exploring on-site generation as the United States races to expand grid capacity and meet AI demand.
- Who
- US utilities, technology companies, policymakers, data-centre developers and electricity consumers.
- What
- The United States is expanding generation, on-site power and grid infrastructure to meet rapidly rising electricity demand from AI data centres and other electrification.
- When
- Electricity demand is projected to set records in 2026 and 2027; related infrastructure plans extend through 2035.
- Where
- Across the United States, with particular pressure in Texas and at major data-centre locations.
- Why
- AI data centres require very large, reliable amounts of electricity, while existing generation and grid connections may not be sufficient.
Faster, More Reliable Expansion
Lower Emissions and Consumer Protection
Natural gas versus emissions
Faster, More Reliable Expansion
Natural-gas plants can provide dependable electricity around the clock and respond relatively quickly to changes in demand, helping address AI’s near-term power needs.
Lower Emissions and Consumer Protection
Greater reliance on natural gas could make it harder for the United States to reduce power-sector emissions.
On-site generation versus grid planning
Faster, More Reliable Expansion
Data centres that generate power near their facilities could reduce dependence on slow grid connections and begin operating sooner.
Lower Emissions and Consumer Protection
A larger and better-coordinated electricity network is still necessary, and the growth of major private power projects could intensify infrastructure and resource pressures.
AI expansion versus household costs
Faster, More Reliable Expansion
Technology companies argue through the Ratepayer Protection Pledge that they will build, bring or buy generation and pay for power-delivery infrastructure serving their data centres.
Lower Emissions and Consumer Protection
A November 2025 survey found that 78% of respondents were concerned new data centres would increase household electricity bills; the pledge has no enforcement mechanisms, according to Carnegie.
- Projected US electricity consumption in 2026
- 4,270 billion kWh
- Projected US electricity consumption in 2027
- 4,349 billion kWh
- Planned 2026 utility-scale solar capacity
- About 43.4 GW
- Planned 2026 battery-storage capacity
- About 24 GW
- Natural gas share of generation
- About 40% in both 2026 and 2027
- Texas projects awaiting grid connection
- More than 470 GW of data-centre projects and other large energy users
- Projected US data-centre capacity in 2035
- 194 GW