Record US power demand brings AI-energy nexus into sharp focus
Electricity consumption is set to hit more record highs in the US this year and in 2027, mostly due to data centres serving AI and cryptocurrencies, according to the Energy Information Administration (EIA).
According to the EIA's Short-Term Energy Outlook, power demand set a second consecutive annual record at 4,195 billion kWh in 2025, with projected rises to 4,248 billion kWh in 2026 and 4,379 billion kWh in 2027.
As well as data centre demand, the EIA says the increase is also driven by homes and businesses using more electricity and less fossil fuels for heat and transportation. However, it says power demand growth is led by the commercial sector, which is expected to outpace residential demand in 2027 for the first time.
Data centre energy appetite
The International Energy Agency (IEA) says data centres globally currently consume 485-500 TWh of power annually, accounting for about 2% of global electricity demand.
This rise in global demand reflects a rapidly increasing trajectory, primarily driven by the AI boom. As a result, projections indicate AI-specific hardware consumption will triple by 2030, nearly doubling overall data centre power use in that period.
The IEA says the AI-energy nexus is continuing to evolve rapidly, with capital expenditure by the largest technology companies exceeding $400 billion in 2025 and anticipated to rise a further 75% this year, fuelling data centre growth.
“Capital expenditure of just five technology companies is now larger than global investment in oil and natural gas production,” said the agency.
IEA satellite tracking found cutting-edge AI data centres more than tripled capacity in the past 18 months.
AI efficiencies in context
The IEA said worldwide data centre electricity demand rose by 17% in 2025, in line with its projections.
AI energy efficiency per task is improving at a rate "unprecedented in energy history," due to software and hardware advances.
However, this is offset by the increasing use of new energy-intensive AI applications, such as video generation, reasoning, and agentic tasks, which the IEA says can consume thousands of times more energy than simple text generation.
Managing bottlenecks
The agency said bottlenecks across energy supply chains have tightened since its last report amid a wave of data centre applications and a broader trend of rapid load growth and electrification, as the AI ecosystem scrambles for electricity and grid connections.
“An individual server rack within an advanced data centre is only the size of a large refrigerator, but by 2027 it could have peak power demand equivalent to that of 65 households,” the report noted.
The IEA suggested a possible upside in its mid- to longer-term outlook for data centre power demand, driven by investments to relieve bottlenecks across energy equipment and chip manufacturing.
On-site gas power is emerging in the US as developers bypass slow grid connections, with 15–27 GW expected by 2030, though the IEA cites uncertainties, such as a global gas turbine supply crunch.
Driving power innovation
Scaled battery storage could be critical to ensuring reliable supply to AI training and model-use data centres that induce large, rapid power swings.
The IEA predicts that 20-25 GW of global battery storage could be installed by 2030. With the right incentives, this could become a grid asset in the broader AI-driven acceleration of electricity sector deployment and innovation.
The agency cautioned that data centres could be a prominent flashpoint for concerns around energy prices.
But the IEA report also highlighted that AI has potential as an important tool in enhancing energy security and sustainability, such as optimising existing grid capacity.
However, it warned that the energy sector is yet to fully seize these opportunities.
An IEA survey of companies found that a lack of digital skills was the single largest barrier to greater AI adoption in the energy sector, while globally, less than half of energy demand was covered by policy frameworks promoting AI uptake in the sector.
