One 2025 estimate says AI could need as much electricity within three years as it takes to power 22% of U.S. households — a scale industry insiders sometimes call “bragawatts.” That helps explain why cloud and chip companies are buying power and building huge data centers. Researchers comparing single-query costs with whole-fleet demand say the gap between promised clean power and the grids that actually supply data centers is wide, shaping costs, emissions and investment decisions.

How big is big?

Measuring AI’s power appetite starts small and scales fast. Analysts tallied the electricity used by individual chatbot queries, image generations and video renders, then aggregated those numbers and projected future demand. That exercise produced a striking headline figure: by one estimate in 2025, AI’s draw could equal enough electricity to power about 22% of U.S. households within three years.

That figure isn’t a single machine’s rating. It’s a system-level projection — the compute, cooling and continuous operations across scores of data centers that support models and services. And it’s why companies are planning multi-gigawatt builds and signing long-term power contracts. They need steady, round-the-clock capacity, not occasional bursts.

Why data centers demand constant power

“AI data centers need constant power, 24-7, 365 days a year,” said Rahul Mewawalla, chief executive officer of Mawson Infrastructure Group, which builds and operates high-density facilities that support AI workloads. That always-on requirement changes the power equation.

Unlike variable consumer loads, AI operations can’t simply pause when grid supplies dip. They run regardless of sunshine or wind. That makes intermittent renewable sources harder to rely on unless backed by storage or firm generation. The practical effect is that many AI facilities end up drawing from grids that still lean on fossil fuels.

Where the power comes from

Grid mixes matter. A preprint from Harvard’s T.H. Chan School of Public Health used carbon intensity — grams of CO2 per kilowatt-hour — to compare emissions footprints. It found the electricity used by many data centers had a carbon intensity about 48% higher than the U.S. average.

Part of that comes down to geography: major data-center clusters sit near grids with heavier use of natural gas and coal, including the mid-Atlantic region covering Virginia, West Virginia and Pennsylvania.

In 2024, fossil fuels including natural gas and coal made up just under 60% of U.S. electricity supply. Nuclear supplied roughly 20%, and the rest came from renewables and other sources. That mix shapes the emissions tied to AI operations because, so far, many facilities draw on whatever the local grid provides.

Industry responses: buy power, build or promise clean energy

Big cloud providers and tech firms have pitched a range of fixes:

  • Signing long-term power purchase agreements and building dedicated capacity.
  • Pursuing on-site generation and backing pledges to expand low-carbon sources globally, including commitments to increase nuclear capacity.
  • Facing near-term gaps with bridging projects — some of which have raised regulatory and environmental concerns in local communities.

For example, satellite imagery and local filings showed that a new supercomputing center for X near Memphis was using dozens of methane-fired generators to supplement grid power, a move that environmental groups say violated approvals and air rules.

The finance angle: investment, contracts and risk

AI’s power needs translate directly into capital decisions. Building multi-megawatt data centers requires site work, power-substation capacity and long-term power contracts. Those commitments shape how companies assess upfront capital, contracting risk and permitting hurdles.

Investors and energy buyers now weigh not just construction and operating costs but also the emissions profile of the grids that will supply these facilities. That calculus affects where data centers are sited, the structure of power deals and which projects win funding.

Related Articles

Companies are planning for near-constant loads even though fossil fuels still supply just under 60% of U.S. electricity — a reality that will affect costs, emissions and where investors place their bets. Some facilities have even turned to methane-fired generators, drawing regulatory and environmental scrutiny in local communities.

This article was created with AI assistance.