US AI boom needs $110bn of new power plants, Moody’s says

America’s data centre boom will need $110 billion to build 45 gigawatts of new power generation through 2030, according to an analysis by Moody’s Ratings.

The vast majority of that new supply, more than 30 gigawatts, will come from natural gas-fired units and represents about 4 billion cubic feet of incremental gas supply, according to a report Monday.

The bulk of the remaining supply will come from solar and energy storage, with nuclear restarts accounting for less than 5%. One gigawatt is roughly equivalent to the output of a traditional nuclear reactor.

The projection is based on the International Energy Agency’s forecast that data centres will consume 426 terawatt-hours in 2030, doubling its share of US electricity usage to 10% from 2025 levels. 

Read: The $3 trillion AI data center build-out becomes all-consuming for debt markets

ADVERTISEMENT

CONTINUE READING BELOW

Moody’s forecast quantifies what it would take to produce enough megawatts to feed the data centres needed for artificial intelligence and cloud computing that are driving the biggest expansion in US electricity demand in decades, fuelling a massive build out of infrastructure to support it.

The cost of America’s AI build-out

The White House has called the AI race a national imperative, backing some of the world’s biggest companies as they compete for potentially enormous profits.

But the build out comes at a cost, drawing opposition from Americans worried that the average household and business is subsidising AI through rising power bills, pollution, water shortages and threats to public lands and private property.

The $110 billion build-out of new power plants alone could add $25 billion to $30 billion a year to electricity costs in the US; how that is passed on to consumers will depend on the rate-making process and other rules for allocating costs in each area, Ryan Wobbrock, senior vice president of ratings at Moody’s Ratings’ global infrastructure finance group, said in an email.

Read: AI’s volatile power demand is damaging its own data centres

Data centres will pay for as much as $15 billion of those costs directly, because they are paying for power plants to be built on their AI campuses, he said. These behind-the-meter projects at data centres represent about 30% of the total build-out through 2030.

ADVERTISEMENT:

CONTINUE READING BELOW

After two decades of lacklustre demand growth, the US has been struggling to jump-start the construction of gas generators in an industry that typically moves with years of lead time.

These gas generators are coveted because they can be ordered to ramp up, unlike renewables that are beholden to sunshine and wind speeds.

Amid affordability concerns, “federal and state politicians and regulators are requiring more time to review and structure rates, creating a new source of potential delays for new data centre developments,” according to the report led by Wobbrock.

“The pace of data centre development is increasingly exceeding the speed at which the electric grid can add generation, transmission and interconnection capacity.”

© 2026 Bloomberg

#boom #110bn #power #plants #Moodys