Moody's: U.S. AI Power Demand Requires $110 Billion to Build 45GW of New Power Plants
nashnova research
Moody's Ratings says the U.S. must spend $110 billion to build 45 GW of new power generation by 2030 to keep up with AI and cloud-driven electricity demand — the largest expansion in decades, now facing a rising regulatory drag that could delay the entire buildout.
Where does 45 GW of new power come from?
Moody's estimates the U.S. needs 45 GW of new generation capacity by 2030, at a total cost of roughly $110 billion.
More than 30 GW will come from natural-gas plants, requiring about 4 billion cubic feet of incremental gas demand. The rest is mostly solar and storage; nuclear restarts account for less than 5%.
This means → natural gas is the dominant fuel for this buildout. Clean-energy alternatives alone cannot close the gap in the near term.
Why will data-center power consumption double?
The IEA projects U.S. data-center electricity use will reach 426 billion kWh by 2030.
Data centers' share of total U.S. power demand will jump from 5% in 2025 to 10% — a doubling in five years.
In plain terms = AI training and cloud computing are extraordinarily power-hungry. Within five years, data centers go from a minor grid user to a one-tenth-scale consumer of the entire country's electricity.
Who picks up the $25–30 billion annual tab?
Moody's senior VP Ryan Wobbrock says new plant construction alone could add $25–30 billion per year to U.S. power costs.
Data-center operators will directly absorb up to $15 billion of that — some are building their own plants inside dedicated AI campuses. These "behind-the-meter" projects (on-site generation that bypasses the public grid) are expected to account for roughly 30% of total buildout.
How the remaining cost reaches ordinary consumers depends on regional rate-setting procedures and cost-allocation rules.
Could regulatory friction slow everything down?
The report notes that federal and state regulators are demanding more time to review rate structures, creating potential delay risk for new data-center development.
Wobbrock writes: "The pace of data-center development is increasingly outstripping the grid's ability to expand generation, transmission, and interconnection capacity."
This reflects a deeper tension: the AI industry needs speed, but power infrastructure — permitting, construction, grid integration — is structurally slow. Gas-fired generation, after two decades of depressed demand, faces long lead times just to restart its build pipeline. Whether these regulatory frictions resolve before 2030 is the key test for this investment wave.
市场有风险，内容仅供研究参考，不构成投资建议。