Data centers powering AI are straining the grid faster than planners can build capacity, putting reliable 24/7 electricity—and the energy mix that provides it—squarely on the line.
Story Highlights
- U.S. data center power demand is set to more than double by 2035, potentially reaching about 8–9% of U.S. electricity use.
- Global data center electricity is ~2% in 2025 and could roughly double by 2030—far below some inflated claims.
- Reliability needs firm capacity; utilities are adding gas peakers and exploring nuclear while pairing renewables with storage.
- Slow permitting and long build times risk higher costs and local grid stress if supply lags AI demand.
How fast AI-driven data centers are changing U.S. electricity demand
BloombergNEF projects U.S. data center power demand will rise from roughly 35 GW in 2024 to about 78 GW by 2035, with data centers reaching an estimated 8.6% of national electricity use by the mid‑2030s. Deloitte places global data centers at around 2% of electricity in 2025 and expects usage to roughly double by 2030. MIT, citing the IEA, underscores a potential doubling of global data center electricity demand between 2022 and 2026 as AI scales.
Claims that data centers will consume 20% of global electricity by the early 2030s are not supported by these mainstream benchmarks. Instead, the trajectory shows rapid growth but still a minority share of global demand, with efficiency improvements helping temper total load even as AI workloads expand. This distinction matters for honest planning: overstating the scale can mislead policy, while understating the pace can jeopardize reliability and drive sudden rate shocks.
US DATACENTER POWER DEMAND WILL DOUBLE BY 2028 — per S&P Global Market Intelligence.
Utility power demand from datacenters is forecast to hit 58 GW in 2025, up 23% from 2024, and nearly 95 GW by 2028. By 2030, demand could reach over 117 GW, a 147% jump from last year. pic.twitter.com/LT2rbDz9Tu
— Wall St Engine (@wallstengine) August 5, 2025
What powers 24/7 reliability: the evolving mix of firm and clean resources
Utilities are fast‑tracking firm capacity additions, including gas peakers for near‑term flexibility, while some investigate small modular reactors for future baseload; both pathways face cost and permitting constraints. In parallel, hyperscalers sign long‑term contracts that bundle renewables with storage and pursue 24/7 carbon‑free strategies when feasible. The practical outcome is a portfolio approach that values dispatchability without abandoning cost‑effective clean energy procurement.
Reliability requirements for clustered, always‑on campuses—often exceeding 100 MW—create tight siting pressures on stressed grid nodes. Transmission constraints, interconnection queues, and local community impacts force careful sequencing of generation, network upgrades, and cooling infrastructure. BNEF estimates roughly seven years from site selection to full operation on average in the U.S., making today’s permitting decisions decisive for the early 2030s buildout and the prices ratepayers will face.
Watch: AI Data Centers to Double Electricity Demand by 2030 | Certrec Energetic News
Cutting through spin: separating facts from agenda-driven claims
The narrative that “renewables are simply not up to the task” overstates the case; leading operators are combining variable renewables, storage, and contracts with firm resources, including natural gas and potentially nuclear, to meet 24/7 needs. Conversely, rosy assumptions that efficiency alone will offset AI demand growth are not borne out in current forecasts. The credible midpoint: demand is rising fast, portfolios will diversify, and firm capacity remains critical to safeguard reliability.
For constitution-minded readers wary of government overreach and costly missteps, sound planning grounded in verified numbers is the best defense against avoidable blackouts, ballooning power bills, and rushed mandates that miss the reliability mark.
Sources:
Generative AI power consumption creates need for more sustainable data centers
Power for AI: Easier Said Than Built
AI has high data center energy costs. There are solutions.
Data Center Energy and AI in 2025
Commercial computers could account for 20% of commercial electricity consumption by 2050
