Surprising AI STRAIN: Will the Grid Survive?

U.S. data center electricity use—driven by AI—could more than double by 2035, reaching nearly 9% of the nation’s power, according to BloombergNEF, intensifying debates over energy supply, reliability, and long permitting timelines.

Story Highlights

U.S. data center power demand is forecast to more than double by 2035, reaching an estimated 8-9% of U.S. electricity use.
Global data center consumption is about 2% of electricity in 2025 and could roughly double by 2030.
The need for reliable power is driving utilities to add new capacity, including gas, nuclear, and renewables with storage.
Permitting and grid bottlenecks mean long timelines from site to operation, slowing the delivery of new power.

What Credible Forecasts Say About AI-Driven Electricity Demand

BloombergNEF projects U.S. data center load rising from roughly 35 GW in 2024 to about 78 GW by 2035, taking data centers from nearly 3.5% of nationwide electricity to approximately 8.6%. Deloitte estimates global data centers consume about 2% of electricity in 2025 and could roughly double by 2030. The MIT Sloan School of Management, citing the International Energy Agency (IEA), reports global data center electricity could double between 2022 and 2026 as AI workloads expand.

Some claims circulating—such as data centers hitting 20% of global electricity by 2030—do not align with these benchmarks. A 2023 assertion that AI data centers alone used 4.4% of U.S. power runs higher than leading estimates that place all U.S. data centers near the mid-single digits today. Planners should consider these figures when making investment decisions.

https://t.co/WOCX7Q7Jrt

In 2023, AI data centers consumed 4.4% of electricity in the US. By 2035, data centers “could account for 20% of global electricity use, putting an immense strain on power grids.”

There are more than 10,000 data centers globally, each one a huge…

— ChetTheJet (@TheJetNamedChet) August 10, 2025

Why Firm, 24/7 Power Is Central

Data centers demand continuous power with redundancy and strict uptime; operators are increasingly seeking carbon-free 24/7 supply. Utilities are fast-tracking capacity additions to meet clustered 100+ MW campus loads, evaluating gas speakers for near-term dispatchability and exploring nuclear options, including potential SMRs, for future low-carbon baseload. Developers are also expanding renewables paired with storage and 24/7 clean energy contracts, reflecting a portfolio approach that balances reliability, cost, and decarbonization.

Analysts’ outlooks show that operators are combining variable renewables, storage, and contractual structures with firm resources to meet reliability requirements. That mix varies by region, fuel prices, and siting constraints. In the near term, gas additions appear in several utility resource plans due to speed and controllability, while nuclear remains a potential longer-term pillar where licensing, timelines, and economics can be resolved.

The Bottlenecks: Siting, Interconnection, and Time-to-Power

BloombergNEF estimates roughly seven years from site selection to full operation for major data center projects in the U.S., including about five years pre-construction and over two years of building. These delays reflect transmission constraints, interconnection queues, local permitting, cooling and water considerations, and supply chain realities for power equipment. Such bottlenecks elevate the value of existing firm capacity and transmission access, while pushing buyers to pursue on-site generation and innovative procurement to secure timely, reliable electrons.

Competition for power is intensifying. Hyperscalers are signing long-term deals across the stack: clean energy PPAs with storage, exploratory nuclear offtake discussions, and on-site solutions where feasible. As energy costs become a larger share of AI service costs, efficiency gains—from chips to model architecture—matter. Techniques like Mixture-of-Experts and hardware roadmaps can moderate growth, but forecasts still show substantial net load increases through the 2030s, preserving the need for new firm capacity and grid upgrades.

Implications for Ratepayers, Communities, and Policy

Host communities may gain jobs and tax revenue but will have to debate land use, emissions of gas-fired, water for cooling, and noise. Utilities and regulators will face cost allocation choices for new plants and transmission lines. The U.S. EIA notes data centers’ high energy intensity drives additional ventilation and cooling demand, signaling broader system investments. Policies that accelerate permitting for reliable, affordable power—while preserving consumer protections—will shape whether America can lead in AI without sacrificing grid stability or household budgets.

Key takeaways for readers: demand is real, large, and sustained; reliability requires firm capacity in the mix; credible forecasts contradict the most extreme claims; and timelines are long, making permitting reform and all-of-the-above energy strategies crucial.