BloombergNEF: U.S. Data Centers to Consume 20% of National Electricity by 2035, Driven by AI Surge
Summary
- • BloombergNEF projects data centers will account for 20% of U.S. electricity generation by 2035 — four times today's share
- • AI compute is the primary driver, with data center capacity projected to nearly 200 gigawatts, nearly half devoted to AI training and inference
- • BNEF's 2035 estimate is already 83% higher than its own December 2025 forecast, with EPRI doubling its estimate and S&P raising theirs by a third
- • Grid stress is mounting: PJM electricity prices rose 76% in one year, AEP threatened to leave the interconnection, and PJM paused connections for four years
Details
4x U.S. Electricity Growth by 2035
BloombergNEF projects U.S. data center electricity consumption will quadruple by 2035, reaching 20% of all U.S. electricity generation
200GW Data Center Capacity
AI-driven data center capacity expected to reach ~200 gigawatts over the next decade; nearly half dedicated to AI training and inference
83% Forecast Revision in 7 Months
BNEF's 2035 electricity demand estimate is 83% higher than its December 2025 projection, reflecting the accelerating pace of AI infrastructure buildout
Cross-Agency Consensus Revision
EPRI (electrical nonprofit) more than doubled its 2024 estimate; S&P's forecast rose by more than a third between October and April — systematic underestimation across agencies
PJM Grid Under Extreme Strain
PJM Interconnection (Virginia to Illinois) will route 34% of its electricity to data centers; it paused new connection applications for four years and AEP threatened to leave
76% Electricity Price Surge in PJM
Supply-demand imbalance in PJM pushed electricity prices up 76% over the past year — before the bulk of new AI data centers come online
Global AI Electricity Demand
Worldwide, aggressive AI adoption could add 1,935 TWh of new electricity demand by 2033 — nearly equal to India's entire annual electricity consumption
U.S. Hosts 64% of AI Chips by 2033
The U.S. is expected to host 64% of global AI chip capacity by power demand by 2033, making it the concentrated focal point for energy infrastructure pressure
Source: BloombergNEF report (July 2026). Grid data from PJM Interconnection and ERCOT. Cross-referenced with EPRI and S&P Global forecasts.
What This Means
AI infrastructure is transforming the U.S. energy landscape at a pace that grid operators, energy planners, and policymakers are struggling to match. The systematic upward revision of electricity demand forecasts across BloombergNEF, EPRI, and S&P indicates that current energy policy and grid investments are being planned against already-outdated baselines — a compounding problem as build timelines for new generating capacity stretch years. Strained interconnection queues, rising electricity prices, and the threat of utility withdrawals are early warning signs of structural bottlenecks that could constrain AI development timelines or impose significant costs on regions that host data centers. This is now as much an energy and infrastructure policy story as it is a technology story, and the 83% revision in just seven months suggests more forecast upgrades are coming.
Sentiment
Concerned about grid bottlenecks and brownout risks, with strong focus on energy infrastructure investment opportunities
“U.S. data centers are projected to consume roughly 20% of the country’s electricity by 2035, up from 5.9% today, according to BloombergNEF. BNEF now expects data-center power capacity to reach 194 gigawatts by 2035, an 83% increase from its forecast just seven months ago. The AI race is no longer only about chips. It is becoming a race for electricity, grid capacity and energy infrastructure.”
“Data centers jump from 5.9% to 20% of U.S. electricity by 2035. That’s 194 gigawatts of new demand in 11 years. Grid infrastructure can’t scale that fast. PJM, Ercot, and Miso are already constrained. You’re looking at rolling brownouts in AI-dense regions unless transmission gets rebuilt yesterday. Energy stocks and utilities are the real play here.”
“The US heatwave just exposed the real AI bottleneck. The largest grid operator asked for the power to order data centers off the grid within 15 minutes during emergencies. Data centers eat 4% of US power today — heading to 9% by 2030. Models aren’t the constraint. Chips aren’t the constraint. Electrons are. The AI trade is quietly becoming an energy trade.”
“BloombergNEF now expects U.S. data-center power demand to reach 194GW by 2035, up 83% from its previous forecast. Data centers currently use 5.9% of U.S. electricity, rising to around 12% by 2030 and 20% by 2035. Even with faster grid connections and on-site gas generation, BloombergNEF still sees a 19GW supply shortfall by 2035. AI infrastructure is expanding faster than the grid.”
Split
~65/35: Infrastructure alarm (grid delays, brownouts) vs. investment opportunity (utilities, nuclear, power plays).