Data centers expected to use 4x more electricity by 2035
TechCrunch Tim De Chant ● Covered by 2 sources
Data centers could gobble up a fifth of all US electricity by 2035, four times today's usage. AI's power appetite keeps outpacing every forecast, and the grid is already buckling.
Based on reporting by TechCrunch, Tim De Chant — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Every few months another energy forecaster tears up its spreadsheet and writes a bigger number. The latest comes from BloombergNEF, which now says data centers will consume roughly 20% of U.S. electricity by 2035 — quadruple current levels. That estimate is already 83% higher than what the same firm predicted back in December. This isn't a one-off correction; EPRI has more than doubled its 2024 projection, and S&P bumped its own forecast up over a third in just six months. Nobody modeling this stuff seems able to keep pace with how fast AI compute is scaling.
The report pegs total U.S. data center capacity at nearly 200 gigawatts within a decade, with close to half of that dedicated purely to training and running AI models. And the country isn't ceding ground globally either — by 2033 the U.S. is expected to house 64% of the world's AI chips, measured by power draw. Worldwide, if adoption keeps accelerating, data centers could add nearly 1,935 terawatt-hours of new demand by then, which is close to what all of India consumes in a year.
The real strain shows up regionally. PJM Interconnection, the grid network stretching from Virginia to Illinois, is projected to send 34% of its electricity to data centers. ERCOT in Texas isn't far behind at 22%. PJM has already been buckling under the pressure — it froze new generator connection requests for four years, only reopening the queue this past April. Things got tense enough that American Electric Power, one of PJM's member utilities, floated the idea of leaving the interconnection entirely. Meanwhile electricity prices in the region have jumped 76% over the last year, a direct symptom of supply struggling to catch demand.
Yet none of this friction is slowing data center operators down. They accounted for 38% of the charges in PJM's most recent capacity auction, proof that companies are still racing to secure grid access even as costs and congestion mount. The pattern suggests a strange standoff: utilities and grid operators sounding alarms about instability, while the AI industry keeps building as if the power will simply show up when needed.
What these dueling forecasts really capture is how badly infrastructure planning is lagging behind AI's growth curve. Grids built for decades of gradual, predictable demand are now being asked to double or quadruple capacity in a fraction of that time, and the bill — literally, in rising electricity rates — is landing on ordinary customers long before the industry sorts out how to generate all that extra power responsibly.
My take — AI-written commentary, not fact-checked reporting
I'll believe these forecasts are anywhere near final once one of them actually holds for longer than a quarter. What strikes me isn't the raw numbers but the fact that AI companies keep signing up for grid capacity that clearly isn't there yet, betting someone else will solve the generation problem while consumers eat the price hikes. That's not innovation, that's externalizing your infrastructure debt onto the public — and it's exactly the kind of unglamorous mess that gets ignored until the lights flicker.
Read more about this at: TechCrunch
