Unleashing the Power of Generative AI: Transforming Business Insights

Quick Summary

• AI data centers in the US could account for about 20% of national electricity consumption by 2035, up from roughly 6% today, per BloombergNEF.
• Globally, the International Energy Agency projects data center electricity demand will roughly double to 945 terawatt-hours by 2030, on its way to around 1,200 terawatt-hours by 2035.
• Rising demand is already pushing up electricity prices in data-center-heavy regions like Virginia and Texas.
• US data centers consumed about 17.4 billion gallons of water directly in 2023, and AI-specific water demand could climb into the hundreds of billions of gallons annually by 2030.
• Microsoft, Amazon, Google, and Meta have each signed nuclear power deals in the past two years to secure carbon-free electricity for AI.
• Local opposition to new data centers is growing fast, driven largely by concerns over power costs, water use, and grid strain.

The Two Numbers Behind AI Data Centers’ Electricity Use

You’ve probably seen some version of the claim that AI could “quadruple” electricity use. It’s true, but it’s really two different numbers getting mashed together, and the distinction matters if you want to understand what’s actually coming.

The International Energy Agency’s Energy and AI report warns that AI-optimized data centers could become a major driver of electricity demand growth, with consumption expected to more than quadruple by 2030. The report estimates global data center electricity use could reach around 945 terawatt-hours as AI adoption expands.  Separately, energy and technology research firm BloombergNEF projects that total U.S. data center electricity demand, including but not limited to AI workloads, could reach roughly four times today’s level by 2035. The analysis estimates that data centers could represent about 20% of total U.S. electricity consumption, compared with less than 6% today.

Both numbers are real. They’re just answering slightly different questions, one about AI workloads specifically and globally by 2030, the other about all US data centers by 2035. Treat any single “quadruple” headline, including this one, as shorthand for a genuinely fast-moving target. BloombergNEF itself revised its own 2035 estimate upward by 83% compared to its forecast from just seven months earlier.

Training One Model Can Use More Power Than a Small Town

A single Nvidia H100 GPU, the chip most commonly used to train and run large AI models, draws around 700 watts on its own. Multiply that across the tens of thousands of chips packed into a modern AI training cluster, running nonstop for weeks or months, and you get facilities that behave less like traditional data centers and more like industrial power consumers.

A conventional data center, the kind that hosts cloud storage or video streaming, draws roughly as much electricity as 10,000 to 25,000 households, according to IEA estimates. AI-focused facilities pack far more power-hungry hardware into the same footprint. A rack of conventional servers dissipates 5 to 15 kilowatts of heat. A rack loaded with the latest AI accelerators can dissipate 40 to 120 kilowatts, and newer liquid-cooled designs are pushing toward 200 kilowatts per rack. That heat has to go somewhere, which is a big part of why the water story below matters just as much as the electricity story.

The Grid Is Already Feeling It

This isn’t a 2035 problem waiting to happen. It’s showing up on electricity bills right now. In the mid-Atlantic region served by grid operator PJM, where many of the largest AI data center campuses are concentrated, electricity prices have risen 76% over the past year as new data center connections have piled pressure onto an already tight grid. Data centers represented 38% of the charges in PJM’s most recent capacity auction.

That strain is starting to trigger real friction. One utility, American Electric Power, has threatened to pull out of the region’s interconnection process altogether over concerns about how fast new data center demand is being added. And communities are pushing back directly: Data Center Watch tracked more than 75 proposed projects, worth an estimated $130 billion, blocked by local opposition in just the first three months of 2026, matching the total for all of 2025 in a single quarter.

The Water Problem Nobody’s Talking About

Electricity gets most of the headlines, but water is arguably the more locally acute problem, especially since many new data centers are being built in drought-prone regions like Arizona and Texas.

Here’s the scale of it:

• US data centers directly consumed about 17.4 billion gallons of water in 2023, the most recent year with comprehensive figures, roughly equivalent to the water use of 160,000 American households.
• AI-specific water demand is projected to grow far faster than that baseline: one industry estimate puts US AI server water consumption at 200 to 300 billion gallons annually between 2024 and 2030.
• In Texas alone, a study from the Houston Advanced Research Center and the University of Houston found data centers are projected to use 49 billion gallons in 2025, rising to as much as 399 billion gallons by 2030. Researchers compared that to draining Lake Mead, the largest reservoir in the US, by more than 16 feet in a single year.
• Amazon disclosed in June 2026 that its global data centers consumed 2.5 billion gallons of water in 2025, the first time the company has published that figure. Google has made similar disclosures for its own facilities in prior years.

That scale can be misleading on its own, though. Some of it is direct cooling water, lost to evaporation in the towers that keep server racks from overheating. But a 2026 analysis by Xylem and Global Water Intelligence found that on-site cooling actually accounts for a relatively small share, about 4%, of AI’s total projected water footprint through 2050. The bigger drivers are the power plants generating the electricity in the first place (roughly 54%) and semiconductor manufacturing (about 42%). In other words, even a data center with a highly efficient, closed-loop cooling system still carries a substantial water footprint upstream, wherever its electricity comes from.

Big Tech’s Answer: Go Nuclear

Facing both a power shortage and pressure to keep AI’s carbon footprint in check, the largest AI infrastructure operators have converged on the same solution over the past two years: nuclear power, which provides steady, round-the-clock, carbon-free electricity that solar and wind can’t match without significant battery storage.

Microsoft signed a 20-year, roughly $16 billion power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1 in Pennsylvania, the reactor now renamed the Crane Clean Energy Center, targeting commercial operation around 2027. Amazon invested more than $650 million to acquire a data center campus next to the Susquehanna nuclear plant from Talen Energy, and has since committed additional capital toward expanding that site and toward small modular reactor developer X-energy. Google signed a deal with Kairos Power for up to 500 megawatts of small modular reactors, with the first site expected online by 2030. Meta has issued requests for proposals covering 1 to 4 gigawatts of new nuclear generation and signed a long-term agreement to buy power from Illinois’s Clinton Clean Energy Center.

Across the industry, tech companies have now committed to nearly 10 gigawatts of new or restarted nuclear capacity in roughly the past year, according to industry trackers. It’s a striking reversal for an industry that, a decade ago, was investing almost exclusively in wind and solar.

Fossil Fuels Are Still Filling the Gap

Nuclear deals get headlines because they’re new and dramatic, but they won’t come online for years. In the meantime, most of the immediate gap is still being filled by fossil fuels. Roughly 40% of the electricity used in data centers in 2024 came from natural gas, followed by renewables at about 24%, nuclear at 20%, and coal at 15%, according to Brookings’ analysis of the sector.

That mix is expected to shift only gradually. More than half of new global data center capacity built between 2023 and 2035 is expected to come from renewable sources, but because existing fossil fuel plants remain cheaper to run in the near term, roughly 64% of the incremental electricity generation added over that same period is still projected to come from fossil fuels. In other words, renewable capacity is growing, but it’s not growing fast enough on its own to keep pace with AI’s demand curve, which is why the industry is now betting so heavily on nuclear as a bridge.

Who Ends Up Paying For This

The clearest real-world effect so far is on electricity prices in regions where data centers are concentrated, and that effect is likely to spread as more capacity comes online in more states. Utilities in these areas face a genuine balancing act: data centers bring jobs, tax revenue, and infrastructure investment, but they also add load that existing transmission systems weren’t built to handle, and the cost of upgrading that infrastructure often shows up on everyone’s bill, not just the data center operator’s.

The environmental picture is similarly mixed. AI data centers aren’t inherently more polluting than any other large industrial electricity user, but their growth is happening fast enough, and concentrated enough in specific regions, that local grids, water systems, and communities are absorbing the impact before longer-term fixes like nuclear capacity or expanded renewable generation are ready. That timing gap, not the total electricity number itself, is the part of this story most worth watching over the next few years.

Discover how AI is reshaping technology, business, and healthcare—without the hype.

Visit InfluenceOfAI.com for easy-to-understand insights, expert analysis, and real-world applications of artificial intelligence. From the latest tools to emerging trends, we help you navigate the AI landscape with clarity and confidence

Frequently Asked Questions