Electricity Up 37% — Google Publishes Its Own 'AI Bill'

On June 30, Google released its 11th annual Environmental Report. And buried inside was one number it couldn't hide: in 2025, its electricity use jumped 37% year over year. That's the single largest one-year increase in Google's history. Compared with 2019, its electricity consumption has ballooned by more than 250%. The reason converges on a single word. AI.

Here's the interesting part: Google didn't hide it. It put the number right up front. Normally a company files an environmental report to brag about how much it cut. This one reads almost like the opposite — a document admitting how much more it consumed. In effect, one of the biggest AI companies on the planet confirmed, in its own words, just how much electricity the AI boom is swallowing.

But then the story twists. Despite burning through all that extra power, Google's own operational emissions (Scope 1 and 2) actually fell 2%. Meanwhile, its supply-chain emissions (Scope 3) grew 25%. Same company, same year — one bucket of emissions went down while the other went up. That split is the real heart of this report, and it compresses the entire dilemma facing Big Tech in the AI era.

In this piece I'll unpack why the numbers diverged like this, what "100% renewable matching" actually means, and what all of it changes from the standpoint of consumers, the AI industry, policymakers, and climate-minded readers.

The Cast: Google, the Data Centers, and the Grid

Let's set the stage. This story has three characters: Google, the data centers it's building at a furious pace, and the grid and energy suppliers that plug electricity into those data centers.

Google is smack in the middle of an AI infrastructure buildout. Training large models like Gemini and stuffing AI features into Search, YouTube, Cloud, and Android across the board has caused its compute needs to explode. More compute means more GPUs and TPUs, which means more data centers to run them, which means more electricity to keep those data centers humming 24/7. At the very end of that chain sits the grid. AI's ambition ultimately gets denominated in electricity.

The data center plays a double role here. It's both a "mouth that eats electricity" and a "construction site that emits carbon." The power a data center draws while running counts as operational emissions; the concrete, steel, and semiconductors poured into building it count as supply-chain emissions. The more and the faster Google builds data centers for AI, the more both kinds of burden grow at once.

The grid and energy suppliers are the variable Google can't fully control. No matter how many clean-energy contracts it signs, if the local grid where a data center is physically plugged in runs mostly on coal and gas, the electricity being consumed at that moment still emits carbon. The more Google sources semiconductors and expands infrastructure in regions with relatively "carbon-heavy" grids — Taiwan, Japan, Vietnam, India — the more this problem grows outside Google's own ledger. That's the key to why operational and supply-chain emissions diverged.

To put it simply: Google can turn the switch it directly pays to flip (operations) green fairly quickly. But the concrete, steel, and chips used to make that switch and that building (the supply chain) come out of someone else's factory on someone else's grid, so they turn green far more slowly. That difference in speed runs through the entire report.

What Happened: The Report's Core Numbers

Now let's get concrete. Google's 2026 Environmental Report covers 2025 performance. Here are the key metrics in one table.

Metric 2025 result Direction
Electricity use (YoY) +37% (largest single year ever) Surging
Electricity use (vs. 2019) +250% or more Surging
Operational emissions (Scope 1-2, YoY) -2% Falling
Supply-chain emissions (Scope 3, YoY) +25% Rising
Renewable energy matching 100% for 9th straight year Held
New clean-energy deals (2025) 12+ GW Expanding
Cumulative clean-energy deals (since 2010) 240+ deals / ~35 GW Expanding
Avoided emissions (2025) 58M+ metric tons CO2eq —

The first thing that jumps out is that 37%. Electricity use has never climbed this much in a single year in Google's history. Being up more than 250% versus 2019 means power consumption has more than 3.5x'd in just six years. It's the most honest gauge we have of how power-hungry AI training and inference really are.

Then comes the strange contrast. Google burned all that extra power, yet operational emissions fell 2%. How? Because Google matched 100% of the electricity it used with renewable-energy purchases for the ninth consecutive year, and signed more than 12 GW of new clean-energy deals in 2025 alone. In other words, "we used more, but we bought more to offset it," and operational emissions actually ticked down. Cumulatively since 2010, that's over 240 deals and roughly 35 GW of net-new clean energy secured.

On the other side sits the 25% rise in supply-chain emissions. This isn't electricity Google flips on directly — it's the emissions baked into making the AI infrastructure. Data center construction alone added about 2.3 million tons of CO2eq, much of it coming from semiconductor suppliers running on carbon-intensive grids in Taiwan, Japan, Vietnam, and India. As of 2025, supply-chain emissions make up nearly 80% of Google's total carbon footprint. Put differently: the emissions Google really has to wrestle with aren't in the switch it flips, but in the concrete and chips it orders.

So on a total basis, Google's overall emissions rose year over year. The 25% jump on the supply-chain side dwarfs the 2% trimmed on the operational side. The headline may be "electricity up 37%," but the genuinely hard problem is the Scope 3 buried underneath it.

What Each Side Gains and Loses: The Trade-offs

What Google gets out of this is clear: staying in the AI race. In today's Big Tech landscape, if you can't build data centers and can't lock down power, you fall out of the AI race entirely. From Google's seat, using 37% more electricity isn't a "cost" — it's the ticket to entry. Skip it and you don't get to play.

At the same time, Google is trying to protect an asset called reputation. It can build the narrative "we used a lot of power, but cut operational emissions and held 100% renewable matching for nine straight years." Half of that is genuinely true. Google really has spent serious money on clean-energy procurement, and 12 GW of new deals is on the scale of a small country's generation plan. But the other half — supply-chain emissions and rising absolute totals — doesn't get hidden by that narrative. That's the crux of the trade-off.

What do the grid and local communities gain and lose? Google's clean-energy deals pump capital into new solar, wind, and nuclear projects, pulling the whole grid greener. That's a clear plus. But at the same time, regions where data centers cluster shoulder surging power demand, upward pressure on electricity prices, transmission strain, and water use. AI's benefits get shared worldwide, while the power burden concentrates in specific places — an asymmetry worth naming.

And here's the thing you have to pin down: what "100% renewable matching" actually means. It does not mean "Google's data centers literally run on nothing but renewables 24 hours a day." It's closer to annual matching — "over the course of a year, we bought enough renewable energy somewhere to balance the ledger against our total consumption." In reality, at night or when the wind isn't blowing, fossil-fuel electricity still flows, and Google offsets it with surplus daytime renewable purchases. The stricter standard Google aspires to — "24/7 carbon-free energy" (CFE, hourly matching) — is far tougher than annual matching, and when absolute usage and absolute emissions keep climbing, the gap between the two can actually widen. Remember: a 100% matching ratio and the absolute quantity of carbon actually emitted are two different things.

Past Cases: Big Tech's Carbon Pledges, Wins and Failures

This picture isn't new. Big Tech's climate pledges have cycled through flashy targets and quiet retreats for the past decade.

Start with Google itself. It declared "carbon neutrality" back in 2007 and has hit annual 100% renewable matching since 2017. In 2020 it went further, setting the ambitious goal of running every data center and office on 24/7 carbon-free energy by 2030. Then the AI boom crashed head-on into that roadmap. Power demand spiked far faster than expected, and what was once a point of pride — that "net-zero moonshot" — now looks much farther away. This report reads, in effect, as a numeric admission that the goal has gotten harder.

Microsoft walked a similar path. In 2020 it set the industry's most aggressive target: "carbon negative" (absorbing more than it emits) by 2030. But as it aggressively scaled AI infrastructure, its emissions actually rose sharply, and the company itself has acknowledged the goal has become "harder" to reach. Amazon pledged net zero by 2040 under "The Climate Pledge," yet reining in absolute emissions amid cloud and logistics expansion remains an unsolved chore.

See the pattern? Operational emissions (the electricity you flip on) can be addressed relatively quickly by buying renewables, so there are lots of success stories there. But supply-chain emissions (steel, concrete, semiconductors, logistics) come out of other people's factories, so they're hard to control — and that's where the targets collapse. Big Tech's climate wins came mostly from Scope 2; the failures came mostly from Scope 3. Google's 2026 report is the newest and biggest version of that old pattern.

Rivals' Counterplay: The Energy Math of Microsoft, Amazon, and Meta

Google isn't alone with this problem — every Big Tech player doing AI is hitting the same wall. And each is responding a little differently.

Microsoft is betting big on nuclear. As AI demand exploded through its relationship with OpenAI, Microsoft went so far as to sign deals to revive shuttered reactors (most notably restarting Three Mile Island) in a bid to secure "always-on carbon-free power." Its read: renewables are intermittent and fall short for data centers that run 24/7. Microsoft is also one of the few companies that has publicly admitted AI drove its emissions up.

Amazon leans on scale. It's the world's largest corporate buyer of renewable energy and has jumped into nuclear investment too, including small modular reactors (SMRs). Still, Amazon is under its own emissions pressure amid AI and cloud expansion, and several analyses this summer grouped it alongside Google as a marquee case of "AI pushing emissions back up."

Meta has a slightly different flavor. As it builds mega data centers, Meta is pursuing large solar and wind contracts alongside nuclear procurement, and it stays exposed to controversy over local grid strain and electricity prices. All three companies hit the exact same choke points: renewable matching can manage operational emissions, but (1) the stricter 24/7 carbon-free standard and (2) the supply-chain emissions of building the infrastructure still don't get solved.

That's why the center of gravity in this competition is shifting. Where the contest used to be "who buys more renewables," it's now "who locks down 24/7 carbon-free power (nuclear, geothermal, long-duration storage) first." Google is already into small modular reactors and next-gen geothermal deals, and this report's 12 GW of new contracts is an extension of that direction. The real battlefield of the AI era is moving beyond model performance to who holds the "clean, always-on electricity."

So What Changes: Broken Down by Who You Are

If you're a consumer — you won't feel it much day to day, but the direction is worth knowing. AI features may look free, but enormous amounts of electricity churn behind them, and that cost gets passed somewhere eventually. If you live in a region packed with data centers, you may collide with it more directly through higher electricity bills or transmission and water issues. These are numbers worth recalling the next time you reach for AI and think, "this isn't actually free."

If you're in the AI industry — it's now certain that power has become a bottleneck as important as silicon. No matter how many GPUs you lock down, they're useless without the electricity and data center sites to run them. Going forward, Big Tech's competitiveness will hinge not just on models and chips but on "the ability to secure power" — and grabbing 24/7 carbon-free electricity first will rise to the center of strategy. Google publicly disclosing a 37% jump in electricity use reconfirmed for the whole industry just how real this bottleneck is.

If you're a policymaker — it's a signal that it's time to redesign regulation and incentives. There's now a clear need to reveal precisely how much a company's self-reported "100% renewable matching" diverges from its actual absolute emissions, and how to measure and assign responsibility for supply-chain (Scope 3) emissions has emerged as the core task. Real-world policy — grid expansion, streamlined clean-energy permitting, data center siting rules — is the true variable in the AI era's climate response.

If you're a climate-minded reader — you need to look at this more coldly. A "2% cut in operational emissions" is a genuine achievement, but making that the headline misses the whole picture. Absolute power use and absolute emissions keep rising, and supply-chain emissions make up 80% of the total. The accounting feat of a 100% matching ratio and the absolute quantity of carbon that actually went into the atmosphere are different stories. Seeing that gap honestly is the starting point for thinking about AI and climate together.

In one sentence: Google can turn the switch it flips green quickly, but the concrete, steel, and chips that make that switch and that building turn green far more slowly. And AI is widening that gap in speed a little more every year.

🥄 Three Things You're Probably Wondering

— How did operational emissions fall when electricity use rose 37%? The key is an accounting method called "matching." Google buys enough renewable energy to offset, on paper, the total electricity it actually used. So even as usage rises, if it grows clean-energy purchases proportionally (12 GW of new deals in 2025), operational emissions can drop. The catch: this is annual total-basis matching, so at night or when the wind is calm, fossil-fuel electricity still flows. That's why "100% matching" and "zero actual emissions" are not the same thing.

— So is Google unable to control the 25% jump in supply-chain emissions? Right now, mostly not. Supply-chain emissions come from the concrete and steel that build data centers and the semiconductor fabs that make the chips — that's someone else's factory on someone else's grid, so Google can't just flip that switch on and off. Semiconductor suppliers in carbon-intensive grids like Taiwan, Japan, Vietnam, and India make up a big share. As long as Google has to keep building AI infrastructure, these emissions are likely to keep growing alongside it for a while.

— Is this a Google-only problem, or industry-wide? Industry-wide. Microsoft, Amazon, and Meta are all seeing power and emissions spike because of AI, and each is diving into 24/7 carbon-free power like nuclear, SMRs, and geothermal. If anything, Google is on the more candid side for writing all of this openly into its report. The real thing to watch has shifted from "who buys more renewables" to "who grabs always-on clean electricity first."

References

Numbers and criteria are as of announcement and may change.