Google's AI boom pushed electricity usage up 37% — but carbon emissions still fell

There’s a paradox at the heart of Big Tech’s AI buildout, and Google’s latest environment report lays it out in stark numbers.

The company’s electricity consumption surged 37% in 2025, hitting an all-time high. The culprit is obvious: the massive expansion of AI infrastructure — more data centers, more TPUs, more GPUs running around the clock. Yet Google’s operational carbon emissions went in the opposite direction, falling 2% from the previous year.

How? Google bought enough renewable energy to cover every kilowatt-hour it consumed — for the ninth year running. It pushed data center efficiency well past industry baselines. And it rethought the entire AI stack, from the silicon up to the scheduling software that decides which model runs where and when.

Kate Brandt, Google’s chief sustainability officer, described the dynamic as a “decoupling” of business growth from emissions growth. The company’s data centers now deliver about 1.83 times more compute per unit of energy than the industry average — a gap that widens at scale.

Google also introduced what it calls the “AI Stack” framework, arguing that AI energy efficiency can’t be solved at the chip level alone. It takes coordinated work across hardware, system software, model architecture, and compute resource scheduling. On the hardware side, Google has been steadily improving the performance-per-watt of its in-house TPU chips. Compiler optimizations, better memory management, and smarter compute scheduling all feed into higher utilization rates.

At the data center level, the playbook is more familiar but no less effective: lower PUE scores, denser server packing, smarter building design, longer hardware lifecycles, and better component recycling.

On the energy procurement front, Google signed more than 12 gigawatts of new renewable energy purchase agreements in 2025 alone. Since 2010, the company has signed over 240 green power contracts totaling roughly 35 gigawatts of installed capacity. According to Google, these efforts cumulatively avoided about 58 million tons of CO2 emissions in 2025.

There’s a second, less visible number in the report. Google estimates that its AI-powered features — smarter route optimization in Maps, intelligent temperature control in Nest thermostats — helped external users avoid roughly 41 million tons of CO2 last year. That’s about three times Google’s own annual operational emissions.

The bottom line is straightforward: AI is power-hungry, but a well-designed grid-sourcing and efficiency strategy can keep its carbon footprint from exploding. Google’s numbers show it’s possible — if you’re willing to invest at the scale of a small country’s energy budget.