Google’s electricity consumption rises by 37% amid AI expansion

The deployment of artificial intelligence is already impacting Google’s energy bill. The company increased its electricity consumption by 37% in 2025, the largest annual increase recorded so far, according to its latest environmental report. The figure summarizes a tension affecting all major cloud providers: AI demands more data centers, more chips, increased cooling, and more energy—just as tech companies strive to meet their climate commitments.

Google maintains that it matched 100% of its annual electricity consumption with renewable energy purchases for the ninth consecutive year and reduced its operational emissions by 2% compared to the previous year. However, the company acknowledges the core issue: infrastructure construction for AI is advancing faster than grid decarbonization.

The distinction between these statements is important. Matching annual consumption with renewable purchases doesn’t mean every data center operates with clean energy at all hours. It means that, on an annual basis, Google purchases or contracts enough renewable energy to cover its consumption. Its more ambitious goal remains operating with carbon-free energy 24/7, which is much harder as demand grows at the current pace.

AI Breaks the Normal Growth Rhythm of Energy Demand

The 37% leap is not a minor increase within a company that already consumes electricity on a large scale. Data Center Dynamics reports that, following this increase, Google’s total electricity consumption has risen more than 250% since 2019. The growth is attributed mainly to AI demand, with more computing capacity for training, inference, products like Gemini, and cloud services for third parties.

Google tries to offset this pressure through efficiency and clean energy procurement. In 2025, it signed agreements for over 12 GW of new clean energy, the largest annual figure in its history, surpassing the combined total of the previous two years. The company highlights that between 2010 and 2025, it has signed over 240 agreements to buy nearly 35 GW of new clean energy.

The company also reports an average PUE of 1.09 across its data center fleet, a very efficient figure for the sector. PUE measures how much total energy an installation consumes relative to the energy delivered to IT equipment. Closer to 1 is better. The issue is that efficiency per unit does not always compensate for the overall volume increase: deploying many more servers, GPUs, and accelerators can still lead to significant total consumption growth.

Google offers another data point: the energy associated with a typical text query on Gemini decreased 33-fold within 12 months. This is a significant technical improvement, but it doesn’t prevent total consumption from rising if usage multiplies, models grow, and infrastructure expands.

Renewables, Nuclear, Hydropower, and Demand Response

The report also shows how the energy strategies of major hyperscalers are evolving. Google isn’t just purchasing renewable certificates; it’s signing long-term supply agreements, supporting new energy sources, and seeking flexibility to reduce load when the grid is strained.

In 2025, notable moves include collaboration with NextEra Energy to reactivate the Duane Arnold nuclear plant in Iowa, 600 MW; a framework agreement for 3 GW of hydropower with Brookfield in grid regions like PJM and MISO; and a fusion energy purchase agreement with Commonwealth Fusion Systems, hailed as the biggest of its kind to date.

Google also reports having integrated 1 GW of demand response capacity into long-term electric contracts in the U.S. Practically, this allows shifting or reducing machine learning loads during grid stress periods. It’s becoming an increasingly important piece: AI data centers not only consume a lot of energy but can concentrate significant loads in areas with existing grid constraints.

This point will gain importance. The debate on AI and sustainability will no longer only depend on how many solar or wind parks are contracted, but also on where the data centers are located, when they consume, if they can modulate load, and their impact on the local grid.

Supply Chain’s Carbon Footprint

Although Google reduced its operational emissions, its supply chain emissions grew 25% year-over-year. The company attributes part of this increase to the scale of new AI infrastructure and a supply chain in Asia-Pacific operating on grids with insufficient clean energy sources.

Data Center Dynamics adds that data center construction contributed about 2.3 million tons of CO₂ equivalent in 2025, with Google pointing to semiconductor suppliers operating on grids with limited access to clean energy in Taiwan, Japan, Vietnam, and India.

This detail is key because the AI footprint isn’t only at the data center plug. It’s also in chips, servers, HBM memory, advanced packaging, construction materials like steel and concrete, cooling systems, and all associated logistics. As hyperscalers buy more accelerators, supply chain sustainability becomes a core part of their own climate accounting.

Here’s a paradox: Google can procure more clean energy for its direct operations, but it has less control over the electricity used by its semiconductor suppliers. Much of the advanced chip supply chain is located in regions where electricity still has a high carbon footprint or where access to new clean energy isn’t growing fast enough to meet demand.

Water: The Other Limit of AI Infrastructure

Water consumption also increased. According to data from Data Center Dynamics, Google used 10.9 billion gallons in 2025, a 34% rise. The company states it replenished about 7.7 billion gallons through 165 water management projects across 97 watersheds—about 78% of its fresh water use, based on its official report.

This figure should be interpreted carefully. Replenishing water doesn’t mean returning exactly the same water, at the same place and time. These programs may improve watersheds, restore ecosystems, or offset consumption through local projects, but they don’t automatically reduce pressure on communities where data centers compete for water resources.

Google aims to be “water positive” by 2030, with a goal to replenish more water than it consumes. It’s an ambitious target, but the 34% increase in consumption shows that the challenge grows alongside infrastructure expansion.

AI’s Sustainability Entering a More Challenging Phase

Google’s report doesn’t indicate that the company has abandoned its climate targets. On the contrary, it demonstrates significant investments in clean energy, efficiency, demand response, and water projects. But it also makes clear that AI is scaling the problem in new ways.

The narrative a few years ago was simpler: increasingly efficient data centers, large renewable procurement commitments, and a cloud capable of reducing emissions compared to dispersed infrastructure. That story still holds some truth. But AI introduces a new variable: computing demand is growing so rapidly that it forces the construction of more physical infrastructure than clean grids can comfortably absorb.

Google claims that its hardware, software, and energy purchases prevented over 58 million tons of CO₂e in 2025, and that without these measures, its footprint would have been much larger. That’s valuable context, but it doesn’t erase the central question: how much electricity, water, and industrial supply chain will AI need as it shifts from a growing product to an essential infrastructure?

The answer won’t only depend on Google. Microsoft, Amazon, Meta, Oracle, xAI, OpenAI, and major chip providers are in the same race. AI’s sustainability will no longer be measured by a single metric but by the ability to align data centers, electric grids, semiconductor manufacturing, cooling, and actual product consumption.

Google has laid out some figures. And those numbers say that AI is not just software—it’s energy, water, land, materials, and the electric grid.

FAQs

How much did Google’s electricity consumption increase in 2025?
Google reported a 37% increase, the largest annual rise so far.

Does Google only use renewable energy?
Google states it matched 100% of its annual electricity consumption with renewable energy purchases for the ninth year in a row. This does not necessarily mean it uses clean energy every hour in every data center.

Why is the consumption rising so much?
The main factor is the expansion of AI infrastructure, with more data centers, servers, accelerators, and machine learning loads.

What about water?
Google’s total water use grew to 10.9 billion gallons in 2025. It claims to have replenished about 7.7 billion gallons through water management projects.

Does efficiency offset the growth?
It helps but isn’t enough alone. Google reports an average PUE of 1.09 and significant efficiency improvements in Gemini, but total consumption continues to grow due to scaled deployment.