
Google has released its eleventh annual environmental report, focusing on the fiscal year 2025. The company saw a staggering 37% increase in electricity consumption year-over-year, marking its largest historical surge, largely driven by its expanding artificial intelligence infrastructure. The report acknowledges that this rapid deployment is outpacing the decarbonization of electrical grids.
This release comes at a time when the energy appetite of artificial intelligence is putting all major cloud providers under scrutiny. Google is the first to present full-year 2025 figures, while Microsoft, Amazon, and Meta are still reporting on their 2024 operations. Its report stands out as one of the most comprehensive in the sector. However, comparing these tech giants on equal terms can be misleading, as their scopes and methodologies differ; Google uses an adjusted footprint, while Amazon reports total gross figures. Therefore, this report should be viewed as a snapshot of the trajectory of a prominent player amidst evolving needs.
For Google, this trajectory is primarily reflected in its electricity consumption. The report indicates that electricity use skyrocketed to 43.6 terawatt-hours in 2025, compared to 31.7 terawatt-hours the previous year, a staggering increase of over 250% since 2019. Despite this surge, the company has managed to reduce its operational emissions by 2% and those from purchased electricity (scope 2) by 3%, illustrating a genuine decoupling from the growth of energy demand. However, total carbon emissions rose by 18%, reaching 14.5 million metric tons of CO2 equivalent. This increase is largely attributed to indirect emissions in the supply chain (scope 3), which climbed by 25% and now account for 80% of the overall total. The carbon footprint has shifted upstream, largely from semiconductor manufacturing and data center construction.
Efficiency Hits Its Limits
For more than a decade, Google has relied on the efficiency of its data centers as a primary lever to manage energy costs. This efficiency is measured using Power Usage Effectiveness (PUE), a ratio of total energy consumption at a site to the energy used solely for computing. A PUE of 1.09 means that for every kilowatt-hour used for computing, an additional 9% is consumed for cooling and electrical distribution. The company has consistently improved this ratio until it approached its theoretical minimum. The industry average remains at 1.56, placing Google among the most efficient in the market.
However, in 2025, this ratio has stagnated at 1.09, the same as in 2024. This stagnation is not accidental; a PUE cannot drop below 1.0, the benchmark for a site using no extra energy outside of computation. Google has thus reached the physical limits of this lever, exhausting easy gains at the building level. Future efficiency improvements are now sought higher up the technological stack, from chips to models, utilizing in-house designed TPU processors aimed at maximizing computational output per watt.
While site efficiency was improving, it absorbed some of the growth in computing demand. As these efficiencies have plateaued, each new unit of computation has increasingly translated into additional electricity consumption. The energy cost of AI can no longer be offset by the existing infrastructure, which will inevitably influence cloud pricing and impact the token economics of large models that client companies will eventually bear.
Rising Water Costs
To cool its servers cost-effectively, Google has opted for water cooling instead of air cooling. In 2025, the company consumed 41 million cubic meters of water, a 34% increase year-over-year. This volume has more than doubled since 2021, mirroring the trajectory of computing demand.
The company highlights its replenishment strategy, with 78% of its freshwater consumption compensated through local projects, aiming for 120% by 2030. However, this offset is currently failing to keep pace with rising consumption. Water is becoming a tangible and local concern, as a data center draws from resources shared with the surrounding community.
In Europe, data center approvals increasingly hinge on their water footprint. The availability of computing resources will soon be negotiated on these terms.
Renewable Energy Commitment Counts on Annually
Google reiterates its achievement of covering 100% of its electricity consumption with renewable energy purchases for the ninth consecutive year. This matching occurs on an annual global scale, whereas actual consumption plays out hour by hour and grid by grid. The discrepancy is significant; the indicator that measures this reality—carbon-free energy 24/7 (CFE)—flatlined at 66% in 2025, unchanged from 2024.
In reality, nearly one third of operational hours consume electricity that remains carbon-intensive, with offsets managed through certificates acquired from elsewhere. Therefore, the “100% renewable” claims emphasized by providers and mirrored in their clients’ reports come with an asterisk. A decision-maker constructing their CSRD report based on their hosting provider’s numbers should be careful to distinguish annual matching from real-time carbon intensity.
Google is aware of this gap, and its strategy is reshaping the energy landscape. The company signed new clean capacity agreements totaling more than 12 gigawatts in 2025, a figure eight times its 2019 contracts. Google is now banking on controllable energy sources, from nuclear to geothermal and even fusion, given the insufficient solar and wind energy matching on an hourly basis. This shift toward reliable electricity will determine where computing infrastructure can be deployed.
These three ceilings stem from the same pressure—a demand for growth that exceeds the levers designed to curb Google’s footprint. At both Microsoft and Meta, indirect emissions constitute over 95% of their totals. The carbon footprint of AI is shifting from consumption to manufacturing, landing in the scope 3 categories for both suppliers and their clients.
For IT decision-makers, Google’s report provides a framework to understand the glass ceiling that today’s initiatives face. Efficiency appears to be reaching its limits while demand continues to surge, escalating costs in which water and energy will play an increasingly significant role in determining the pricing and location of computing resources. It also casts a critical eye on claims of renewables, suggesting they merit close examination. The challenge remains to integrate these factors into cloud service decisions, alongside performance and sovereignty considerations.



