The artificial intelligence (AI) revolution is reflected in Google’s electricity meters. In its 2026 environmental report, the group acknowledges that in 2025 it experienced “the sharpest rise in electricity demand” in its history: its total electricity consumption rose by 37% year-on-year, following a 27% increase in 2024. In absolute terms, it rose from 31.7TWh in 2024 to 43.6TWh in 2025. In 2021, it stood at just 18.1TWh. In four years, Google has therefore more than doubled its electricity consumption.
This increase is primarily driven by data centres. In 2025, they consumed 42.4TWh, compared with 30.6TWh a year earlier. Offices and other facilities, by contrast, accounted for just 1.17TWh. In other words, more than 97% of the electricity used by Google is now spent on running its digital infrastructure – the systems that support Search, YouTube, Google Cloud and, increasingly, AI models and services.
The historical perspective is even more telling. Google reports that its total electricity demand has risen by more than 250% since 2019. The company attributes this increase to the growth of Google Cloud, investments in Search, the expansion of YouTube and AI capabilities. This is one of the report’s most significant revelations: AI is presented both as a tool for climate transition and as a factor placing immediate strain on electricity grids.
Clean energy, but…
To defuse the criticism, Google is highlighting its strategy for purchasing clean energy. The company claims that, for the ninth consecutive year, it has offset 100% of its annual global energy consumption through the purchase of renewable electricity. It says that in 2025 it signed agreements for more than 12GW of new clean energy, the largest annual volume in its history. Since 2010, Google claims to have secured more than 240 agreements for nearly 35GW of new capacity.
But the report also highlights the limitations of this approach. Annual “matching” with renewables does not mean that every data centre operates at all times using locally available carbon-free electricity. Google acknowledges this indirectly by explaining that it wants to move from an annual approach to an hourly one, with carbon-free electricity “every hour, on every grid”. The group also highlights the bottlenecks: connection delays, fragmented grids, slow regulatory processes and a lack of controllable carbon-free generation.
The challenge is therefore clear: Google is improving the efficiency of its machines, but its energy requirements are growing even faster. The company is targeting an average PUE of 1.09 for its data centres by 2025, energy efficiency three times higher per unit of computation compared with five years earlier, as well as a 33-fold reduction in the energy required for a median text query in Gemini Apps over a 12-month period. These gains are significant, but they are not enough to stem the surge in total demand.
Emissions related to the supply chain
The group is attempting to demonstrate that this growth can be decoupled from its operational emissions. In 2025, Google claims to have reduced its operational emissions by 2% year-on-year, despite an increase in its electricity consumption. It also highlights that it has reduced its emissions per gigawatt-hour by 47% since 2019. However, the picture is less favourable across the entire value chain: emissions linked to the supply chain have risen by 25%, largely as a result of the construction of new AI infrastructure and an Asian supply chain that is still powered by energy networks that have not yet been sufficiently decarbonised.
The report’s key message therefore lies less in Google’s green PR than in the sheer scale of its energy requirements. The company continues to secure a growing number of contracts in solar, wind, hydroelectric, nuclear, geothermal and even fusion power. But its own trajectory shows that the next battle in AI will not be fought solely over chips, models or applications. It will also be fought over electricity grids, their capacity to absorb rapidly growing demand, and the speed at which new low-carbon capacity can actually be connected to the grid.



