Summary
- Google’s 2025 environmental reporting shows higher resource demand linked to AI infrastructure growth.
- Electricity, water consumption, and construction-related supply-chain emissions remain central pressure points.
- The figures show how hyperscale sustainability targets are being tested by the physical scale of AI build-out.
Google has reported higher energy and resource demand as AI infrastructure growth expands the physical footprint behind its services, putting electricity, water, and construction emissions under renewed scrutiny.
The company’s data centre sustainability materials link its environmental performance to clean-energy procurement, water stewardship, energy efficiency, circularity, and construction emissions. Google’s 2025 Environmental Report sets out the detailed reporting behind those themes.
Trade coverage of the report highlighted a 37% rise in Google’s electricity consumption in 2025, alongside water consumption of 10.9bn gallons and around 2.3m tonnes of CO2e associated with data centre construction in the supply chain. The figures show how AI demand can stretch even the most advanced corporate sustainability programmes.
Efficiency is fighting scale
Google continues to operate some of the industry’s most efficient facilities and remains a major buyer of renewable and carbon-free energy. It also reports that its data centres deliver more than six times more computing power per unit of electricity than five years ago. Those gains are significant, but absolute growth can still outrun efficiency.
AI is accelerating that tension. A facility can produce more compute per watt and still consume more total electricity if workloads grow faster than efficiency improves. The same pattern applies to cooling. Water-based cooling can reduce energy consumption in some climates, but it can increase local water exposure where evaporative systems are used.
Google describes water cooling as an energy-efficient way to remove heat and says it uses a climate-conscious approach when choosing cooling systems at each campus. That is a real engineering trade-off rather than a simple failure. In some markets, water can reduce energy use and carbon impact. In others, water stress may make the same approach harder to defend.
Construction emissions add another layer. Data centre sustainability debates often focus on operational electricity, but rapid AI capacity growth also increases embodied carbon through steel, concrete, mechanical plant, electrical equipment, backup systems, and site works. As campuses become larger and denser, supply-chain emissions become harder to keep in the background.
Europe will demand local evidence
Europe’s regulatory environment will make the resource picture more visible. Significant data centres are already subject to energy performance reporting under the Energy Efficiency Directive framework, and further rating or minimum performance measures are moving through policy discussions.
Corporate sustainability reports and local infrastructure impacts do not always line up neatly. A hyperscaler may match electricity consumption with renewable energy globally or regionally, while a specific facility still connects to a local grid node, uses local water or alternative cooling, and triggers local planning, reinforcement, and heat-reuse questions.
That gap is becoming a planning risk. Municipalities and regulators increasingly want to understand where the power comes from, whether the grid can absorb the load, what water source is used, whether waste heat can be recovered, and how construction impacts are being reduced.
Google’s scale gives it tools that smaller operators do not have: long-term energy contracts, utility partnerships, custom cooling designs, advanced efficiency programmes, and supplier influence. Rising electricity and water figures under that operating model show how powerful the AI demand curve has become.
The reporting also creates a communications challenge. Sustainability language built around net zero, carbon-free energy, and responsible water use becomes harder to sustain if absolute resource use rises quickly. The stronger position is detailed disclosure around trade-offs, local decisions, and infrastructure constraints.
Google’s figures do not show a company ignoring sustainability. They show a company encountering the physical consequences of the AI build-out. The environmental performance of cloud infrastructure is now being determined by substations, cooling choices, water basins, equipment supply chains, and construction materials at a scale efficiency gains cannot fully offset.
