Google's 2025 environmental report, published last week, delivered numbers that would once have seemed unthinkable for a company that has spent years trumpeting its green credentials. Electricity consumption rose 37% year-over-year — the company's largest single-year jump ever — while water use climbed 34% to 10.9 billion gallons and greenhouse gas emissions rose 18%. The culprit, stated plainly in the report itself, is the AI infrastructure buildout.
That 37% increase in load translates to roughly 42 million megawatt-hours consumed by Google's data centers across 2025 — a number comparable to the entire annual electricity consumption of New Zealand. Since 2019, Google's total energy demand has climbed 250%. The company signed a record 12 gigawatts of new clean energy agreements last year, but demand is running ahead of the grid's ability to decarbonize. Google's own language is unusually candid: "our AI infrastructure buildout is currently accelerating faster than the grid is decarbonizing."
The water problem is harder to see
Electricity figures tend to dominate these conversations, but the water numbers are just as stark. Cooling high-density GPU clusters requires enormous volumes of water, and that consumption has more than doubled since 2021. At 10.9 billion gallons, Google's operational water draw rivals mid-sized municipal water systems. Unlike carbon emissions — where renewable energy certificates offer at least a partial accounting hedge — there is no "renewable water" certificate. The water is consumed.
Emissions rose 18%, the largest annual increase Google has reported. A significant portion traces not to data center electricity but to manufacturing AI hardware — chips, servers, networking equipment — where renewable energy certificates can't offset upstream supply chain carbon. That embedded carbon is notoriously difficult to account for and frequently left out of headline figures.
Why this matters beyond Google
Google is the only major hyperscaler currently required to publish detailed sustainability data with this level of granularity, which makes its report a useful proxy for the sector. Microsoft, Amazon, and Meta are all on similarly steep AI infrastructure growth curves. The IEA has estimated that global data center electricity consumption could double by 2026; Google's numbers suggest that trajectory may be conservative if major AI workloads continue compounding at current rates.
None of this means AI development should stop, or even slow — but the environmental cost has shifted from hypothetical to quantified. For years the industry pushed the narrative that AI and sustainability were compatible, even synergistic: smarter grids, optimized logistics, climate modeling. That story isn't false, but the 2025 numbers reveal it was incomplete. The energy drawn to train and serve models is currently growing faster than the clean energy capacity added to power it.
What Google is doing about it
The company points to its 12 GW clean energy signing, advanced geothermal investments, and ongoing work on more efficient chips as mitigation paths. It is also developing new nuclear power purchase agreements. These are real commitments, not greenwashing, but the report's own math shows they haven't kept pace with the demand surge. Google has set a goal to run on 24/7 carbon-free energy by 2030; the gap between that target and the 2025 trajectory is now public and quantified.
For the AI industry as a whole, Google's transparency is arguably a service — it puts numbers on a problem the sector has collectively preferred to discuss in vague terms. Whether competitors will follow with equivalent disclosure remains to be seen.