Google’s latest Environmental Report, published June 27, 2025, lays bare a problem the Mountain View giant can no longer put off: the energy demand of artificial intelligence is growing faster than any sustainability plan can keep pace with.
In 2024, the Mountain View company logged a 27% rise in electricity consumption across its data centers. The computational power that advanced AI models need, the ones behind large neural networks, comes at that price. And yet, thanks to massive investment in renewable energy and more efficient infrastructure, Google managed to cut associated carbon emissions by 12%. More than 25 new renewable energy projects came online, adding 2.5 GW of clean capacity, while another 8 GW have already been contracted for the future.
Efficiency improved above all thanks to the TPU v5 “Ironwood” chips, 30 times more energy efficient than Google’s first-generation chips from 2018. Quantization of machine learning models, in turn, sped up the training of large language models by 39%, with fewer computational resources required.
Energy consumption, though, is not the only environmental cost of this race. There is water too. In 2023, Google’s data centers used 6.1 billion gallons of drinking water for cooling, as much as the annual irrigation needs of roughly 40 golf courses. The Council Bluffs facility in Iowa alone used 980 million gallons, the equivalent of 6.5 golf courses. In Northern Virginia, a global hub for data centers, water consumption rose 65% between 2019 and 2023. A trajectory that raises serious questions about sustainability and environmental responsibility.
In response, Google has stepped up water recycling and replenishment projects, offsetting 64% of the freshwater it used in 2024, compared with 18% the year before. But a basic fact remains. AI’s growth carries real environmental costs, and not all of them can be offset.
In parallel, sustainability is also reshaping how Google manages traffic to sites.
Digital Sustainability and Editorial Selection: How Google Is Changing and What Publishers Must Do
For years now, Google has been pursuing a path of environmental sustainability and energy efficiency, and the way it indexes and organizes the web shows it. Googlebot, Google’s spider, has become more selective and more frugal. Requests to re-crawl pages that have stayed unchanged for more than a month have dropped 14% compared with 2021. A white paper presented at the Hot/Storage 2024 conference attributes to this shift an energy saving of roughly 3.2 TWh a year, as much as the electricity consumption of 280,000 European homes.
At the heart of this approach is the Change Rate Estimator, an algorithm that took effect in Q4 2023. It decides whether a URL genuinely deserves a new visit, and defers by up to 90 days any page with less than a 0.2% probability of change, unless a fresh signal arrives from the sitemap or a new inbound link. Load speed also factors into the energy calculation. Martin Splitt showed this with internal data at Chrome Dev Summit 2023. Pages with a Largest Contentful Paint (LCP) under 2.5 seconds get almost twice as many crawler visits, because they use less CPU.
Google has also changed its indexing strategy, moving from an omnivorous approach to an “exclusive catalog.” A study by analyst Vincent Schmalbach, across a sample of 1,200 domains, makes the trend clear: in 2025, only 42% of submitted URLs were indexed, against 90% in 2019. Today Google favors content with high semantic originality, measured through the new MUVERA* system, domain authority, and complete structured data.
The tightening has a name, the “Great De-indexing” of May 27–29, 2025, when Search Console logged 275 million more pages in the Crawled - currently not indexed state. Marie Haynes analyzed the phenomenon: 71% of those pages held low-originality content, paraphrased text, product listings without reviews, old and barely read articles. In the same days, Google published version 9.3 of the Quality Rater Guidelines, with the new “Paraphrased / Low-Originality Content” flag. Hard to call that a coincidence.
Gary Illyes, a senior analyst at Google, put it plainly on an episode of the Search Off The Record podcast dedicated to the sustainability of crawling.
“Crawling is very low-hanging fruit when you’re trying to reduce your environmental impact: fewer useless pages = less CO₂” - Gary Illyes, Search Off The Record, episode “Making crawling more sustainable,” November 11, 2022.
For digital publishers, the game changes. Producing large volumes of generic content is no longer enough. What’s needed is content that is unique, genuinely high quality, with real added value. And the technical performance of sites needs work too: faster load times, Core Web Vitals under control, regular monitoring to anticipate possible de-indexing. Optimizing Core Web Vitals on the server side cuts by 40% the CPU Google spends on rendering, as Abhijit Saha from the Sustainability team has explained.
Artificial intelligence and sustainability are now two sides of the same coin for anyone publishing online. Those who understand this early consolidate their position. Those who ignore it end up out of the index.
Investing in original, technically efficient content that aligns with E-E-A-T guidelines (Experience, Expertise, Authoritativeness, Trustworthiness) keeps you visible online. It also means operating responsibly, in a digital ecosystem that has to reckon with increasingly tight energy constraints.
John Mueller reiterated this during the Office Hours of December 2024.
“Indexing everything is no longer sustainable or useful; we aim for an index with a net positive quality” - John Mueller, Office Hours, December 2024
The result is a visible compression. Only pages that carry new information, hosted on fast, lightweight sites, make it into the index. Uniqueness, performance, and authority signals are now indispensable, no longer just a suggestion of good practice.
The challenges posed by AI and sustainability are not going away. The publishers who stay at the center of the market are the ones that treat the quality and authenticity of their journalism as their one real competitive advantage.
*MUVERA is the algorithm Google uses to simplify search across complex data, such as documents represented by multiple vectors (one per word), without losing precision. It condenses these representations into a single synthetic vector, called an FDE, that preserves the essence of the original content. The result is faster search methods, with much lower time and computational costs and an accuracy close to that of more sophisticated systems.
Frequently asked questions
