Commentary on the environmental impact of AI often swings wildly between doom-and-gloom catastrophism and blind techno-optimism. But where’s the truth in all this? On July 24, 2025—the symbolic date of Earth Overshoot Day—we sat down with Yves Grandmontagne, founder and editor-in-chief of DCMAG (Data Centre Magazine*), to get his take on AI and its real environmental impact. It is worthy of note that Yves and I explored Silicon Valley’s infrastructure innovators together through extensive press tours some time ago. This provided us with firsthand insight into the tech industry’s approach to these challenges. The current annotated transcript of our interview is a summary of our thorough, nuanced, and let’s admit it, quite lengthy discussion. You are therefore encouraged to treat this article as your starting point for diving deeper into this extremely complex topic.

Exploring the Real Environmental Impact of AI

* DCMag is only available in French

This post summarises what turned out to be an incredibly rich hour-long conversation. The sheer complexity of this topic forced us to dig into multiple technical, economic, and environmental angles—making any kind of comprehensive analysis near inconceivable.

Drawing on his deep expertise in the data centre and AI sectors, Yves Grandmontagne gives us some much-needed factual perspective on a debate that’s often polarised between doomsday scenarios and over-the-top techno-optimism. To tackle this properly, we decided to take recent quotes—both positive and negative—and fact-check them with our expert.

Yves’s analysis helps us cut through the noise and understand what’s really at stake in this technological breakthrough.

---

Environmental Impact of AI: Reality Check Time

TLDR: Environmental Impact of AI

The electricity consumption issue is more nuanced than you think* – AI will represent 20-30% of data centre consumption (not twice that number), and only 2-4% of overall electricity consumption

1. Energy efficiency gains are actually remarkable – Over the last decade: number of data centres x2, floor space x4, but energy consumption up only 6%





2. Beware of dubious comparisons – Comparing a ChatGPT query to Google search is methodologically flawed (completely different technologies and services)





3. Water consumption varies massively by geography – Huge issue in the US, but Europe has been using smarter closed-loop systems for ages





4. Tech innovations look promising – New technologies (direct liquid cooling, immersion cooling) are slashing water and energy consumption





5. AI might actually be part of the solution – Can optimise energy mix management and electricity transport, which is currently our main bottleneck





6. Let’s get some perspective here – Data centre impact remains pretty marginal compared to the chemical industry (32% of French energy consumption) or agriculture

*All numbers by Yves Grandmontagne at Data Centre Magazine

Bottom line: The impact is “real but massively overstated”—we need to put things in context and remain cool and collected.

---

So here are the quotes about the environmental impact of AI that we wanted to fact-check with Yves.

Rumours vs Reality: Decoding the Doomsday Predictions

Predictions of Booming Electricity Consumption

The first claim I put to Yves Grandmontagne was Synth Media’s prediction [Fr] that “AI’s growth could double data centre electricity consumption by 2026.” His response immediately throws cold water on this alarmist take:

“It’s absolutely true that AI is rolling out infrastructure at breakneck speed and eating up more space in data centres. Sure, it’s going to significantly bump up their energy consumption—no question about that. But will consumption actually double? I seriously doubt it. AI should account for somewhere between 20 and 30% of global data centre consumption worldwide.”

<figure class="wp-block-image size-large"><figcaption class="wp-element-caption">A data centre server rack — photo antimuseum.com</figcaption></figure>

AI should account for somewhere between 20 and 30% of global data centre consumption worldwide

This reality check reveals something consistent throughout Yves Grandmontagne’s analysis: the absolute need to put numbers in context. The expert points out that this increase is just part of the natural progression tied to our ever-growing digital habits. “What’s driving this consumption increase is our daily usage, whether that’s for work or for personal reasons,” he reminds us, highlighting our collective responsibility in this evolution. It’s a topic we’ve tackled before with a broader focus on digital consumption.

The most striking part of his analysis is about the energy efficiency angle. Contrary to popular belief, data centres aren’t following a consumption curve that mirrors data growth. This improving efficiency is something that gets completely overlooked in public debates about the environmental impact of AI.

ChatGPT’s Carbon Footprint: More Context Needed

When it comes to the 10,113 tonnes of CO2 equivalent attributed to ChatGPT usage in January 2023 (Basta Media – Data for Good, AI has the potential to destroy the planet), Yves Grandmontagne takes a refreshingly pragmatic approach:

“I can’t verify that exact figure. Getting that precise—down to 10,113 tonnes—represents a massive methodological challenge, especially when you’re dealing with AI infrastructures that are distributed systems.”

<figure class="wp-block-image size-large"><figcaption class="wp-element-caption">We asked Yves Grandmontagne, editor-in-chief of DCMAG (Data Centre Magazine) to give us the real story about the environmental impact of AI — He gave us facts and figures, which we’ve compiled in the infographic at the end of this post</figcaption></figure>

This observation raises a crucial methodological point: just how tough it is to accurately measure the carbon footprint of distributed infrastructures. The expert does acknowledge this pollution is real, but puts it in perspective: “Those 10,113 tonnes of CO2 still represent volumes significantly smaller than what many other industries pump out.”

This contextualisation isn’t about downplaying the issue—it’s about keeping things proportional. Yves Grandmontagne reminds us of a basic truth that is often overlooked: “The moment we use our smartphones, we become CO2 producers.” This highlights the inconsistency in criticisms that single out AI from our overall digital consumption.

The Google vs ChatGPT Comparison: A Methodological Trap

<figure class="wp-block-image size-large"><figcaption class="wp-element-caption">Watch out for convenient shortcuts between pollution and digital tech—they’re everywhere and pretty handy when you want to hide overall industrial pollution — image created with Midjourney</figcaption></figure>

MIT’s claim that “<a href="https://www.lecese.fr/sites/default/files/pdf/Avis/2024/2024_14_IA_E