Fact-Checking the "AI Environmental Crisis"

In recent months, a wave of alarming headlines is sweeping through our news feeds: “AI is Thirstier Than We Thought,” “AI is Eating the World’s Power,” and “ChatGPT’s Water Footprint is Sustainable Only in Your Nightmares.”  While an honest discussion about the use of environmental resources is never a bad thing, we should avoid moral panics fueled by out of context data points. 

Yes, training and running Large Language Models (LLMs) requires significant resources. In particular energy and water for cooling.  The rise in AI infrastructure investment has been lightning fast and is bound to create tensions. However, if we look at the actual usage data in a wider context, as well as the motivations behind the loudest critics, we see a very different story emerge.

The narrative that AI is an environmental doomsday machine is not just a huge overstatement – it is often being weaponized by specific interest groups to feed on societal angst and slow down innovation.

A Drop in the Bucket

To understand AI’s real impact, we have to move past scary-sounding cherry-picked numbers (usually using billions of gallons or TWhs)  and look at global usage percentages. Currently, data centers (of which AI is only a small part) account for about 1% to 1.5% of global electricity use. They are used to fuel a huge number of applications, from telecommunications, e-commerce to online streaming. 

By comparison, the energy used by traditional cooling and heating (HVAC) systems in buildings or the global transportation sector dwarfs AI by orders of magnitude! As Andrew Ng recently noted in The Batch, while AI’s energy growth is real, it is often compared to the total energy of small nations only to create a sense of shock in the reader. But what does that comparison really mean? Somehow we don’t see the same alarmist headlines for the global laundry industry or video streaming services, both of which have massive energy footprints. Not to mention traditional resource gobblers like agriculture and heavy industry. Why? Because AI is the “new” scary tech and it’s easier to intimidate people with something they don’t know or understand.

Global catastrophe or local shortage?

One of the most persistent misconceptions is that AI is “draining the world’s water.” When you look at the numbers it becomes clear however, that while sometimes a real concern, the water usage in AI is a local issue, not a global one.

When a data center is built in a region with abundant water and a closed-loop cooling system, its impact on the global water cycle is negligible. Any “crisis” only occurs when such infrastructure is poorly placed in already water-stressed areas. This is a challenge of urban planning and corporate responsibility, not a fundamental flaw of Artificial Intelligence as such. Framing a local utility management issue as a global environmental catastrophe is a classic case of scapegoating and creating a moral panic.

To put numbers behind this – agriculture alone uses around 70% of the world’s sweet water, with industry eating another 20% and individual consumption accounting for the rest. To put things into perspective golf courses in the US consume more water than the AI industry. Eating a dinner meal of beef with your family will use more water than you need for 30 000 (thirty thousand!) interactions with ChatGPT – more than you will may have in your life. Are beef eaters killing the planet?

Every statistic, taken out of context, can be made to sound scary – that is the tactic of choice in these campaigns. What we really need is to see everything in the proper perspective and weigh the costs against benefits. 

The “Hidden Interests” Behind the Alarmism

We must ask: why are we seeing this wave of similarly framed and seemingly concerted narratives? 

As Andrew Ng argues, we are seeing the rise of “professional campaigning organizations,” often funded by unclear interests, that use environmental concerns as a convenient hook to tap into societal fears and further their agenda. By convincing the public that AI is an environmental hazard, these groups can more effectively advocate for stricter regulation of AI and hope to divert investment away from it, as a potentially risky endeavour. These tactics often go by the acronym FUD – as they aim to cover the target in a web of Fear, Uncertainty and Doubt. 

Oddly this has been previously directed at open source by corporate providers of IT solutions and generally is often what legacy industries rely on to slow down competition from emerging technologies. While some caution is always prudent, there is a clear downside for society. If only the largest, entrenched  corporations can afford the “environmental compliance”, the upstart competition (including open source projects) faces an even more tilted playing table.

Blowing the ecological footprint issue out of proportion doesn’t save the planet, but it may serve some incumbent players with deep pockets. But why has this narrative become so pervasive lately? It turns out that the campaigners behind them ran studies that helped them identify which anti-AI arguments are most effective at mobilizing public support. The early favourite, the doomsday scenario, where AI would lead the extinction of humanity, turned out to be too far-fetched for people to care about. So the campaigners turned to things that hit close to home – job security and the environment. The focus research gave them a clear answer – people can be galvanized to act to save the environment and jobs, so that is what they have latched onto. 

AI can be a solution, not just a cost

Finally, the “cost” side of the equation is always highlighted, while the “savings” side is ignored. AI is currently being used to:

• Optimize energy grids, reducing waste by percentages that far exceed the power AI itself consumes.
• Accelerate material science, finding new catalysts for carbon capture and more efficient battery chemistries.
• Streamline logistics, cutting thousands of tons of CO2 from global shipping routes.

To focus only on the electricity used by the server while ignoring the massive efficiency gains AI provides to the rest of the physical world is intellectually dishonest.

The truth is that large-scale data centers are very efficient in what they do. Running the same tasks on individual computers would be far more taxing on the environment at every stage of the process. These are simple economies of scale. 

The Path Forward

The AI industry is committed to reaching net-zero. Data center operators are the largest corporate buyers of renewable energy in history. Leading hyperscalers like Google and Microsoft are pioneering small modular reactors (SMRs) and advanced liquid cooling.

Every industry must be ready to discuss its environmental and societal impact, and AI is no exception. But we must move the conversation away from sensationalism and into solid evidence and facts. The “water and energy crisis” of AI is a manageable engineering challenge, not a reason to halt progress lest we destroy civilization. Let’s focus on building better infrastructure rather than falling for narratives expertly researched and designed to feed on our fears to further sometimes less-than noble interests.