The Ecological Footprint of AI is No Small Matter

Global warming was displaced by a new trend: AI - and not because of ecological concerns.

Jul 21, 2025

Si quieres leer o compartir este Artículo en Español.

There has been a steady progression in how fast humans adapt to new technology. For example, Facebook took 4.5 years to reach 100 million users, Instagram 2.5 years, TikTok 9 months and ChatGPT only 2 months. There is also a correlation between the amount of energy and operating costs these technologies require for their advancement. ChatGPT is the highest generator CO₂e per user of all the technological applications out there – and people are beginning to use it as a search engine (at the end of the article, I will share some comparison charts for your consideration).

Apps like Facebook, which have 3 billion users and a history of 21 years, are consuming roughly the same CO₂ per user as ChatGPT, but the total scale is vastly different. True, for ChatGPT the estimation is that it will take 5 years to reach 1.6 billion users whereas TikTok took 7 years to reach it. Also, ChatGPT’s training cost of GPT-4o, at around $200 million, is billions of times greater than the cost of just one user asking their first prompt. That’s the nature of model innovation: colossal upfront investment to make micro-cost usage possible.

And to top it all off, there are at least 15 maybe 20 foundation model families, that is AI companies developing and consuming energy and water at scale.

The cost of a Prompt

The average cost of a single prompt is ~0.2 Wh (conservative average for GPT-4).
The average user makes 10 prompts per day.

These city-scale comparisons are based on GPT-4o being used by 200 million users daily, which is its current estimated global reach. If scaled to a billion, these impacts would multiply fivefold.

Sources: Statista, OpenAI (ChatGPT milestone confirmed in Dec 2022 press brief)

And with M³ of water:

Sources: Boston Water & Sewer Commission, Tokyo Waterworks Bureau

So, you get the idea: one “thank you” from every user to their ChatGPT will cost today a million dollars a day, 40 MWh, 100,000 M³ of water, and 16 tons of CO₂ per day. And this is nothing if we start using the AI models as a web search engine – or when we factor in that there are around 20 AI companies already attracting massive user bases.

Stop with the numbers and let’s address the meaning

Some two years ago, most of the world was concerned about climate change and the ecological footprint of the leading and most industrialized countries. Everyone was demanding cuts on energy, the closure of coal-base power generation, and increase of wind and solar energy. Today, there is a global shift in the conversation from climate change to AI. But curiously, not with regard to energy consumption, or ecological footprint.

Was our concern for climate ever truly deep — or was it just trend-driven? If we meant it, why are we not alarmed now by the ecological footprint of ChatGPT and its siblings?

There is no snappy comeback possible

This desolate scenario is a hard thing to swallow. But I am less concerned with AI’s ecological footprint than I am with social, political, cultural and individual devastation it is bringing. We are driving asleep at the wheel. This is nothing new – we’ve been doing it since, perhaps, the 19^th century. The difference now is that, in the midst of this illusory grand AI mission, we are giving the vehicle full throttle – while our eyes are dazzled by our dreamlike fantasy.