About climate change and AI
Recently, a researcher from a huge company published a featured article in Nature, which defends not only that AI is not detrimental to the climate, but instead that promoting AI has exponentially beneficial effects for the climate [1]. I am not happy with what I read. Below, I’ll briefly expose the points that I find disturbing. Of course, if you have not read the article, some of my points might feel out of context, but I think that it is very likely that you have heard similar comments somewhere else.
It is just 0.5% more CO2
An argument that we find in the first lines of the article is the following:
But, from a global perspective, AI’s operational footprint remains modest. Data centres use about 1.5% of global electricity and contribute roughly 0.5% of total CO2 emissions.
You might have heard the argument in other forms, such as “people make more emissions than driving their SUV to work than what they generate using AI”. True. The problem, obviously, is that we are not reducing emissions by driving fewer SUVs. We are not reducing emissions, full stop. Does the fact that you are not the largest emission producer give you an alibi to start polluting? I do not think so.
What do we call AI?
For example, AI-enabled dynamic line ratings (a technology used to monitor the real-time capacity of overhead power transmission lines) use real-time data — such as temperature, wind speed, and other weather conditions — to continually adjust the safe operating capacity of power lines.
Let me ask you a rhetoric question: What is the meaning of AI at this point? For instance, the author highlights an application that seems quite interesting (monitoring). Now: is that employing an LLM? How does an LLM enable better monitoring? It does not matter, because the description of the project the author provided does not even mention LLMs or AI (to my knowledge). It is simply a monitoring initiative, and though I am not an expert on gas pipeline monitoring or anything alike, I would say that it is possible that this technology was ready much before we spent $300 billion on data centers. It is very alarming that this is the best example that the author can come up with to highlight AI’s potential. More importantly, it seems that all planned improvements connected with AI come from efficiency gains, which is not something that has allowed to decarbonize the economy yet.
The AI-kool aid
Yet, only 39% of people globally who use AI at work report receiving employer-led AI training, even though 66% of business leaders now consider AI skills essential […] Without targeted support, a growing AI skills gap could limit progress in climate-crucial sectors such as energy, agriculture and public policy. […] Building this fluency is essential to unlocking AI’s full climate potential.
Oddly enough, it seems that diverting blame from their big companies for climate change is not enough: instead, the author tries to blame workers! This is quite similar to other stories about the future of AI that I have read recently. The narrative goes like:
- The future will be 10000% better in ways that we cannot predict or tell you.
- But you have to trust us.
- You either board now into the future or you will be left behind forever.
Not very different from a cult. The rapture is about to happen, but those who refuse to drink the Kool-Aid will not depart.
Beware of economic models
Researchers from the London School of Economics and Political Science estimated even greater potential: 3.2–5.4 gigatonnes of annual reductions by 2035 across electricity, transport, and food sectors. […] While this analysis comprehensively uses a bottom-up method to estimate the impact of AI on emissions reductions, it has its limitations […] It does not include the impact of AI on accelerating capital deployment or supporting better policymaking, which are key levers to speeding up the climate transition. […] Our analysis also does not consider any rebound effects such as efficiency gains in AI leading to increased consumption or unintended expansion of carbon-intensive processes
In other words, the limitations of the study were simply that they could not put an upper bound on AI efficiency gains. I guess I do not need to convince anybody about how deranged this is. As a computational biologist expert in molecular modeling, I would recommend being wary of models in general. Some of their results are not due to the consequences of their model but due to the assumptions of the model. In this case, the assumption is that AI solves everything, so why would the result surprise us.
The cost of opportunity
There might be a much bigger issue that few people pay attention to: the cost of opportunity. LLMs run in huge data centers employing hardware that becomes obsolete very fast. As a result, many firms are incurring huge investments that so far are not getting monetized.
Much attention has focused on AI’s high energy consumption and water use […] But they are not major drivers of global climate change. Framing them as such distracts focus from AI’s transformative climate-mitigation potential and from its biggest influence on the climate, which will depend on how it is used.
The issue is that the ecological transition to a decarbonized economy requires money too. A lot of money. Invest all the money in building data centers, and there will not be enough money to make other things that are much more pressing from a climate perspective (e.g., a modern power infrastructure, high-speed trains, solar power centrals). Keep diverting money from where it is really needed, and we will have amazing deep fake porn generators, but we will be living under the sea level.
The tobacco model, into tech
For a long time, the tobacco industry hired statisticians and doctors to avoid the causal link between tobacco and lung cancer. Oil companies tried to make us believe that climate change was due to volcanic activity. Now, tech companies seem to be very worried about climate change, at the same time that they become important polluters. Is there a pattern here?
References
[1] https://www.nature.com/articles/d41586-025-02641-4 [3] https://www.nature.com/articles/s44168-025-00252-3