Ecological impact of AI
A ChatGPT prompt can consume far more energy than a Google search. The invisible cost of an apparent convenience.
Generative AI presents itself as immaterial. It runs in "the cloud". The physical reality is entirely otherwise.
The training of a single large language model can emit as much CO2 as five cars over their whole lifetime (Strubell et al., 2019). Data centres already represent 1 to 2% of global electricity consumption, a figure expected to rise sharply by 2030 according to the IEA. Microsoft acknowledged in 2023 that its water consumption had increased by 34% in a single year, largely because of the expansion of its AI capacities.
"AI is not immaterial. Its ecological footprint is significant."
In the same spirit: Planned obsolescence · Digital sovereignty
↑ Contents
Planned obsolescence
The device that still works has become unusable. No one stole its battery. But someone decided it.
In 2017, Apple acknowledged that it was deliberately slowing older iPhones through software updates. The company later paid fines amounting to several hundred million euros in France and Italy.
Planned obsolescence designates the strategy of conceiving products so that they become obsolete within a controlled delay: physically, through non-replaceable batteries; functionally, through updates that degrade older models; or through marketing that makes what is still usable appear outdated.
In France, a smartphone is kept on average for two to three years, whereas its technical life expectancy is seven to ten years. Electronic waste is the fastest-growing waste category in the world: 62 million tonnes in 2022.
"Digital waste has an ecological cost."
In the same spirit: Ecological impact of AI · Precarity 2.0
↑ Contents