Large language models like ChatGPT have entered the daily lives of millions of people. Every request they process passes through data centers packed with power-hungry GPUs, raising legitimate concerns about their climate impact. As a result, it’s not uncommon for people to criticize those working with AI by pointing to its environmental footprint. That footprint certainly exists — but it’s also fair to ask whether critics avoid higher-impact actions themselves, or whether there’s a double standard when it comes to artificial intelligence.

To place AI’s ecological impact in a global context, we need a comparative analysis with everyday activities whose weight we often forget — to avoid deluding ourselves that skipping a few prompts to ChatGPT will somehow fix the climate crisis.

All figures in the table below are operational estimates (usage phase only — no hardware production included, for either AI or other tools) and should be read as orders of magnitude. The goal is to encourage verification and the search for ever more accurate data — which, unfortunately, AI companies often obscure (a bad sign). Training is excluded here, since its considerable energy cost gets diluted across hundreds of millions of users and becomes negligible compared to everyday usage. Training GPT-4o, for example, consumed tens of GWh — as much energy as an entire neighborhood uses in a year. But after just a few billion queries — which ChatGPT processes in a matter of months — training energy falls below 1% of the total. For the most popular models, the real impact lies in daily use, not training. Conversely, “one-shot” research projects or niche proprietary models may have a much higher training-to-usage ratio.

* EU average electricity emission factor 2023 = 242 g CO₂/kWh
** Prompt eq = CO₂ of activity ÷ 0.073 g (CO₂ footprint of one prompt)
*** Source: Epoch. Sources: MIT Technology Review (May 20, 2025) and University of Michigan — prompt energy use ranges from 0.00003–0.002 kWh depending on model, hardware, and complexity.

Of course, electricity doesn’t always equate to CO₂: if a data center runs entirely on renewables, the impact of a prompt drops to nearly zero, whereas reliance on fossil fuels can dramatically raise it. The numbers also show how quickly AI is becoming more efficient: in just two years, energy consumption per prompt has dropped by an order of magnitude thanks to better chips, model quantization, and batching.

The message I want to convey isn’t to absolve AI’s ecological impact — which remains a systemic issue, as this excellent MIT article notes — but to contextualize its role within broader priorities. If we truly want to reduce our climate footprint, it’s infinitely more effective to skip a dryer cycle, shorten a shower, or give up cars and meat than to stop asking questions to a chatbot. If a non-vegan person thinks they’re helping the climate by avoiding AI, for example, they’re clearly a bit confused.

AI’s climate impact is still real — due to opaque data from companies, the potential for massive future use, and water consumption (which I haven’t addressed here, though that too should be compared to other activities, like a simple toilet flush). But blaming individual users — who may need cars or AI for work — is always ethically wrong, as it shifts attention away from those most responsible for the climate crisis: multinational corporations, governments, and the socioeconomic systems in which they operate. These should be the true focus of climate justice.