We asked ChatGPT about its carbon footprint – A greener life, a greener world

By Anders Lorenzen

This new monthly series focused on asking a Large Language Model (LLM) such as ChatGPT a relevant climate, energy or sustainability-potent question.

This months question

With the growing debate about ChatGPT‘s and other LLMs’ carbon footprints, the question that would kickstart this series was rather obvious.

With the controversies around biases, Deepseek’s sensitivity, and criticism of the Chinese government, this question would test how well ChatGPT scrutinises itself.

Anders Lorenzen (AL): What is your carbon footprint, ChatGPT?

ChatGPT response:

Interestingly, ChatGPT did not provide a clear and direct answer but instead mentioned what contributes to its carbon footprint, such as data centres, training and inferences and hardware and infrastructure.

I went back to the drawing board, and I realised I needed to do more targeted questioning.

AL: What was the ChatGPT US carbon footprint in 2024?

This prompt produced a much better response from ChatGPT.

ChatGPT response:

It provided me with data showing that in 2024, ChatGPT accounted for 4% of electricity usage and represented roughly 2.18% of the US’s total emissions. 

It further informed me that the LLM produced 260.93 metric tons (mt) of CO2 monthly, equivalent to the emissions of 260 transatlantic flights.

Across 2024, ChatGPT data centre emissions accounted for 105 million tons of CO₂e.  

Per query, ChatGPT used 2.9 watt-hours (WH) which is roughly 1.07 grams of CO₂.

AL: What are the other concerns aside from energy usage?

While its data center emissions are the biggest issue, it would be unfair to ignore other environmental implications of data centres.

ChatGPT response:

Beyond energy consumption, data centers also impact local resources:​

• Water usage: Cooling systems in data centers can consume significant amounts of water. Some facilities use up to 4 million gallons daily. ​
• Local infrastructure strain: In regions like Ireland, data centers have been reported to consume more electricity than all urban homes, raising concerns about energy stability and environmental goals. ​

In conclusion, while the proximity of data centers can reduce certain aspects of carbon emissions, the broader environmental impact, including energy sources and resource consumption, must be considered.

AL: How can we make AI and ChatGPT greener?

If this made you disheartened, ChatGPT was ready to offer ideas for how it could become greener:

ChatGPT response:

• Optimise model usage: Employ smaller, task-specific models where appropriate to reduce energy consumption.
• Leverage renewable energy: Prioritise training and deploying models in regions or data centers powered by renewable energy sources​
• Enhance hardware efficiency: Invest in energy-efficient hardware and cooling systems to minimise energy waste
• Monitor and report emissions: Utilise tools like CodeCarbon to track and report the carbon footprint of AI models, promoting transparency and accountability

AL: So where does all this leave us?

ChatGPT response:

While offering advanced capabilities, its substantial energy and water consumption raise concerns about sustainability, especially given the increasing demand for AI services.

Calculate your LLM footprint

Have you tried calculating the carbon footprint of your ChatGPT usage or any other LLM you use?

If yes, we would love you to share the result with us.

Next month’s AI question

What question should we ask ChatGPT next month?

Fill in the form below, and we may feature your idea.