Saying ‘please’ and ‘thank you’ to ChatGPT wastes millions of dollars a year, according to OpenAI chief executive Sam Altman, and his admission raises serious questions about the sustainability of popular chatbots.
Altman revealed the cost in response to a post pondering the electricity consumption of polite users, noting that it’s “tens of millions of dollars well spent”.
While Altman’s comment suggests this is a ballpark estimate, it highlights the huge impact of the generative AI boom from a sustainability perspective. Over the last two years, we’ve had repeated warnings about the environmental issues created by the technology, and yet consumers appear to be burying their heads in the sand while developers reap the rewards of our curiosity.
User queries on ChatGPT, or any AI chatbot for that matter, come at a cost – one that many users might not realize. These powerful models rely on data center infrastructure, which requires vast amounts of power and water to provide you with those lightning quick responses.
The figures on chatbot energy consumption are startling. Statistics from Goldman Sachs, for example, show that each GPT-4 query uses around ten-times more electricity than a standard Google query.
Similarly, analysis by the Washington Post last year revealed that if one-in-ten Americans used GPT-4 just once on a weekly basis, the power required to respond to that solitary query is astounding.
It equates to the same amount of electricity consumed by every single household in Washington DC over a 20 day period. The US capital has more than 670,000 inhabitants – do the math (and try not to lose your mind thinking about it).
What’s most concerning about Altman’s admission is that this centers around these simple terms. Imagine the impact of the rest of the mundane queries ChatGPT is bombarded with each day.
Users are turning to ChatGPT as an alternative to simple Google searches, using the chatbot to self-diagnose, ask basic questions that could’ve been solved by a few simple clicks on Google, and using the tool for a whole multitude of tasks on a daily basis.
Jumping headlong into using these tools so frequently has a range of implications, not least of all the potential for dumbing us down, but we’re also at risk of digging our own graves from an ecological perspective.
And that’s just ChatGPT. Have a browse through X today and you’ll see user upon user asking Grok, the social media platform’s in-built chatbot, to summarize threads or clarify information contained in posts.
While energy consumption has an obvious environmental impact – take a look at what’s powering the data center used to support Grok 3 in Memphis, per Data Center Dynamics – concerns over water consumption can’t be ignored.
Microsoft data centers used to train GPT-3 consumed 700,000 litres of water, for example, and now we’re multiple iterations on from this initial model.
Research from the University of California, Riverside shows that water consumption rates are expected to surge in the coming years. Global AI demand is projected to account for between 4.2 – 6.6 billion cubic meters of water withdrawal by 2027.
That’s more than the total annual water withdrawal of half of the United Kingdom.
Moreover, fresh water usage is occurring in areas with ‘high water scarcity’. In 2023, Google revealed that 15% of all its freshwater consumption came from high-risk areas, while 42% of Microsoft’s freshwater withdrawals in the same year came from ‘areas with water stress’.
To their credit, big tech companies are making efforts to improve water sustainability.
In response to an ITPro query on this topic in 2024, Microsoft said it has set goals to be ‘water positive’ by 2030. The tech giant is also ramping up efforts to conserve water and exploring options for more efficient data center cooling techniques.
The fact still remains though that big tech is sucking the earth dry, so think twice next time you ask Grok for cliffnotes on a Twitter thread.
