ChatGPT is so convincingly human-like in its responses that it almost feels rude not to say “please” and “thank you” to the AI chatbot, especially when it’s helping with a tricky email or prepping you for a job interview.
But all that extra politeness is quietly costing OpenAI millions.
Earlier this week, an X (formerly Twitter) user named @tomieinlove mused: “I wonder how much money OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you’ to their models.”
OpenAI boss Sam Altman replied, presumably with tongue firmly in cheek: “Tens of millions of dollars well spent — you never know.”
The account that asked the question leans heavily into comedy, and Altman’s response was likely tongue-in-cheek. However, the exchange underscored a serious point: running AI at scale is expensive, and there are real questions about how those costs are being passed on to consumers.
How much does ChatGPT cost to run?
tens of millions of dollars well spent--you never know
— Sam Altman (@sama) April 16, 2025
Behind the friendly chatbot lies a mountain of computing power. Every time you ask ChatGPT to polish a cover letter, brainstorm a lesson plan or revamp your dating bio, you're tapping into a system that could be racking up costs of around $700,000 (£525,999) a day, according to estimates from tech analyst Dylan Patel.
That figure, from 2023, was based on OpenAI’s older GPT-3 model — so the latest version, GPT-4o, is likely even pricier to run. We can only guess at the eye-watering cost of keeping ChatGPT’s servers humming following the viral launch of OpenAI’s image generator, when social media was awash with Studio Ghibli-style self-portraits and group pics.
Why is it so expensive? Responding to your prompts, be they tricky maths problems or a courteous please or thank you, takes serious horsepower. This is generated using vast server farms packed with specialised chips called GPUs (graphics processing units), which chew through electricity and generate huge operational costs. Experts say these daily running costs, known in tech-speak as "inference costs", have overtaken even the sky-high price of training the models in the first place.
it's super fun seeing people love images in chatgpt.
— Sam Altman (@sama) March 27, 2025
but our GPUs are melting.
we are going to temporarily introduce some rate limits while we work on making it more efficient. hopefully won't be long!
chatgpt free tier will get 3 generations per day soon.
Is AI bad for the environment?
And then there's the environmental toll. The International Energy Agency says AI — and data centres and crypto mining — comprised nearly 2 per cent of global electricity demand in 202. The agency says that figure could double by 2026, roughly matching the energy needs of a country like Japan.
According to research from the University of Massachusetts Amherst, training a single large AI model can emit more carbon than five average American cars do over their entire lifetimes — manufacturing included. And with new models constantly replacing old ones, much of that energy is ultimately wasted.
Meanwhile, tech giants are guzzling water to keep their AI data centres cool. Microsoft’s water use jumped more than 1.7 billion gallons over the course of a year — enough to fill 2,500 Olympic swimming pools. Google’s wasn’t far behind, reporting a 48 per cent rise in emissions since 2019, driven largely by AI.
Passing the cost of AI on to consumers
With tech firms racing to plug AI into every product, the industry is grappling with a tough question: who will pay for all this? After luring users in with free tools, companies may soon look for new ways to recoup their eye-watering energy bills, whether charging more for premium features, throttling usage, or (shock, horror) introducing adverts.
OpenAI, for instance, now offers a $200 (£150) per month ChatGPT Pro subscription with unlimited use of its most advanced features, including its powerful video model, Sora. Despite usage limits for free users, ChatGPT still regularly buckles under the strain of demand. Meanwhile, the company is reportedly considering advertising as a new revenue stream, as it shifts further into for-profit territory.
Maybe if we ask nicely, OpenAI will allow us to keep using the insanely popular bot for free.
