How much electricity are your workers chewing through with AI, and should you be worried?

British tech workers are using AI for 35 tasks per day across their teams, new research suggests. While that works out to just a handful of prompts for most employees per day, it's starting to build up to serious electricity usage.

That's according to Uswitch Business Energy, which found the average workplace team generates 24 prompts per day, including writing emails, automating admin tasks and analysing data.

But that's across the whole team, not per worker. Instead, Uswitch combined its data with YouGov research for some back-of-the-envelope extrapolations.

Survey data shows 32% of British workers use AI at work, and with an average team size of 12, and an estimated average AI prompt generation of two per day, that adds up to an estimated 22 million AI prompts every working day.

That sounds like a lot, but there's just shy of 26 million full-time workers in the UK, and not all of them are sitting at computers.

IT and telecoms workers account the most prompts, at 35 per team, followed by finance at 29, Education at 22, and healthcare at 21, the report found.

Workers aged 25-34 had an average of 30 AI-assisted tasks on their teams each day, versus 27 for younger workers and just 18 for those aged over 55.

Are energy concerns warranted?

Uswitch's energy division warned that the shift to AI is eating up a fair amount of electricity.

The switching company points to Google's own estimate that a Gemini text prompt uses 0.24 watt-hours (Wh) of energy, meaning British employees are using 5.28 MWh of electricity each and every work day, or about 1.37 gigawatt-hours (GWh) each year.

That works out to enough electricity to power 508 British homes annually. Of course, that energy use is over in a Google data center, rather than at point of use.

"AI has rapidly evolved from an emerging technology into an everyday business tool," said Ben Gallizzi, energy expert at Uswitch. "As more organizations adopt AI to support tasks such as content creation, customer service and administration, the cumulative energy demand from these tools will continue to grow."

Developing and training a model is an energy hungry process, with estimates that training current cutting edge models uses between 20-25MW of power over three months, or between 43 GWh and 54 GWh in total.

However, research by MIT Technology Review suggests that those massive figures make up a small amount of the total energy use of AI.

As much as 90% of its footprint lies in inference, with small text based queries, image generation, and the rest adding up when used by so many people.

"Although each prompt uses a relatively small amount of energy, the scale of adoption means businesses are collectively generating billions of AI interactions every week," said Gallizzi, although the billions would be globally, not just in the UK.

Indeed, OpenAI said last year that it was seeing 2.5 billion queries on ChatGPT each day.

Gallizzi added: "As organizations continue to embrace AI, understanding the impact this could have on future electricity demand will become increasingly important."

Time to track

Uswitch suggested businesses need to be aware of these figures, just as they track energy use from more mundane aspects of their offices. As noted, this usage will be off-site in a data center run by AI companies, so it's more about considering overall usage and impact than direct costs.

"Businesses already carefully monitor energy use from equipment, heating and lighting," Gallizzi said. "As AI becomes more deeply embedded in workplace processes, organizations may also need to consider how digital tools contribute to their overall energy footprint."

Wanton AI use could be reined in — not by electricity footprint concerns, but by rising prices and shifting billing models. That's led to companies pushing to end so-called "tokenmaxxxing" and encouraging more sensible use of AI credits.

FOLLOW US ON SOCIAL MEDIA

Follow ITPro on Google News and add us as a preferred source to keep tabs on all our latest news, analysis, views, and reviews.

You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.