The Human and Environmental Cost of Doing More (and More... And More)
Researched and written by Katy Wicks - Happy Citta Founder
In this article, I'm going to talk about some complicated facts and figures and how they've changed over time, but here are some definitions to clarify the three categories of the jargon you might come across here:
Productivity: How much work we complete
Wellbeing: How people feel about their jobs
Environment: How our outputs impact the earth
If the statistics and explanations are not of great interest to you, you can skip to the end for a pretty graph!
Since the changes to working life in 2020, many of us have become used to doing more of our work through screens, shared drives and online tools. There’s also a growing sense that we should be doing more than the job we were actually employed to do: taking on the extra task, the new initiative, the side quest that becomes a standard part of the job.
AI can be genuinely useful in that context. It can help us get started, summarise information or tackle a repetitive task. But I find myself wondering what happens to the time it saves. Does it give us breathing space? Or does it make room for another task, target or expectation?
There is a physical infrastructure behind our digital work which includes servers, storage systems, networks and cooling. The International Energy Agency estimates that data centres used around 415 terawatt-hours of electricity worldwide in 2024 — about 1.5% of global electricity use. Its central projection puts that at around 945 terawatt-hours by 2030. These figures cover data centres supporting many kinds of digital services, not AI alone, though the IEA identifies AI as an important driver of future growth.
Closer to home, the Department for Energy Security and Net Zero estimated that data centres used 4.5 terawatt-hours of electricity from the Great Britain grid in 2024, up from 3.2 terawatt-hours in 2020. Again, that is a measure of data centres overall, not a personal footprint for office workers or a figure for AI alone.
Breaking down the jargon
Data centres didn’t appear with AI. In the 1990s, organisations already had mainframes and server rooms, alongside the computers, screens and network equipment in their offices. But the infrastructure looked different, and it wasn’t measured as one neat global “data centre” category. A US estimate for 1999, for example, counted office and network equipment across several types, including servers, minicomputers and mainframes.
One peer-reviewed estimate gives us a rough starting point for global data-centre electricity use: about 71 terawatt-hours in 2000, rising to around 153 in 2005, so more than double. A later estimate put 2010 use at roughly 1.1% to 1.5% of all global electricity!
A Bank of England study estimated that ICT capital (computers, servers, communications networks and related technology) contributed 28% of UK market-sector productivity growth in 1990–2000, and 47% in 1995–2000. Alongside the rise in data-centre electricity and the broader increase in output per hour since 1997, that helps us see a plausible relationship between technology, productivity and energy demand.
It still doesn’t show that one caused the other. These figures cover different activities and periods, and data centres support far more than office work.
And what does one tonne of CO₂e look like in everyday terms?
CO₂e is a way of expressing the warming impact of different greenhouse gases using a common unit. Using the 2026 UK electricity conversion factor, one tonne of CO₂e is roughly the emissions associated with generating 7,600 kilowatt-hours of electricity. According to Offgem, that is about three years of a home’s entire electricity use.
Bananas for scale...
A global-average estimate puts bananas at about 0.86 kilograms of CO₂e per kilogram of fruit. On that basis, if we assume each banana weighs about 100 grams, one tonne of CO₂e is roughly the growing-and-delivery footprint of around 12,000 bananas.
Let's bear the above in mind here, so:
A public-sector case study
In 2022, greenhouse-gas emissions were estimated at 3.3 tonnes of CO₂e per employee in the “public administration and defence” industry. That is a sector average, not the measured footprint of an individual civil servant or of their computer use.
The same table gave different averages for other largely desk-based industries, including 1.4 tonnes for administrative and support services and 0.6 tonnes for information and communication.
The source also described how ONS specifically was trying to reduce its own impact. It reported a 65% reduction in greenhouse-gas emissions between 2017/18 and 2023/24, alongside practical changes such as using solar and biomass energy, auditing resource use, replacing lighting and improving waste recycling.
Its Greening Government target for 2021–25 was a 38% reduction; that target and the reported 65% reduction use different periods, so they shouldn’t be read as a like-for-like comparison, but still an overall positive story.
The source also flagged that international flights are an area needing further attention, while recognising that travel and office attendance can be shaped by operational requirements.
For me, that makes the example more useful than a simple “switch off your computer” message. It shows how an organisation can measure its impact, make changes to the parts it can control and keep scrutinising the areas where emissions remain difficult to reduce.
So, People?
I wouldn’t be me if I didn’t pause to think about the people impact too. Productivity figures can make work sound like a neat equation of more output per hour, more tools, more things completed. But there is a person at the other end of every target.
One measure I looked at was how much ownership people have over their day-to-day tasks. In Britain’s Skills and Employment Survey, the share of employees with significant influence over their immediate work fell from 62% in 1992 to 44% in 2012 and 34% in 2024. That isn’t a direct wellbeing score, and it doesn’t tell us what caused the change. But it does give us something important to ask: how much say do people have in what they do, how they do it, and how hard they work?
The survey also found that 64.4% of people with high task discretion were completely or very satisfied with their jobs, compared with 46% of those with low or no discretion. That relationship doesn’t prove that control alone creates wellbeing. It does remind me that how we pursue productivity matters.
If technology gives us a way to do more and do it quicker, does that mean people get more space, choice and recovery?
Personally, I feel like the answer is no. If you read last week's blog about your best work becoming your expected baseline standard for your employer, you may recognise that I'm pursuing the same topic this week but with a new environmental angle!
What does comparing all of the datasets look like?
I've said in previous talks and blogs, I am neither an analyst, nor a data engineer, but I do appreciate a good visual to help put all of these figures and studies into perspective. Call me a hypocrite, but I asked AI to kindly create a line graph to represent all of the different datasets I've talked about here, so we can see a comparison, particularly from the 90's through to the 2024 figures.
This is not the bold claim that AI and increasing productivity is the sole cause of either increased emissions, or decreased workplace wellbeing, but I'd say there is certainly a correlation, even if not causation!
Since workplace wellbeing isn't something that's received consistent measurement, I have included the decreasing scores for 'task discretion', which I mentioned earlier have been linked to decreases in workplace satisfaction.
It’s tempting to ask how much carbon one AI prompt produces. But there isn’t one reliable number that applies to every prompt and the answer depends on the model, the work it does and the infrastructure supporting it. What we can say is that data storage has an ongoing environmental cost.
Government guidance describes that cost as coming largely from the electricity and water used to run and cool data centres, and recommends reviewing whether stored information is still needed.
That makes me think about all the screenshots, drafts, duplicate files and AI-generated text we save because it might be useful one day.
Any one item may seem insignificant, but the habit of creating and keeping more digital material, alongside the expectation that we should produce more, deserves a little thought.
I don't write about this to instil guilt for using a tool that helps with your work, but I'd like us all to consider the questions:
When a tool saves time, who gets that time back?
When a task becomes quicker, do we stop and consider whether it still needs doing? And when we keep more data, do we know why we’re keeping it?
I think that more useful measure of productivity would look less at how much we can produce in a day and more at whether the work is worthwhile, whether the people doing it have enough room to think, and whether the digital resources we use are serving a real purpose.
What do you think would change if we (and the organisations we work for) treated our attention, our working hours and our digital storage as resources to use thoughtfully, rather than as a continuously expanding capacity to fill with more things?
Comments