Nearly 20 years ago, the British mathematician Clive Humby coined a snappy phrase that has turned into a platitude: “data is the new oil”. He wasn’t wrong. We have an insatiable appetite for data, we can’t stop generating it, and, just like oil, it’s turning out to be bad news for the environment.
So the Guardian set me a challenge: to try to give a sense of how much data an average person uses in a day, and what the carbon footprint of normal online activity might be. To do that, I tried to tot up the sorts of things I and millions of others do every day, and how that tracks back through the melange of messaging services, social networks, applications and tools, to the datacentres that keep our digital lives going.
My own carbon tally gets off to a bad start, and it is not even my fault. The email from my editor asking me to try to quantify quite how much data a single person uses in a day is itself contributing to my footprint. If the editor took 10 minutes to write the email – likely, given it was quite detailed – and it took me three minutes to read, and if it was sent from a laptop and received on one, then we have generated 17g of carbon dioxide (CO2) emissions already, according to estimates by Mike Berners-Lee, a professor at Lancaster University’s Environment Centre, and the author of How Bad are Bananas? The Carbon Footprint of Everything.
My frantic emails to people asking to speak to them for this story pump out more carbon at a prodigious rate. And though 17g of CO2 is insignificant compared with the 384.2m tonnes of net emissions the UK as a whole is responsible for each year, it all adds up.
All those emails and videos and games don’t just appear on our screens by magic. Everything we do digitally involves the vast transfer of data through the internet from one place to another, brokered through datacentres. Datacentres are vast premises full of computer servers that store data. The idea behind them is to reduce what the data industry calls “latency”, the time between you typing in a web address or clicking on an app button, and the content you are requesting being delivered to you. Everything on the internet, every link you click, every video you watch, is physically stored in a datacentre somewhere.
Datacentres are big business, and vast numbers of them are being built around the world. In the UK, Amazon has just announced plans to invest £8bn over the next five years building new datacentres and maintaining those it already has, “supporting 14,000 jobs annually”. That comes on top of £3bn already spent in the UK by Amazon’s cloud computing arm since 2020. Google is spending $1bn on a new centre at a 133,500 sq metre (33-acre) site in Hertfordshire, and at the end of last year Microsoft committed to £2.5bn of investment in the next three years, more than doubling its datacentre footprint in this country.
The reason for this is simple: demand is increasing at alarming rates. Americans used 100tn megabytes of wireless data in 2023, a record-breaking increase of 36% on the previous year – that’s enough to download Candy Crush Saga 265bn times.
It is a lot of data, and a lot of energy is required to serve that data to us, plus a lot of water to keep all those servers cool. In fact, Ireland, the Netherlands and Singapore are so worried about the energy impact of datacentres that they have imposed moratoria on new developments. When Google announced its environmental impact earlier this year, it revealed its own greenhouse gas emissions had risen 48% in the last five years, and 13% in the last 12 months, largely driven by increased datacentre demand to service its AI needs. Now big tech companies have come up with another solution to try to solve the looming energy crisis: their own nuclear power plants. Microsoft has struck a deal to recommission the Three Mile Island site in Pennsylvania, Google recently announced plans to build six or seven new small reactors to meet its anticipated energy needs. There’s no way round it: a steady stream of environmental harm is coming from our everyday actions – activities that we often don’t think about in relation to the target of limiting global heating to below 1.5C.
“You will run into this pretty much anywhere during the day,” says Alex de Vries, who researches the carbon footprint of our day-to-day lives at the Vrije Universiteit Amsterdam in the Netherlands. “Digital applications are so deeply embedded in our lives nowadays, it’s really hard to avoid. The thing is, when you’re using them, it’s not like you have something popping up in the screen telling you, like: ‘Hey, be aware, this activity has this carbon footprint.’”
De Vries also runs the Digiconomist website, which tries to track – where possible – the environmental impact of these things. That “where possible” is an important caveat. “It’s incredibly hard to figure out that information,” says de Vries.
In the absence of reliable figures from the companies themselves, educated guesses are often all we can rely on. Case in point: estimates of the proportion of world energy use that the internet makes up range from 3.7% to 10%, depending on who is counting. One estimate by Zero Waste Scotland suggests all our online activity generates an average of 8.62kg of CO2 a week (about 448kg a year), or about 30 miles in an average-sized petrol car. But a German estimate (which also includes the emissions created by the production digital devices themselves) says we expend around twice that, roughly 850kg a year.
People struggle with two key problems when trying to wrap their heads around their data usage and resultant carbon impact, says De Vries. One: everything is digital, and therefore not tangible. “If you’re holding a pen and a piece of paper, you can get some idea of what might be necessary to make this product,” he says. “But if you’re using a digital application, what’s really going into that to make all of that happen? A lot of people simply will have no clue what that looks like.”
The other issue is that the tech companies are really good at making things work. “You probably don’t even know what is in [an application],” says De Vries. You press the button, and the Netflix series starts.
Companies such as Netflix are disarmingly honest about their data usage: if you keep your video quality on “low”, you use a paltry 300MB an hour of data on a streaming service such as Netflix. If you want to watch things in HD, though, you ramp up to 3GB an hour when looking at the most detailed scenes. If you are a movie connoisseur, your 4K streaming uses up to 7GB an hour.
But while few would argue we should spend less time in front of streaming services, the environmental impact of all that binge-watching appears to be comparatively low. A 2020 analysis by the International Energy Agency (IEA) found that watching an hour of Netflix was equivalent to boiling a kettle once: about 36g of CO2.
There are other variables to take into account, though: the energy consumption of the device you are watching on, for example (Netflix says 70% of its viewers use televisions, which are more energy-hungry than mobile phones); or how the electricity you are using is generated (nuclear or wind is less carbon-emitting than coal or gas).
If you want to gossip about the latest episode with your friends, that also comes with an environmental toll. The average WhatsApp group chat uses 2.35kg of CO2 a week, Zero Waste Scotland calculated. (To blunt the impact slightly, rely more on emojis – which are stored locally on your device – than reaction gifs, which have to be downloaded afresh from datacentres.) Listening to music online also comes at an environmental cost, although it is estimated that you can stream music for five hours before you will emit more CO2 – 288g – than is involved in making a CD in a case. Like many tech companies, Spotify has committed to reaching net zero emissions, in its case, by 2030.
Big tech companies buy carbon credits and offsets to try to mitigate the impact of their activity, but it’s often seen as a poor attempt at atonement for the environmental impact they cause. There are also questions about the extent to which firms’ reported datacentre emissions are capturing the whole picture. A recent Guardian analysis found that real emissions between 2020 and 2022 from datacentres owned by the four big tech companies, Google, Microsoft, Meta and Apple, were likely to be 662% higher than officially reported.
The tech industry’s warm embrace of generative AI has complicated things even further. It is becoming increasingly difficult to avoid. Type certain searches into Google and you will be given an “AI overview”, as Google calls them, which summarises key information from the results the search engine finds and presents it in a simple set of bullet points, alongside associated links. And you can’t turn it off. “AI Overviews are a core Google Search feature,” the company says.
“Generative AI hasn’t necessarily added very many new use cases,” says Sasha Luccioni, AI and climate lead at AI company Hugging Face. “It’s adding more compute and more environmental impacts to existing use cases.” The problem is that we don’t fully know how much. “None of the corporates, and none of the proprietary models, have published any numbers,” she says. De Vries’s research suggests that AI-powered search results use 10 times the power that non-AI searches do.
All this is before you get into the conscious use of generative AI tools such as ChatGPT or Anthropic’s Claude chatbot – where you are going to their websites or opening their apps, and taking part. Here, we are also in the dark about how much data, and therefore how much energy and water, generative AI uses. The best information we have is from informed third-party estimates: training GPT-3, a precursor to the current model, used an estimated 5.4m litres of water, according to one academic study, and produced as much CO2 as would be generated by flying between New York and San Francisco 550 times.
I recently published a book on AI and as part of that, I have been touring and giving talks about AI’s impact on our world. In my favourite set of slides that I present there is a party trick. To highlight concerns around copyright in generative AI, I ask ChatGPT’s image generator, Dall-E, to produce a depiction of whichever place I’m in, in the style of Vincent van Gogh’s The Starry Night.
The gimmick always gets a laugh and serves its purpose: it shows how often the AI system has seen that painting by the ability to mimic its brushstrokes. But I always feel guilty. Because each time I do that, whether in Chipping Campden or Vilnius, I’m using data. About halfway through my book tour, I started adding a couple of slides immediately afterwards on the environmental impact of AI.
So besides stopping generating bootleg Van Goghs, what should those of us conscious about our environmental footprint do? Luccioni advocates for “digital sobriety”: being mindful about how we use AI. “You don’t need to be using these new AI tools for everything,” she says. “There are applications that are useful, but there’s a lot of cases where you really don’t need them.” The same approach holds true for everything digital: think twice, text once.
Your data diet
Estimating how much data your daily activities use is an art not a science, but here are best estimates of how much you are gobbling up online.
Listening to a podcast: 20-100MB an hour
Watching Netflix: 3GB an hour at HD quality
Online shopping: Consider the data size of any images you browse, which can be big, before even thinking of the environmental impact of your delivery
WhatsApp text message: 1-5KB a message, on average
WhatsApp voice call: 400KB-1MB a minute
WhatsApp video call: 2.5-15MB a minute
Average pre-AI Google search: 500KB for a text-based search
Average post-AI Google search: No one knows …
Sending an email: Depends on the size of the message, but about 75KB on average
Sending an email with photo attachment: As above, plus the size of the attachment
Downloading an album on Spotify: Depends on your audio quality, but around 72MB for an hour-long album
Playing a game of Fortnite: Between 45 and 100MB an hour