One evening in the spring of 2015, I filmed a 15-second video out the window of an Amtrak train as it rattled across the barren flatlands of southern New Jersey. There’s nothing artful or interesting about the clip. All you see is a slanted rush of white and yellow lights. I can’t remember why I made it. Until a few days ago, I had never even watched it. And yet for the past nine years, that video has been sitting on a server in a data center somewhere, silently and invisibly taking a very small toll on our planet.
At some point since I made the video, the emissions of information and communications technology began to match those of the entire aviation sector. Data centers and data-transmission networks now account for as much as 1.5 percent of global electricity consumption, according to the International Energy Agency. In the years ahead, the advent of ubiquitous artificial intelligence could, as Matteo Wong wrote for The Atlantic last year, “push the web’s emissions to a tipping point”: Earlier this week, Google released a report showing that its emissions have grown substantially as a result of the AI boom, a major leap backwards from the net-zero goal it set just a few years ago.
With other forms of consumption that are bad for the planet, we all understand that the main burden of responsibility falls on the big players—industry, government, the rich and powerful. But we also acknowledge that everyone else has a part to play too. I stop running the water while I’m brushing my teeth. I carry groceries in a burlap tote. I turn off the lights whenever I step out of my apartment, regardless of whether I’m leaving for five minutes or a week.
[Read: AI is taking water from the desert]
Every time we make a new video or send an email, or post a photo of our latest meal, it’s like turning on a small light bulb that’ll never be turned off. This points to an uncomfortable, and eminently modern, question. “Everyone says it’s really bad to fly,” Tom Jackson, a professor at Loughborough University, in England, who studies the environmental impact of data, told me. “But also we’ve got to think about whether it’s really bad to carry on with our current digital practices.”
In other words: To help save the planet, should we be using less data? Given how much of modern life depends on megabytes and teraflops, the answer could be a key facet to living nobly in the AI age.
As the harms of global data use and storage have grown, it was only a matter of time until digital environmentalism became a thing. In 2020, the BBC reported on “a growing number of eco-conscious consumers trying to reduce their environmental impact online and on their phones.” An infographic for aspiring digital environmentalists, from the carbon-credit company Climate Impact Partners, says that you shouldn’t play videos if you only need the audio, for example. You should unsubscribe from newsletters you don’t read. And whenever possible, you ought to go to websites directly instead of finding them via a search engine.
The reality is that it’s not easy to get a clear sense of exactly how much these interventions help, or how much pollution our personal data are actually responsible for. “All these numbers are ferociously difficult to be exact about,” Mike Berners-Lee, the author of The Carbon Footprint of Everything, told me. Headlines about how, for example, sending fewer unnecessary emails and cutting down on our Netflix time could save thousands of tonnes of emissions every year have been shown to be wildly exaggerated. But even our minor contributions can add up quickly.
By my estimate, following a formula included in a recent research paper, storing my train video has created about 100 grams of CO2 over the past decade. At first blush, this is effectively nothing: less than one three-100ths of a percent of the yearly CO2 emissions from a pet cat. But data slough off us like skin cells. Last year, I sent 960 videos to the cloud. Because phones record videos in much higher quality these days, most of these clips are larger than that 15-second video from 2015. And like many other people, I have a sprawling digital footprint; many of my stored videos have been either sent to or received from at least one other person who is also storing them on one or two cloud platforms.
And 100 grams is just a ballpark figure, anyway: The real number could be radically higher or lower depending on variables such as the type of server data the video is stored on. A major factor is where it’s being housed. Google Cloud storage in Ohio, for example, creates five times more emissions than it does in Oregon; whether the grid is powered by clean energy makes a huge difference. Emissions will also fluctuate according to how often data are accessed and distributed: In 2020, Rabih Bashroush, a professor at the University of East London, estimated that every time Cristiano Ronaldo posts a photo on Instagram, the energy that would be needed to show the image to each of his followers—190 million at the time—could power a household for five to six years.
“We just need to start to think around the impact of every button we press ‘Send’ or ‘Upload’ on,” Jackson told me. As a first step, he suggests going back through your phone and computer and getting rid of all the data that you’ll never use again. (The industry term for such detritus is dark data; much of Jackson’s research focuses on teaching companies to reuse old information instead of making new bytes.) That’s easier said than done. When I was looking through old videos for this story, I found many clips that sparked cherished memories. None of these videos was particularly fascinating. But a data center had conserved the data for so long that watching them now transported me, joyfully, to a simpler time. Deciding whether to scrap any of these is not the same as deciding whether to turn a light bulb off when you step out of a room. “The light bulb, you can just come back and switch it back on,” Jackson admitted. “Once you’ve gotten rid of data, it’s gone.”
Even my feelings about the train video—which did not spark any fond memories—remain unresolved. For now, it’s still up there.
In a report published in 2021, Berners-Lee and a team of researchers found that if the information-and-communications sector is going to match the reductions necessary to keep global warming under the 1.5 degrees Celsius threshold, it will have to cut its carbon emissions by 42 percent by the end of this decade, and 72 percent by the end of the next.
Thinking that we could all chip in might be comforting. But others regard the whole premise of personal responsibility with profound suspicion—“micro-consumerist bollocks,” as the columnist George Monbiot puts it. Generating data is a fact of life. The goal of climate action, Berners-Lee said, is “to leave the world’s fossil fuel in the ground.” In the absence of systemic change, agonizing over our individual data-usage decisions is “like trying to hold back a flood with a bucket and spade.”
And if anything, the big players are making it harder for individuals to do even a very small part, as they roll out difficult-to-avoid AI features. Google recently announced that it is testing a Gmail feature that writes your emails with AI. In April, Meta announced new features for WhatsApp users in about a dozen countries, which means that when you search for a text message, you have to scroll past multiple suggested queries to “Ask Meta AI” before seeing your results. With Apple Intelligence, anyone with an iPhone, an iPad, or a Mac may struggle to check the weather or set a reminder without activating a large language model.
A similar phenomenon is happening with internet-connected products. Today, about 400 million homes worldwide have at least one “smart” device—refrigerators, dog collars, sex toys. By 2028, that figure is expected to nearly double. According to IBM, 90 percent of the material uploaded by these devices is dark data; it’s never used. Not all of that growth is fueled by customer demand. Recently, I had to settle for an electric toothbrush with less battery life and less scrubbing power because the top-of-the-range model had “smart” internet-enabled features that I don’t want anywhere near my mouth.
[Read: Phones will never be fun again]
We are, in other words, being ushered into a world where it’s harder to do anything without turning on a digital light bulb. This puts eco-conscious consumers in a bind. “Do you really want to be someone who can’t get their information properly because you’re too busy keeping the carbon footprint of your searches down?” Berners-Lee asked me.
In this context, straining to reduce our data emissions out of a sense of responsibility for the planet might be exactly what the industry wants, Kate Crawford, a professor at the University of Southern California at Annenberg and the author of Atlas of AI, told me. In an email, she explained that the term carbon footprint, which was popularized by British Petroleum, “contributed to a strategic shift away from systemic harms and industrial activities at scale toward individual accountability.”
It’s not hard to imagine the tech industry, which continues to be highly reliant on fossil fuels, leading a similar crusade. Multiple digital-carbon-footprint calculators are already available online. I recently used one such calculator, developed by a carbon-reduction consulting firm with an obvious interest in making prospective clients believe that their footprint is huge. According to the calculator, my yearly work activities generate as much CO2 as a Ford Focus driving 3,000 miles. The dazzle of such figures—believable or not—easily washes out larger questions. Could the industry do more so that a year of working from home isn’t as polluting as crossing the continent in a hatchback? And if ubiquitous AI turns that Focus into a diesel SUV, is it because the vox populi actually asked for it?
The digital carbon footprint could also obscure a deeper point. When we face a digital decision, the less data-thirsty option will often be the smarter one, regardless of whether we can measure how much it helps the environment. Blocking third-party tracking when we’re on the web, for example, not only reduces the power consumption of internet browsing; it’s also good for your privacy. AI-based search tools not only use more fossil fuels and water; they are prone to providing patently false information. Disconnecting our home appliances from the cloud makes them less vulnerable to hacking.
More fundamentally, maybe we don’t need to turn everything into data. If I put down my phone the next time I’m on a train, it won’t save the planet. But I’ll be looking out the window with my own eyes, creating a memory that emits no carbon at all.
