I can still remember the floppy disks used by the first computer my family ever owned, a Commodore 128. They were the truly floppy type, 5 ¼ by 5 ¼ inches, a flexible circle of plastic with a magnetic coating, encased in a plastic sleeve with a few strategic holes. They could hold up to 360 kilobytes of data, which was useful at the time but today ten of them wouldn’t be enough to hold a single photograph from my mobile phone.
My family’s second computer was one of the early Apple Macintosh computers, which used the smaller size of floppy disk – 3 ½ inches. These were floppy in name only. The magnetic disk inside had a rigid plastic casing which meant they were much less vulnerable to damage. They could also hold more data, up to 2 megabytes. Two of them combined might just about hold some of the photographs on my phone. On the other hand, my whole MSc thesis could be fitted on one of them (excluding the images, which were film photographs pasted in by hand and then copied by the printer at great expense).
It's hard to imagine now. These computers ran without the internet. The only way to transfer data from one to the other was by temperamental pieces of plastic. But I can remember how revolutionary floppy disks were. For my first two years at university, all my assignments were handwritten. By the time of my MSc, I could carry my assignments from place to place, writing them on one computer or another, depending on what was available, and then taking them elsewhere to print.
There was something else new when I was working on my MSc. Through the institute where I did part of my research, I was given an email address. Then, when I got my first full-time job, I was able to search the internet for information. I admit, I didn’t find much. Books were still the mainstay.
But things were changing, and changing fast. Less than ten years later, I was fighting to keep any books at my workplace at all. Although there was a lot of information which was not on the internet, the leadership of government departments had decided that libraries were unnecessary. It was only through some determined effort that we kept any books at all.
It’s a strange feeling, taking this nostalgia trip through my personal history with computer data storage. I still have memories, as vivid as photographs, of sitting in the spare bedroom in the eerie light of the computer screen, playing the first game we had for our first computer – Summer Games, which featured various sports from the Olympics. I have another image from the time when the computer sat in the front hall and I typed up assignments which I had handwritten first, because in those days I couldn’t type and think at the same time. I remember sounds too, the high scratching sound of the dot-matrix printer and the sporadic scraping of a floppy disk being read in the disk drive. And of course, the sound of a computer dialling up to the internet.
So much has changed. If I’m in a positive frame of mind, I think about how much information is available to me now, and how easy it is to find. I think about how wonderful it is to be able to find academic papers on Google Scholar. I can get around 90% of what I need for my work free, with a bit of ferreting about. For recent papers which aren’t available free, I’m often able to get them by emailing a researcher on the other side of the country, or the world. I download the papers into my reference management software (Zotero, which I recommend highly as an excellent free tool). It’s easy to keep track of what I’ve used where, all without using a single piece of paper.
Of course, the huge amount of information available creates a problem. It isn’t all equally useful. When I was a student, I never had to think about the quality of information that I was finding in the university library. If there was a journal in the library, someone had already made a judgement that it was worth looking at. Today, evaluating the quality of online information is a real skill, something I work to improve all the time. It’s something that is a struggle for many (I’ve put together some information here and here if you’d like to know more). I went to a talk recently where the situation was described as polluted information. That’s a good description. Our information streams are as toxic as some of our lakes and rivers.
There’s another issue with having everything online. A lot of paper has been saved by having fewer printed journals and not photocopying everything. But online data still has an environmental cost. How much of a cost? That is what I was curious to find out.
Until recently, most of the information I had on my personal computer was stored on that computer’s hard disk and backed up on external hard drives. When I worked in office jobs, however, we stopped storing much information on our individual computers at least 25 years ago. Instead, information was stored on servers which were connected to our computers by cables. I can remember bulky cabinets in out-of-the-way parts of the offices, which made a constant loud hum.
The hum that I heard in those old office servers mostly came from cooling fans. Computers generate heat when they operate, an inevitable consequence of running on electricity. My previous computer used to have the occasional glitch where it would heat up alarmingly and stay hot for hours – I solved the problem with that mainstay of tech support: turning it off and then on again. But that was just a small laptop. When a lot of computer power is concentrated in one location, such as in a server room, the heat can be considerable.
But by the time I stopped working in offices (three years ago, when I switched to fully working from home), server rooms were long-gone. By that time, my office was storing all its information in data centres. At home, I still had some files which were stored only on my computer and backed up on external hard drives, but that was becoming a rarity. Eventually, I too reached the point where everything I had was backed up in what we casually call the cloud.
It's such an evocative term, cloud, conjuring images of delicate formations in the sky, something nebulous and insubstantial, or perhaps the harbinger of refreshing rain. When we talk about data storage, the cloud is something quite different. It’s made up of data centres, also known as server farms, but even that term is euphemistic. There’s nothing bucolic about a server farm.
The largest server farms are massive. The world’s largest is 67 hectares, which is considerably larger than the island of Rasgetheemu in the Maldives, which I wrote about a few weeks ago. They are basically enormous warehouses filled with cabinets. And they use huge amounts of energy. Globally, around 1.5% of energy consumption is used to run server farms, and a similar amount is used in transmitting data around the world. In terms of carbon emission, it’s around 1% of energy-related carbon emissions. That isn’t huge, it’s just under half the amount from aviation and only a tenth of that emitted by cars. We also aren’t wasting as much as we used to – moving from servers located in offices to server farms has allowed efficiency gains. Perhaps, in proportion to the benefit we receive, it’s worth it, but it’s not nothing. And energy isn’t the only resource used by server farms.
The humming servers which used to sit in remote corners of our offices were cooled by fans, but fans aren’t efficient enough to cool server farms. Instead, they are increasingly using water. A medium-sized server farm uses more water than two 18-hole golf courses. Data allows us to do many useful things, but we also waste a lot. One small example – all of my photographs and emails are backed up online, and I’m basically a hoarder and don’t delete anything except images where I accidentally photographed the lens cap or the floor, and emails which are either advertising or spam. Video is a particularly heavy user of data, so all our video calls and binge-watching TV are contributing.
Partly, the water use of server farms is linked to the electricity consumption. Many methods of electricity consumption use water – for example, fossil fuel power plants use the fossil fuels to heat water to steam to drive turbines. Hydro-electric plants also use water, because large amounts evaporate from hydro lakes. It does depend on the climate where the hydro lake is – the Waikato lakes will be very different from Lake Mead, which is in the Mojave Desert where the dry climate means much more water is lost to evaporation.
This electricity-linked water consumption can’t be ignored, but it’s also an issue with everything which uses electricity – it’s not a problem specific to server farms. But there is another way that they use water – for cooling, as an alternative to running fans. We don’t know exactly how much water is used – companies often argue that the figures are a trade secret. One news source suggested that Google, one of the three major providers of data storage, used the equivalent of 37 golf courses in 2022. Data centres are being built in New Zealand, and it’s unclear how much water they will use.
In some parts of the world, server farms are beginning to compete for scarce water supplies. This is the case in Uruguay, where there are shortages of drinking water in Montevideo, the largest city, despite massive quantities being used in agriculture. Earlier this year, Google created controversy with a proposal for a server farm in Uruguary which would use enough water to supply 55,000 people, although they’ve since scaled back the proposal.
When it comes to energy and water consumption, not all internet use is the same. There are a couple of internet activities which are particularly rapacious, and over the last few years they have quietly been consuming more and more energy and water without much attention.
The first is cryptocurrency. I won’t even pretend to understand how it works, but the short version is that creating new cryptocurrency – termed “mining” – is done by solving complex computer puzzles. This uses staggering quantities of computing power. Cryptocurrency transactions also use a lot of computing power. There are lots of numbers around, but it’s hard to pin down an overall figure for just how much cryptocurrency uses, because the situation is changing so rapidly.
A 2022 article suggested that cryptocurrency used double the electricity of Amazon, Google, Microsoft, Facebook, and Apple combined – crucially, the first three are responsible for the largest server farms. An estimate from January 2023 suggested that in the USA alone, cryptocurrency was responsible for carbon emissions between those of New Zealand, at the lower end, and North Korea at the high end. But an article from December last year suggested that global production of Bitcoin, the most well-known cryptocurrency, doubled in 2023. One estimate suggested that Bitcoin mining alone was responsible for 0.2-0.9% of global electricity use in 2023, about the same amount as Australia.
The other area which is using an increasing amount of energy is artificial intelligence. It’s too new to have a good understanding about just how much energy and water it’s going to use, but even simple activities conducted using artificial intelligence are energy- and water-hungry. An internet search using artificial intelligence uses many times more energy than a standard Google search, for example. In 2022, both Microsoft and Google had significant increases in water consumption (34% and 22% respectively). That’s a long time ago in terms of artificial intelligence – Chat GPT was released at the end of November 2022. How much are they using now?
In the widening debate about all the things which we can now do with computers, and what we might do in the future, we need to consider what they might mean for climate change and the environment, even if the total impact is currently small in comparison with other areas, such as transport.
Great info. I’m curious if the water used for electricity has to be desalinated.
It's an interesting problem and you've (as usual) covered it well. My substantive career has been designing and operating large technology infrastructure including data centres.
You are absolutely correct about "AI," the power required to make it operate is staggering, if not published. NVIDIA is a chip maker to watch, they are increasingly selling into that market. Watch their share price. The current commercial offering, ChatGPT 4, is not just a natural language system it has more and more add-ons including image and film creation.
There is room to lower the impact of processing by moving workloads in real time and designing data centres that use more natural cooling. Iceland has experimented with this, data centres in cooler climates that use the outside temperature to reduce heat.
Loads can also be moved geographically based on heat, again. Why pay large power bills in one place when you can reduce it by moving somewhere colder.
Big companies are also experimenting with oil cooling and even things like underwater ocean based data centres.
There are also some more alarming things happening.
There is a socio-economic divide coming driven by AI. The commercial versions costs money and is "smarter" than the free versions. Those with money will have a knowledge advantage over those who do not.
There are concerns around the world about the amount of power data centres are, and will, consume. Ireland is a good example of the problematic flourishing of data centres without considering power infrastructure, or lack of it. In New Zealand there are roughly seven large data centres being built right now with more planned.
No one here is considering the overload load on power and water in that context.
There is a likelihood that data centre power use will outstrip renewables or even traditional (fossil fuel) generation. There is quite a bit of chat about deploying nuclear power to make up that gap. Not the traditional nuclear power stations but smaller more "portable" devices. Think of how nuclear submarines are powered.
There is also a slightly science fiction outlier theory that once AI reaches a certain intelligence it will realise it needs to compete for resources, energy, and seek self survival by redirecting energy from non-essential things (humans) to itself. And, it will seek to propagate as a further form of survival, creating more energy eating instances of itself and protecting them.
One thing is for certain, the exponential growth of data centres is absolutely assured and the only thing holding it back right now is manufacturing chips. AI is already being used to shorten that timeline and make chips more powerful. Clearly, there is an ecological impact with the need to mine more rare materials.