Andrew Rice and Adrian Friday are members of the Working Group chaired by Andy Hopper FRS and leading the Royal Society’s Digital technology and the planet project.

View of the Earth's atmosphere from space

In recent months we have witnessed the extraordinary power of the Internet. It has allowed doctors and scientists to instantly share information about the spreading coronavirus pandemic, it has been a source of information for billions of people. It has allowed many to work and collaborate from home, and has been a life-line for people trying to stay in touch with friends, family and the outside world. The Internet has brought people together at the very time they are being kept apart.

With more than half of the world’s population confined to their homes in an effort to control the pandemic, there have been many images and stories of the impact this is having on the environment. Air pollution and carbon emissions have plummeted due to reduced levels of industrial activity and transport. Rather than commute to work, many employees have instead logged on remotely over the Internet. But this online activity is not free from emissions either.

Most of us use the Internet largely oblivious to the complex infrastructure of cables, fibres, computers, data centres, routers, servers, repeaters, satellites, radio masts and energy needed to keep it running. And the Internet uses more energy than we might realise – as powerful as the Internet is, it also takes a lot of power to build, install and run the web of infrastructure that supports it. This in turn releases carbon dioxide (CO2) into the atmosphere, contributing to the climate crisis. As the scale of the Internet and its infrastructure grows, we should no longer ignore these environmental impacts.

Some attempts to estimate the carbon footprint of the Internet have led to dramatic headlines about how much CO2 is released with each online search we perform and email we send. But knowing precisely how much carbon is released into the atmosphere from a single online activity like this is actually pretty tricky, depending on which parts of the chain of consumer devices, wireless networks, data centres and Internet backbone you include in your calculations. It also depends on how much of the energy being used by each piece of infrastructure is attributable to your task. At any given moment this huge web of equipment is carrying out many different computational activities simultaneously.

Trying to attribute the amount of carbon dioxide released by sending a single email or firing off a message on social media is rather like working out how much food you need to eat in order to read a sentence out loud. The problem is that while you are speaking, you are also breathing, digesting, moving and performing a myriad of other activities all at the same time.

It’s also the wrong question to ask. What’s interesting is not how much energy goes into a single breath or sound. Rather, how much of the food you ate to grow your body in the first place should you count? This is also true of the Internet. Every device we use to access it and piece of equipment in the chain of infrastructure took energy to manufacture, deliver and install. Manufacturing microchips requires a lot of energy. The high levels of purification needed for the silicon, together with the large amounts of purified water needed to wash them, makes them incredibly energy intensive in comparison to the mass of the final product.

In many cases the use of a piece of computer equipment accounts for little more than a third of the energy it consumes during its lifetime. The rest comes from the energy needed to manufacture and deliver it in the first place. It means years of sending emails, performing searches and watching online videos to generate the same amount of greenhouse gases as are released in the manufacture of our devices.

If we are eager to reduce the impact the Internet has on the environment, it makes much more sense to look at our relationship with the technology hardware we use. While a phone requires less energy to run compared to a laptop or desktop computer, we hold onto them far less long – most of us will replace our phones every year or so while a PC will last four or five.

Extending the amount of time we use our equipment for is perhaps then an easy way to reduce the impact our digital technology has on the environment. One project at the University of Edinburgh, for example, found that using a single desktop computer and monitor for six years instead of four, could avoid the release of 190kg CO2e (CO2 equivalent) (PDF).

But even if we choose to treat our own IT equipment with a bit more affection, there are still problems elsewhere in the system that require a more wholesale change. The data centres that do much of the computational “work” on the Internet are run in ways that accept the processors, chips and hard-drives are effectively disposable items that will have short lifespans of just a few years before they fail. Computers are commodities in the workplace, discarded when they can no longer be supported, not because they are obsolete.

Is there a better way? Certainly data centres have been making great strides in tackling the amount of energy they need to run. Made up of rooms filled with row upon row of racks filled with dedicated server computers, data centres are hot places that need to be cooled, and this is where much of the energy needed to run them is used.

Globally, data centres are thought to have used around 205TWh of electricity in 2018 – about 1% of the world’s electricity consumption. But while the workload of data centres has soared in recent years – by up to 550% since 2010 according to one estimate – the amount of energy they use has remained roughly the same. By moving to larger, purpose built data centres – in some cases tens of thousands of square metres in size – major Internet companies have been able to design these facilities in ways that make them far more efficient. In some parts of the world, the hot exhaust gas has been used to provide heating for communities living nearby. Piping in water, which is a better heat conductor than air, is another trick being tried, while data centre engineers have made other gains to reduce the energy load of their systems – stripping out the banks of flashing LED lights, for example, and customising hardware so they produce less heat.

There is also a question of where the electricity to power all the systems that the Internet relies upon comes from. A data centre in California, for example, will have far more access to renewable energy than one in China or India, for example – principally due to the mix of electricity generation in these locations.

Many major Internet companies, including Google, Microsoft, Facebook and Amazon, all claim to run their data centres using renewable energy where they can, or buying it elsewhere to compensate for sites where they cannot. This commitment by the industry to reduce its carbon footprint from data centres, has been a key driver of growth in renewable energy. And with the industry now required to meet ambitious targets for lowering their emissions, they will have to do more. Some predict that communication technology could contribute up to 23% of global carbon emissions by 2030 unless there are changes.

So, will efficiency improvements always outstrip our growing demand? As our technology becomes more efficient, there is a tendency to simply use the extra capacity for something else rather than bank the savings. The adoption of home fibre broadband and 4G has lowered the energy cost of getting more data to our homes, but this has enabled us to invent streaming video services, UHD smart TVs and high-resolution phones to watch our content.

It is possible to see similar “rebound” effects occurring in situations where moving to digital technologies appears to provide a win in terms of energy use. Streaming music and videos is generally greener than buying physical copies on CDs and DVDs. Smartphones are also smaller than a television, and are used for many other tasks beyond watching videos, so the energy per video view has gone down. But the changes in phone technology has also led to shifts in user behaviour. We now consume videos differently – instead of gathering around the television together to watch a programme, we now watch individually, glued to our own personal screens, gorging on ‘box sets’, which means the energy needed to stream all this data goes up. Similarly, can we really expect online video calls to replace flying entirely, or will it merely facilitate bonds between distant teams that will encourage them to travel even more?

As we embrace other new technologies, such as cryptocurrencies – digital currencies that use cryptography to secure transactions – or machine learning algorithms, we may see other contradictory shifts in the impact our use of the Internet has on the environment.

Many see cryptocurrencies as a more secure and efficient way to conduct financial transactions. But most rely upon blockchain, a system that validates transactions by requiring users to dedicate enormous amounts of computing power through “Proof of Work” algorithms. Some estimate that cryptocurrencies that use Blockchain are responsible for 22 million tonnes of carbon dioxide emissions every year – about the same as the country of Jordan. It is hard to imagine how this can be made sustainable.

Similarly, as we turn to widespread artificial intelligence to find solutions to our problems, the algorithms we use are growing in the energy they need to run our services. A programmer writing software in the 1980s had far less computing power to work with, and so had to ensure their algorithms ran as efficiently as possible within hard limits. Today machine learning (PDF) is expanding to process ever larger datasets, using the growing capacity of the Internet and its cloud data centres.

Demand for machine learning is only likely to keep increasing as we use it more and more in our daily lives. And we are already taking the power of AI for granted – in some cases up to a third of tasks using machine learning are stopped before they are finished, meaning energy they used is simply wasted.

If, as some predict, the current pandemic crisis will lead to a permanent shift in the way we live and work that means we spend even more time online, improvements in energy efficiency and the lifespan of products will only take us so far. The number of people coming online is growing by around one million every day, while the myriad of uses for the Internet will increase just as fast.

The real questions we need to ask is: will there ever be a limit, and should we be imposing one?


  • Dr Andrew Rice

    Dr Andrew Rice

    Dr Andrew Rice is a Reader at the University of Cambridge’s Department of Computer Science and Technology in the Digital Technology Group and the Hassabis Fellow in Computer Science at Queen’s College. He is a visiting scientist at Google.
  • Professor Adrian Friday

    Professor Adrian Friday