In terms of energy conservation, the leaps made in energy efficiency by the infrastructure and devices we use to access the internet have allowed many online activities to be viewed as more sustainable than offline.
On the internet, however, advances in energy efficiency have a reverse effect: as the network becomes more energy efficient, its total energy use increases. This trend can only be stopped when we limit the demand for digital communication.
Although it’s a strategy that we apply elsewhere, for instance, by encouraging people to eat less meat, or to lower the thermostat of the heating system, limiting demand is controversial when applied to the internet, in part because few people make the connection between data and energy.
How much energy does the Internet consume?
How much energy does the internet consume? Due to the complexity of the network and its fast-changing nature, nobody really knows. Estimates for the internet’s total electricity use vary by an order of magnitude. One reason for the discrepancy between results is that many researchers only investigate a part of the infrastructure that we call the internet.
In recent years, the focus has been mostly on the energy use of data centers, which host the computers (the “servers”) that store all information online. However, in comparison, more electricity is used by the combination of end-use devices (the “clients”, such as desktops, laptops and smartphones), the network infrastructure (which transmits digital information between servers and clients), and the manufacturing process of servers, end-use devices, and networking devices. 1
A second factor that explains the large differences in results is timing. Because the internet infrastructure grows and evolves so fast, results concerning its energy use are only applicable to the year under study. Finally, as with all scientific studies, researcher’s models, methods and assumptions as a base for their calculations vary, and are sometimes biased due to beliefs or conflicts of interest. For example, it won’t suprise anyone that an investigation of the internet’s energy use by the American Coalition for Clean Coal Electricity sees much higher electricity consumption than a report written by the information and communication technology industry itself. 23
Eight Billion Pedallers to Power the Internet
Keeping all this in mind, we selected what seems to be the most recent, complete, honest and transparant report of the internet’s total footprint. It concludes that the global communications network consumed 1,815 TWh of electricity in 2012. 4 This corresponds to 8% of global electricity production in the same year (22,740 TWh). 56
If we were to try to power the (2012) internet with pedal-powered generators, each producing 70 watt of electric power, we would need 8.2 billion people pedalling in three shifts of eight hours for 365 days per year. (Electricity consumption of end-use devices is included in these numbers, so the pedallers can use their smartphones or laptops while on the job). Solar or wind power are not much of a solution, either: 1,815 TWh equals three times the electricity supplied by all wind and solar energy plants in 2012, worldwide. 7
These researchers estimate that by 2017, the electricity use of the internet will rise to between 2,547 TWh (expected growth scenario) and 3,422 TWh (worst case scenario). If the worst-case scenario materializes, internet-related energy use will almost double in just 5 years time. Note that further improvements in energy efficiency are already included in these results. Without advances in efficiency, the internet’s energy use would double every two years, following the increase in data traffic. 8
Increasing Energy Consumption per User
Importantly, the increasing energy consumption of the internet is not so much due to a growing amount of people using the network, as one would assume. Rather, it’s caused by a growing energy consumption per internet user. The network’s data traffic rises much faster than the number of internet users (45% versus 6-7% annually). 9 There’s two main reasons for this. The first is the evolution towards portable computing devices and wireless internet access. The second is the increasing bit rate of the accessed content, mainly caused by the digitalization of TV and the popularity of video streaming.
The increasing energy consumption of the internet is not so much due to a growing amount of people using the network, as one would assume. Rather, it’s caused by a growing energy consumption per internet user.
In recent years we have seen a trend towards portable alternatives for the desktop computer: first with the laptop, then the tablet and the smartphone. The latter is on its way to 100% adoption: in rich countries, 84% of the population now uses a smartphone. 94 These devices consume significantly less electricity than desktop computers, both during operation and manufacture, which has given them an aura of sustainability. However, they have other effects that more than off-set this advantage.
First of all, smartphones move much of the computational effort (and thus the energy use) from the end-device to the data center: the rapid adoption of smartphones is coupled with the equally rapid growth in cloud-based computer services, which allow users to overcome the memory capacity and processing power limitations of mobile devices. 4 10 Because the data that is to be processed, and the resulting outcome must be transmitted from the end-use device to the data center and back again, the energy use of the network infrastructure also increases.
High-Speed Wireless Internet
Robbing Peter to pay Paul can improve the total efficiency of some computational tasks and thus reduce total energy use, because servers in datacenters are managed more energy efficiently than our end-use devices. However, this advantage surely doesn’t hold for smartphones that connect wirelessly to the internet using 3G or 4G broadband. Energy use in the network is highly dependent on the local access technology: the “last mile” that connects the user to the backbone of the internet.
A wired connection (DSL, cable, fibre) is the most energy efficient method to access the network. Wireless access through WiFi increases the energy use, but only slightly. 1112 However, if wireless acces is made through a cellular network tower, energy use soars. Wireless traffic through 3G uses 15 times more energy than WiFi, while 4G consumes 23 times more. 13414 Desktop computers were (and are) usually connected to the internet via a wired link, but laptops, tablets and smartphones are wirelessly connected, either through WiFi or via a cellular network.
Growth in mobile data traffic has been somewhat restricted to WiFi “offloading”: users restrict data connectivity on the 3G interface due to significantly higher costs and lower network performance. 4 Instead, they connect to WiFi networks that have become increasingly available. With the advance of 4G networks, the speed advantage of WiFi disappears: 4G has comparable or improved network throughput compared to WiFi. 13 Most network operators are in the process of large-scale rollouts of 4G networks. The number of global 4G connections more than doubled from 200 million at the end of 2013 to 490 million at the end of 2014, and is forecast to reach 875 million by the end of 2015. 101516
More Time Online
The combination of portable computing devices and wireless internet access also increases the time we spend online. 10 This trend did not start with smartphones. Laptops were expected to lower the energy consumption of the internet, but they raised it because people took advantage of the laptop’s convenience and portability to be online far more often. “It was only with the laptop that the computer entered the living room”. 17
Smartphones are the next step in this evolution. They allow data to be consumed in many places in and outside the home, alongside more conventional computing. 18 For example, field research has revealed that smartphones are intensively used to fill ‘dead time’—small pockets of time not focused on one specific activity and often perceived as unproductive time: waiting, commuting, being bored, coffee breaks, or “social situations that are not stimulating enough”. Smartphones also have become to play an important bedtime role, being called upon last thing at night and first thing in the morning. 18
We are using our increasingly energy efficient devices for longer hours as we send more and more data over a worldwide infrastructure.
Noting these trends, it is clear that not every smartphone is a substitute for a laptop or desktop computer. Both are used alongside each other and even simultaneously. In conclusion, thanks to smartphones and wireless internet, we are now connected anywhere and anytime, using our increasingly energy efficient devices for longer hours as we send more and more data over a worldwide infrastructure. 1819
The result is more energy use, from the mobile devices themselves, and — much more important—in the datacenters and in the network infrastructure. Also, let’s not forget that calling someone using a smartphone costs more energy than callling someone using a dumbphone.
Increasing Bit Rates: Music & Video
A second key driver behind the growing energy consumption per internet user is the increasing bit rate of content. The internet started as a text-medium, but images, music and video have become just as important. Downloading a text page requires very little energy. To give an example, all the text on this blog, some 100 articles, can be packed into less than 9 megabytes (MB) of data. Compare this to a single high-resolution image, which easily gets to 3 MB, or a standard quality 8-minute YouTube video, which ticks off at 30 MB—three times the data required for all the words on this blog.
Because energy use rises with every bit of data, it matters a lot what we’re doing online. And as it turns out, we are increasingly using the network for content with high bit rates, especially video. In 2012, video traffic was 57% of all internet traffic (excluding video exchanged through P2P-networks). It’s expected to increase to 69% in 2017. 20
If video and wireless internet access are the key drivers behind the increasing energy use of the internet, then of course wireless video is the worst offender. And it’s exactly that share of traffic that’s growing the fastest. According to the latest Cisco Visual Networking Index, mobile video traffic will grow to 72% of total mobile data traffic in 2019: 10
“When device capabilities are combined with faster, higher bandwith, it leads to wide adoption of video applications that contribute to increased data traffic over the network. As mobile network connection speeds increase, the average bit rate of content accessed through the mobile network will increase. High-definition video will be more prevalent, and the proportion of streamed content, as compared to side-loaded content, is also expected to increase. The shift towards on-demand video will affect mobile networks as much as it will affect fixed networks”.
Power consumption is not only influenced by data rates but also by the type of service provided. For applications such as email, web browsing, and video and audio downloads, short delays are acceptable. However, for real-time services—videoconferencing, and audio and video streaming — delay cannot be tolerated. This requires a more performant network, and thus more energy use.
Does the Internet Save Energy?
The growing energy use of the internet is often explained away with the argument that the network saves more energy than it consumes. This is attributed to substitution effects in which online services replace other more energy-intensive activities. 12 Examples are videoconferencing, which is supposed to be an alternative for the airplane or the car, or the downloading or streaming of digital media, which is supposed to be an alternative for manufacturing and shipping DVDs, CDs, books, magazines or newspapers.
Some examples. A 2011 study concluded that “by replacing one in four plane trips with videoconferencing, we save about as much power as the entire internet consumes”, while a 2014 study found that “videoconferencing takes at most 7% of the energy of an in-person meeting”. 2122 Concerning digital media, a 2014 study concludes that shifting all DVD viewing to video streaming in the US would respresent a savings equivalent to the primary energy used to meet the electricity demand of nearly 200,000 US household per year. 23 A 2010 study found that streaming a movie consumed 30 to 78% of the energy of traditional DVD rental networks (where a DVD is sent over the mail to the customer who has to send it back later). 24
Because the estimates for the energy intensity of the internet vary by four orders of magnitude, it’s easy to engineer the end result you want.
There are some fundamental problems with these claims. First of all, the results are heavily influenced by how you calculate the energy use of the internet. If we look at the energy use per bit of data transported (the “energy intensity” of the internet), results vary from 0,00064 to 136 kilowatt-hour per Gigabyte (kWh/GB), a difference of four orders of magnitude. 1218. The researchers who made this observation conclude that “whether and to what extent it is more energy efficient to download a movie rather than buying a DVD, or more sustainable to meet via videoconferencing instead of travelling to a face-to-face meeting are questions that cannot be satisfyingly answered with such diverging estimates of the substitute’s impact.” 12
To make matters worse, researchers have to make a variety of additional assumptions that can have a major impact on the end result. If videoconferencing is compared to a plane trip, what’s the distance travelled? Is the plane full or not? In what year was it built? On the other hand, how long does the videoconference take? Does it happen over a wired or a wireless access network? Do you use a laptop or a high-end telepresence system? When you’re streaming music, do you listen to a song once or twenty times? If you buy a DVD, do you go to the store by car or by bike? How long is the trip? Do you only buy the DVD or do you also shop for other stuff?
Time and Distance
All these questions can be answered in such a way that you can engineer the end result you want. That’s why it’s better to focus on the mechanisms that favour the energy efficiency of online and offline services, what scientists call a “sensitivity analysis”. To be fair, most researchers perform such an analysis, but its results usually don’t make it into the introduction of the paper, let alone into the accompanying press release.
One important difference between online and offline services is the role of time. Online, energy use increases with the time of the activity. If you read two articles instead of one article on a digital news site, you consume more energy. But if you buy a newspaper, the energy use is independent of the number of articles you read. A newspaper could even be read by two people so that energy use per person is halved.
Next to time there is the factor of distance. Offline, the energy use increases with the distance, because transportation of a person or product makes up the largest part of total offline energy consumption. This is not the case with online activities, where distance has little or no effect on energy consumption.
A sensitivity analysis generates very different conclusions from the ones that are usually presented. For example: streaming a music album over the internet 27 times can use more energy than the manufacturing and transportation of its CD equivalent. 25 Or, reading a digital newspaper on a desktop PC uses more energy than reading a paper version from the moment the reading length exceeds one hour and a quarter, taking the view that the newspaper is read by one person. 26 Or, in the earlier mentioned study about the energy advantage of videoconferencing, reducing the international participant’s travel distance from 5,000 to 333 km makes travelling in person more energy efficient than videoconferencing when a high-end telepresence system is used. Similarly, if the online conference takes not 5 but 75 hours, it’s more energy efficient to fly 5,000 km. 22
The energy efficiency advantage of videoconferencing looks quite convincing, because 75-hour meetings are not very common. However, we still have to discuss what is the most important problem with studies that claim energy efficiency advantages for online services: they usually don’t take into account rebound effects. A rebound effect refers to the situation in which the positive effect of technologies with improved efficiency levels is offset by systematic factors or user behaviour. For example, new technologies rarely replace existing ones outright, but instead are used in conjunction with one another, thereby negating the proposed energy savings. 27
Not every videoconference call is a substitute for physical travel. It can also replace a phone call or an email, and in these cases energy use goes up, not down. 22 Likewise, not every streamed video or music album is a substitute for a physical DVD or CD. The convenience of streaming and the advance of portable end-use devices with wireless access leads to more video viewing and music listening hours 23, at the expense of other activities which could include reading, observing one’s environment, or engaging in a conversation.
A videoconference can also replace a phone call or an email, and in these cases energy use goes up, not down.
Because the network infrastructure of the internet is becoming more energy efficient every year—the energy use per bit of data transported continues to decrease—it’s often stated that online activities will become more energy efficient over time, compared to offline activities. 3 However, as we have seen, the bit rate of digital content online is also increasing.
This is not only due to the increasing popularity of video applications, but also because of the increasing bit rate of the videos themselves. Consequently, future efficiency improvements in the network infrastructure will bring higher quality movies and videoconferencing, not energy savings. According to several studies, bit rates increase faster than energy efficiency so that green gains of online alternatives are decreasing. 222324
Efficiency Drives Energy Use
The rebound effect is often presented as a controversial issue, something that may or may not exist. But at least when it comes to computing and the internet, it’s an ironclad law. The rebound effect manifests itself undoubtedly in the fact that the energy intensity of the internet (energy used per unit of information sent) is decreasing while total energy use of the internet is increasing.
It’s also obvious in the evolution of microprocessors. The electricity use in fabricating a microprocessor has fallen from 0.028 kWh per MHz in 1995 to 0.001 kWh per MHz in 2006 as a result of improvements in manufacturing processes. 28 However, this has not caused a corresponding reduction of energy use in microprocessors. Increased functionality—faster microprocessors—has cancelled out the efficiency gains per MHz. In fact, this rebound effect has become known as Moore’s Law, which drives progress in computing. 2728
In other words, while energy efficiency is almost universally presented as a solution for the growing energy use of the internet, it’s actually the cause of it. When computers were still based on vacuum tubes instead of transistors on a chip, the power used by one machine could be as high as 140 kilowatt. Today’s computers are at least a thousand times more energy efficient, but it’s precisely because of this improved energy efficiency that they are now on everybody’s desk and in everybody’s pocket. Meanwhile, the combined energy use of all these more energy-efficient machines outperforms the combined energy use of all vacuum tube computers by several orders of magnitude.
In conclusion, we see that the internet affects energy use on three levels. The primary level is the direct impact through the manufacturing, operation and disposal of all devices that make up the internet infrastructure: end-use devices, data centers, network and manufacturing. On a second level, there are indirect effects on energy use due to the internet’s power to change things, such as media consumption or physical travel, resulting in a decrease or increase of the energy use. On a third level, the internet shifts consumption patterns, brings technological and societal change, and contributes to economic growth. 2728 The higher system levels are vastly more important than the direct impacts, despite receiving very little attention. 28
“The internet entails a progressive globalisation of the economy that has thus far caused increasing transportation of material products and people… The induction effect arising from the globalisation of markets and distributed forms of production due to telecommunication networks clearly leads away from the path of sustainability… Finally, the information society also means acceleration of innovation processes, and thus ever faster devaluation of the existing by the new, whether hardware or software, technical products or human skills and knowledge.” 27
Nobody can deny that the internet can save energy in particular cases, but in general the overwhelming trend is towards ever-higher energy use. This trend will continue unabated if we don’t act. There’s no constraint on the bit rate of digital data. Blu-ray provides superior viewing experience, with data sizes ranging between 25 and 50 GB—five to ten times the size of a HD video. With viewers watching 3D movies at home, we can imagine future movie sizes of 150 GB, while holographic movies go towards 1,000 GB. 24
Nor is there any constraint on the bit rate of wireless internet connections. Engineers are already preparing the future launch of 5G, which will be faster than 4G but also use more energy. There’s not even a constraint on the number of internet connections. The concept of the “internet of things” foresees that in the future all devices could be connected to the internet, a trend that’s already happening. 410 And let’s not forget that for the moment only 40% of the global population has access to the internet.
In short, there are no limits to growth when it comes to the internet, except for the energy supply itself. This makes the internet rather unique. For example, while the rebound effect is also very obvious in cars, there are extra limits which impede their energy use from increasing unabated. Cars can’t get larger or heavier ad infinitum, as that would require a new road and parking infrastructure. And cars can’t increase their speed indefinitely, because we have imposed maximum speed limits for safety. The result is that the energy use of cars has more or less stabilized. You could argue that cars have achieved a status of “sufficiency”:
“A system consuming some inputs from its environment can either increase consumption whenever it has the opportunity to do so, or keep its consumption within certain limits. In the latter case, the system is said to be in a state of sufficiency… A sufficient system can improve its outputs only by improving the efficiency of its internal process.” 29
The performance of cars has only increased within the limits of the energy efficiency progress of combustion engines. A similar effect can be seen in mobile computing devices, which have reached a state of sufficiency with regard to electricity consumption—at least for the device itself. 29 In smartphones, energy use is limited by a combination of battery constraints: energy density of the battery, acceptable weight of the battery, and required battery life. The consequence is that the per-device energy use is more or less stable. The performance of smartphones has only increased within the limits of the energy efficiency progress of computing (and to some extent the energy density progress of batteries). 29
A Speed Limit for the Internet
In contrast, the internet has very low sufficiency. On the internet, size and speed are not impractical or dangerous. Batteries limit the energy use of mobile computing devices, but not the energy use of all the other components of the network. Consequently, the energy use of the internet can only stop growing when energy sources run out, unless we impose self-chosen limits, similar to those for cars or mobile computing devices. This may sound strange, but it’s a strategy we also apply quite easily to thermal comfort (lower the thermostat, dress better) or transportation (take the bike, not the car).
Limiting the demand for data could happen in many ways, some of which are more practical than others. We could outlaw the use of video and turn the internet back into a text and image medium. We could limit the speed of wireless internet connections. We could allocate a specific energy budget to the internet. Or, we could raise energy prices, which would simultaneously affect the offline alternatives and thus level the playing field. The latter strategy is preferable because it leaves it to the market to decide which applications and devices will survive.
Setting a limit would not stop technological progress. Advances in energy efficiency will continue to give room for new devices and applications to appear.
Although none of these options may sound attractive, it’s important to note that setting a limit would not stop technological progress. Advances in energy efficiency will continue to give room for new devices and applications to appear. However, innovation will need to happen within the limits of energy efficiency improvements, as is now the case with cars and mobile computing devices. In other words: energy efficiency can be an important part of the solution if it is combined with sufficiency.
Limiting demand would also imply that some online activities move back to the off-line world—streaming video is candidate number one. It’s quite easy to imagine offline alternatives that give similar advantages for much less energy use, such as public libraries with ample DVD collections. Combined with measures that reduce car traffic, so that people could go to the library using bikes or public transportation, such a service would be both convenient and efficient. Rather than replacing physical transportation by online services, we should fix the transport infrastructure.
In the next articles, we investigate the low-tech information networks that are being developed in poor countries. There, “sufficiency” is ingrained in society, most notably in the form of a non-existing or non-reliable energy infrastructure and limited purchasing power. We also discuss the community networks that have sprung up in remote regions of rich countries, and the designs for shared networks in cities. These alternative networks provide much more energy efficient alternatives for digital communication in exchange for a different use of the internet.
Kris De Decker (Edited by Jenna Collett)
Page size: 401.7 KB
Even the most complete studies about the internet’s energy use do not take into account all components of the infrastructure. For example, the embodied energy of the energy plants which are used to power the internet is completely ignored. However, if you run a data center or cellular tower on solar energy, it’s obvious that the energy it took to produce the solar panels should be included as well. The same goes for the batteries that store solar energy for use during the night or on cloudy days. ↩
“The cloud begins with coal: big data, big networks, big infrastructure, and big power” (PDF), Mark P. Mills, National Mining Association / American Coalition for Clean Coal Electricity, augustus 2013 ↩
“SMARTer2030—ICT Solutions for 21st Century Challenges” (PDF), Global e-Sustainability Initiative, 2015 ↩↩
Of the total, 852 TWh was consumed by end-use devices, 352 TWh by networks, 281 TWh by data centers, and 330 TWh during the manufacturing stage. ↩
The researchers also provide a “best case scenario” in which energy use increases only slightly. However, this scenario is already superseded by reality. It supposes slow growth of wireless data traffic and digital TVs, but the opposite has happened, as Cisco Visual Networking Index  shows. Furthermore, the best-case-scenario supposes a year-on-year improvement in energy efficiency of 5% for most device categories and an annual improvement in efficiency of the core network of 15%. These figures are well above those of past years and thus not very likely to materialize. The expected growth scenario supposes wireless traffic to grow to 9% of total network electricity consumption, and digital TV to stabilize at 2.1 billion units. In this scenario, energy efficiency improvements for devices are limited to 2% per year, while energy efficiency in the core network is limited to 10% per year. In the worst case scenario, wireless traffic grows to 15% of total network electricity consumption, digital TV will keep growing, and improvements in energy efficiency are limited to 1-5% annually for devices and to 5% in the core network. 4 ↩
“The energy intensity of the internet: home and access networks” (PDF), Vlad Coroama, 2014 ↩↩↩↩
“A close examination of performance and power characteristics of 4G LTE networks” (PDF), Junxian Huang, June 2012. ↩↩
“Energy consumption in mobile phones: a measurement study and implications for network applications” (PDF), Niranjan Balasubramanian, 2009 ↩
Network equipment manufacturer Cisco notes in its 2015 report that “as mobile network capacity improves and the number of multiple device users grow, operators are more likely to offer mobile broadband packages comparable in price and speed to those of fixed broadband.” 10 If this becomes true, and a majority of internet users would routinely connect to the internet through 4G broadband, the energy use of the network infrastructure would more than double, assuming data traffic would remain the same.  That’s because from an energy perspective, the access network is the greedy part of any service provider’s network. The core network of optic cables is much more energy efficient. 4 ↩
“Are we sitting comfortably? Domestic imaginaries, laptop practices, and energy use“. Justin Spinney, 2012 ↩
“Demand in my pocket: mobile devices and the data connectivity marshalled in support of everyday practice” (PDF), Carolynne Lord et al., Lancaster University, april 2015 ↩↩↩↩
“Towards a holistic view of the energy and environmental impacts of domestic media and IT“, Oliver Bates et al., 2014 ↩
“Cisco Visual Networking Index 2012-2017”, Cisco, 2013 ↩
“Screening environmental life cycle assessment of printed, web based and tablet e-paper newspaper“, Second Edition, Asa Moberg et al, 2009 ↩
“Environmental effects of informantion and communications technologies“, Eric Williams, Nature, 2011 ↩↩↩↩
“Computing Efficiency, Sufficiency, and Self-Sufficiency: A Model for Sustainability?” (PDF), Lorenz M. Hilty, 2015 ↩↩↩