How much energy does the Internet use?

At the moment I’m using the Internet, as are you.  In many ways, the Internet is the largest and perhaps most successful global system ever built by humanity.  And yet because of the way it was built—haphazardly, over the course of a few decades—there are no maps, no records documenting its entire structure.

Given my interest in energy, and understanding how society’s energy use is broken down, I wanted to understand how much energy the Internet uses.  Recently my colleague Justin and I did a study to understand just that.  There are many questions in addition to the obvious that we explored, including:

  • Is it worth it to keep the Internet running the way we do today given future energy constraints?
  • Is the Internet a large energy consumer vs. other functions of society?
  • What does the Internet even consist of?

A key part of understanding the energy use of the Internet is that its embodied energy, or emergy, is important to include.

Why measure emergy?

Emergy, a concept introduced by H.T. Odum is perhaps the most ignored aspect of energy use.  Loosely defined, it’s the energy used to make the things that we use or buy or consume.  We often ignore emergy since it doesn’t directly cost us anything, so it’s invisible.  Nevertheless, emergy matters when we want to understand the energy-use impact of something from a society-wide perspective.  As it turns out, the manufacturing of computer hardware is an extremely energy intensive process.  Thus there are two important pieces to look at: the energy use of the Internet (typically viewed as the wall-socket electricity use) and the emergy of the devices themselves.

An Internet census

Before we could even consider the energy use of the Internet, we had to understand how big the Internet is in the first place.  That, is, we needed to take a census.  How many desktop computers are there in the Internet?  How many laptops are there in the Internet?  How many cloud servers are there?  How many miles of fiber optics are there?  How many routers are there?  We were unable to find a single source that answered all of these questions.  So here I’d like to provide the answers to them, along with the sources of our data.  We rounded our numbers since we didn’t want to convey a false sense of precision, and a few values were guesstimates because no good data was available.  Also, since it’s ambiguous whether some of these components are part of the Internet or not, when we include them in our final calculation we weight them with a min (lower-bound) and max (upper-bound) weight to get a range of possible energy use.

Category Count Per-unit emergy
Desktops 750 million 7.5 GJ
Laptops 750 million 4.5 GJ
Cloud 50 million 5 GJ
Smartphones 1 billion 1 GJ
Servers 100 million 5 GJ
Routers 1 million 50 GJ
Wi-Fi/LAN 100 million 1 GJ
Cell Towers 5 million 100 GJ
Telecom Switches 1 million 50 GJ
Fiber Optics 1.5 billion km 10 GJ/km
Copper 3.5 billion km 10 GJ/km

I won’t go through the calculations, because they aren’t particularly exciting.  But the bottom line is that we calculated the wall-socket power consumption as well, and found a few interesting things:

  • The embodied power—the emergy of devices divided by their replacement lifespan—is roughly equivalent to their wall-socket power consumption.  That is, looking at wall-socket electricity ignores half of actual energy use for computing devices.  Specifically, we estimated that the total wall-socket power use of the Internet is between 83-144 GW and the embodied power is between 87-164 GW.
  • The total power use of the Internet, which we estimated to be between 170-307 GW, is about 1-2% of global energy use.  We were a bit surprised by this, because we thought it might be higher than that.  (Though it’s a larger percentage if we were to consider electricity use alone.)
  • It might be reasonable, then, to focus more attention on how the Internet can substitute for other energy-intensive societal functions.  We did a simple calculation for the obvious example—transportation.  Suppose we were to replace 1 out of every 4 air trips with a video conference (we assume business trips that consist of 5 one-hour meetings).  Doing so would decease global energy use by about 285 GW, which is on the order of the Internet’s entire energy consumption.  (And the potential increase in the Internet’s energy use for carrying extra video traffic would only be about 2 GW.)

The bottom line is that the few previous studies that existed were ignoring a large piece—roughly half—of the energy use in question.  Despite this, it seems worth it to keep the Internet going as long as we can because it may help us, for a time, keep our energy use down.  In future posts, I’d like to consider where else looking at emergy can yield insights and reveal shortsightedness.

Leave a Reply


Responses to “How much energy does the Internet use?”

  1. When you are discussing energy, you should use only energy units; and, since the devices being discussed are electrical, I believe it would make the most sense to use units of electrical energy. Since 1 GJ = 2.778 KW-hr, this should not be difficult. Then it would be easy to see how the emergy of the various devices adds up to about 170 million MW-hr (not MW, MW-hr), and it would be easy to compare this to world electrical consumption of 17 billion MW-hr per year.

  2. Hi Art, thanks for your reply! I use units of power (i.e. gigawatts) here for a specific reason: it makes it easy to compare the consumption of energy over different timescales. (Saul Griffith and WattzOn do so for the same reason.) That is, it allows the comparison of wall-socket power consumption (which is usually calculated in Watts or KWh) and embodied energy (usually calculated in GJ) and embodied power (GJ / year) and worldwide energy consumption (which the EIA computes in quadrillion BTUs / year, the IEA computes in million barrels of oil equivalent / year, etc.). Given all these different units, using watts makes things easiest for comparison. Globally we use between 15-18 TW.

  3. This article is relevant – it looks at carbon emissions from burning fossil fuels, rather than energy. But the result is similar to yours —

  4. Thanks for the link – it’s interesting to see that despite the different methodology they are in the same ballpark. Factors that I’m still mulling (and didn’t directly include in this calculation) include the indirect energy / emergy cost due to data center cooling, electricity generation inefficiency, etc. They’re hard to compute accurately, but might increase the topline number a bit.

  5. Would speakers/web cams be factored? Video gaming systems, cable boxes, television’s themselves are a couple others that could be in the grey area.

  6. Andrew -

    We didn’t factor them in explicitly in our initial study (instead we just considered them as misc small devices), but they could be considered. We didn’t consider any single device “part of” or “not part of” the Internet anyway—we used a fractional value as a weight—so it should be fine to consider a much wider range of devices with low weights.

  7. Hi Barath,

    Will you be writing a follow-up article to this one? I’d love to know how much these number have changed considering the lower use of desktop and higher use of mobile as well as the quickly rising number of cloud servers.

    Thank you!

  8. Candice – I do plan on updating this data, and have a project underway to make the analysis more detailed as well. It may be a few months though, but I will be sure to post on it once the analysis is done.

  9. There is one other significant energy cost of the Internet: cooling. All of those devices create heat which must be dealt with. Data centers in particular are heat-monsters, and running the air conditioners can draw nearly as much power as the devices being cooled. Therefore, as a wet-finger-in-the-wind factor, double the energy use for data centers (cloud, servers, switches, routers, telco) and see if that results in numbers that are more in line with what you expected.

    You may (or may not) also wish to include energy consumption for manufacturing the various devices, prorated over the expected lifetime of the device. I don’t have a good handle on how much power it takes to create and assemble all the components in the various devices at hand; it might be statistically insignificant, or it might not.

  10. Hi Lance — Cooling is definitely an issue, and in the calculations we did we added a 50% overhead for cooling for data centers. We also included the energy to manufacture the devices, which to our surprise turned out to be about 50% of the total power consumption of the Internet (i.e. the embodied power is about equal to the wall-socket power). (You can see more detailed calculations of all of this in “The Energy and Emergy of the Internet”.)