index.idyll

[meta title:"The Hidden Cost of Digital Consumption" description:"How much carbon dioxide is released when you load this article? Digital doesn’t mean green."
  shareImageUrl:"https://parametric.press/issue-02/streaming/static/images/share.png"
  shareImageWidth:"880"
  shareImageHeight:"440" /]

[var name:"parametricSlug" value:"streaming" /]

[Nav fullWidth:true /]

[Header
  title:`["The Hidden Cost of Digital Consumption"]`
  longTitle:`["The Hidden Cost of Digital Consumption"]`
  date:"October 19, 2020"
  dek:"Digital doesn’t mean green. How much carbon dioxide was released when you loaded this article?"
  fullWidth:true
  authors:`[{
    name: "Halden Lin",
    role: "",
    url: "https://haldenl.com"
  },
  {
    name: "Aishwarya Nirmal",
    role: "",
    url: "https://twitter.com/AishwaryaNirmal"
  },
  {
    name: "Shobhit Hathi",
    role: "",
    url: "https://twitter.com/ssshathiii"
  },
  {
    name: "Lilian Liang",
    role: "",
    url: "https://twitter.com/liang_lilian"
  }]`
  doi:"https://zenodo.org/badge/latestdoi/245054203"  
  archive:"https://parametric-press-archives.s3-us-west-2.amazonaws.com/issue-02.wacz"
  source:`"https://github.com/ParametricPress/02-" + parametricSlug `
/]

[MediaPreviews /]
// BEGIN ARTICLE

## A Paradigm Shift

In 2014, Google had to [link text:"shark-proof" url:"https://www.wired.com/2014/08/shark-cable/" target:"_blank"/] their underwater cables.
It turned out that sharks were fond of chewing on the high-speed cables that make up the internet.
While these "attacks" are no longer an issue, they are a reminder that both the internet and sharks share the same home.

[div className:"showLargeScreen"]
[LargeAside]
  [ParametricGraphic]

    [AmazonBook/]
    Book [span style:`{color: 'white'}`][b] →[/b][/span]
    [br/][br/]
    [AmazonEbook /]
    E-Book [span style:`{color: 'white'}`][b] →[/b][/span]
    [caption]It is difficult to ignore the environmental consequences of ordering a physical book on Amazon. But what happens when you download an E-Book instead?[/caption]

  [/ParametricGraphic]
[/LargeAside]
[/div]

When you buy a book on Amazon, the environmental consequences are difficult to ignore.
From the jet fuel and gasoline burned to transport the packages, to the cardboard boxes cluttering your house, the whole process is filled with reminders that your actions have tangible, permanent consequences.
Download that book onto an e-reader, however, and this trail of evidence seems to vanish.
The reality is, just as a package purchased on Amazon must be transported through warehouses and shipping centers to get to your home, a downloaded book must be transported from data centers and networks across the world to arrive on your screen.

[div className:"showSmallScreen"]
[LargeAside]
  [ParametricGraphic]

    [AmazonBook/]
    Book [span style:`{color: 'white'}`][b] →[/b][/span]
    [br/][br/]
    [AmazonEbook /]
    E-Book [span style:`{color: 'white'}`][b] →[/b][/span]
    [caption]It is difficult to ignore the environmental consequences of ordering a physical book on Amazon. But what happens when you download an E-Book instead?[/caption]

  [/ParametricGraphic]
[/LargeAside]
[/div]

And that's the thing: your digital actions are only possible because of the physical infrastructure built to support them.
Digital actions are physical ones. This means that Google has to be mindful of sharks.
This also means that we should be mindful of the carbon emissions produced by running the world's data centers and internet networks.
According to one study, [span className:"emphasized"]the Information and Communications Technology sector is expected to account for 3–3.6% of global greenhouse gas emissions in 2020[/span] [Cite
authors:"Lotfi Belkhir and Ahmed Elmelgi"
title:"Assessing ICT global emissions footprint: Trends to 2040 & recommendations"
url:"https://www.sciencedirect.com/science/article/abs/pii/S095965261733233X"
id:"ict-global-emissions"
venue:"Journal of Cleaner Production"
date:"March 2018"
/].
That's more than the fuel emissions for all air travel in 2019, which clocked in at [a href:"https://www.nytimes.com/2019/09/19/climate/air-travel-emissions.html" target:"_blank"]2.5%.[/a]
The answer to "why?" and  "how?" may not be immediately obvious, but that's not the fault of consumers.
A well-designed, frictionless digital experience means that users don't [i]need[/i] to worry about what happens behind the scenes and, by extension, the consequences.
This is problematic: the idea of "hidden costs" runs contrary to principles of environmental awareness.
Understanding how these digital products and services work is a crucial first step towards addressing their environmental impact.

[Aside]
  [Newsletter /]
[/Aside]

## A Look at Different Types of Media
[data name:"mediaPickerData" source:"dist/media-emissions.json"/]
Each type of digital activity produces different levels of emissions.
[span className:"emphasized"]  The amount of carbon dioxide emitted by a particular digital activity is a factor of the quantity of information that needs to be loaded[/span]. More specifically, we can estimate emissions using the following formula:

[div style:`{textAlign: 'center'}`][span className:"emphasized"][Equation display:false]n [/Equation][/span] bytes [Equation display:false] \times [/Equation][span className:"emphasized"][Equation display:false] X [/Equation][/span] KwH/byte  [Equation display:false]\times[/Equation][span className:"emphasized"][Equation display:false] Y [/Equation][/span] g CO₂/KwH[/div][br/]

A [span className:"terms"]byte[/span] is a unit for information.

[span className:"emphasized"][Equation display:false]X = 6 \times 10^{-11} [/Equation][/span] is the global average for energy consumed for transmitting one byte of data in 2015, as calculated by Aslan et al. (2017) [Cite
authors:"Joshua Aslan, Kieren Mayers, Jonathan G. Koomey, and Chris France"
title:"Electricity Intensity of Internet Data Transmission: Untangling the Estimates"
url:"https://onlinelibrary.wiley.com/doi/full/10.1111/jiec.12630"
id:"data-transmission-estimates"
venue:"Journal of Industrial Ecology"
date:"August 2017"
/].

[span className:"emphasized"][Equation display:false] Y = 707 [/Equation][/span] represents the [a href:"https://www.epa.gov/energy/greenhouse-gases-equivalencies-calculator-calculations-and-references" target:"_blank"]EPA's[/a] U.S. national weighted average for grams of CO₂ emitted for electricity consumed.

Ideally, this formula would also include the energy usage of the data source and your device, but these will vary across digital media providers, users, and activities. This formula therefore provides a reasonable lower bound for emissions.

### Websites
[var name:"emissionGraphicPadding" value:`{paddingTop: 0, paddingLeft: 0, paddingRight: 0}`/]
[var name:"emissionSource" value:`["Aslan et al. 2017", {label: "[2]", url:"https://onlinelibrary.wiley.com/doi/full/10.1111/jiec.12630"}, ", ", {label:"EPA", url:"https://www.epa.gov/greenvehicles/greenhouse-gas-emissions-typical-passenger-vehicle"}]`/]
[var name:"emissionSourceMinusEpa" value:`["Aslan et al. 2017", {label: "[2]", url:"https://onlinelibrary.wiley.com/doi/full/10.1111/jiec.12630"}]`/]

[div className:"showLargeScreen"]
[LargeAside]
  [ParametricGraphic
    hed:"Emissions of Websites" style:emissionGraphicPadding
    source:emissionSource
  ]
    [MediaPicker type:"bar" mediaType:"website" data:mediaPickerData width:"100%" headers:false/]
    [caption]Each bar represents the carbon emitted when scrolling through a website for 60 seconds.  Car distance equivalent is calculated using the fuel economy of an average car. [span className:"desktopInstructions"]Click[/span][span className:"touchscreenInstructions"]Tap[/span] each bar to show a preview clip of the scroll.[/caption]
  [/ParametricGraphic]
[/LargeAside]
[/div]

In order to load the _Parametric Press_ article that you are currently reading, we estimate that 51 milligrams of CO₂ were produced. The same amount of CO₂ would be produced by driving a car 0.20 meters (based on the fuel economy of an average car, [link text:"according to the EPA" url:"https://www.epa.gov/greenvehicles/greenhouse-gas-emissions-typical-passenger-vehicle" target:"_blank"/]).
[span className:"emphasized"]  These emissions are a result of loading data for the text, graphics, and visualizations[/span] that are then rendered on your device.

The chart [span className:"showLargeScreen"]to the right[/span][span className:"showSmallScreen"]below[/span] displays the carbon emitted when loading various websites and scrolling through each at a constant speed for 60 seconds (this scrolling may incur more loading, depending on the site). All data collection was done using a Chrome web browser under a 100mbps Wi-Fi connection.
[span className:"desktopInstructions"] Clicking[/span][span className:"touchscreenInstructions"] Tapping[/span] each bar in the chart will show a preview of that scroll, sped up to a 5 second clip.

[div className:"footnote"]
 Note: Subsequent visits to websites may produce fewer emissions, as parts of the site may be saved in something called a [link text:"cache" url:"https://www.bigcommerce.com/ecommerce-answers/what-browser-cache-and-why-it-important/" target:"_blank"/].
[/div]

[div className:"showSmallScreen"]
[LargeAside]
[ParametricGraphic hed:"Emissions of Websites" style:emissionGraphicPadding source:emissionSource]
[MediaPicker type:"bar" mediaType:"website" data:mediaPickerData width:"100%" headers:false/]
[caption]Each bar represents the carbon emitted when scrolling through a website for 60 seconds. Car distance equivalent is calculated using the fuel economy of an average car, according to the EPA. [span className:"desktopInstructions"]Click[/span][span className:"touchscreenInstructions"]Tap[/span] a bar to show a preview clip of the scroll. [/caption]
[/ParametricGraphic]
[/LargeAside]
[/div]

As shown [span className:"showLargeScreen"]to the right[/span][span className:"showSmallScreen"]above[/span], loading websites like Google, which primarily show text, produces much lower emissions than loading websites like Facebook, which load many photos and videos to your device.

### Audio
[div className:"showLargeScreen"]
[LargeAside]
[ParametricGraphic hed:"Emissions of Audio" style:emissionGraphicPadding source:emissionSource]
[MediaPicker type:"bar" mediaType:"audio" data:mediaPickerData width:"100%" headers:false/]
[caption]Each bar represents the carbon emitted when listening to an audio clip for 60 seconds (no audio preview). Note: The amount of data loaded may represent more than a minute's worth due to [link text:"buffering" url:"https://www.satelliteinternet.com/resources/how-to-stop-buffering/" target:"_blank" /].[/caption]
[/ParametricGraphic]
[/LargeAside]
[/div]

Let’s take a closer look at one common type of non-text media: audio.
When you listen to audio on your device, you generally load a version of the original audio file that has been compressed into a smaller size. 
In practice, this size is often determined by the choice of [span className:"terms"]bitrate[/span], which refers to the average amount of information in a unit of time.

[div className:"showSmallScreen"]
[LargeAside]
[ParametricGraphic hed:"Emissions of Audio" style:emissionGraphicPadding source:emissionSource]
[MediaPicker type:"bar" mediaType:"audio" data:mediaPickerData width:"100%" headers:false/]
[caption]Each bar represents the carbon emitted when listening to an audio clip for 60 seconds (no audio preview). Note: The amount of data loaded may represent more than a minute's worth due to [link text:"buffering" url:"https://www.satelliteinternet.com/resources/how-to-stop-buffering/" target:"_blank" /].[/caption]
[/ParametricGraphic]
[/LargeAside]
[/div]

The _NPR_ podcast shown in the visualization was compressed to a bitrate of 128 kilobits per second (there are 8 bits in a byte), while the song "Old Town Road", retrieved from Spotify, was compressed to 256 kilobits per second.
This means that in the one minute timespan that both audio files were played, roughly twice as much data needed to be loaded for "Old Town Road" than for "Digging into 'American Dirt'", which leads to the song having about twice as large of a carbon footprint.
The fact that the song has greater carbon emissions is not a reflection on the carbon footprint of songs versus podcasts, but rather the difference in the bitrate of each audio file.
These audio examples have lower carbon emissions than most of the multimedia websites shown earlier.

### Video
[div className:"showLargeScreen"]
[LargeAside]
[ParametricGraphic hed:"Emissions of Video" style:emissionGraphicPadding source:emissionSource]
[MediaPicker type:"bar" mediaType:"video" data:mediaPickerData width:"100%" headers:false/]
[caption]Each bar represents the carbon emitted when watching a video for 60 seconds at two different qualities, 360p and 1080p (to reduce the size of this article, there is only one 360p preview for both qualities). Note: like audio, videos are buffered, which means that playing the video may have loaded more than 60 seconds of content.[/caption]
[/ParametricGraphic]
[/LargeAside]
[/div]

Videos are a particularly heavy digital medium.
The chart [span className:"showLargeScreen"]to the right[/span][span className:"showSmallScreen"]below[/span] shows emissions for streaming different YouTube videos at two different qualities—360p and 1080p, for 60 seconds each.

[div className:"showSmallScreen"]
[LargeAside]
[ParametricGraphic hed:"Emissions of Video" style:emissionGraphicPadding source:emissionSource]
[MediaPicker type:"bar" mediaType:"video" data:mediaPickerData width:"100%" headers:false/]
[caption]Each bar represents the carbon emitted when watching a video for 60 seconds at two different qualities, 360p and 1080p (to reduce the size of this article, there is only one 360p preview for both qualities). Note: like audio, videos are buffered, which means that playing the video may have loaded more than 60 seconds of content.[/caption]
[/ParametricGraphic]
[/LargeAside]
[/div]

When you view a video at a higher quality, you receive a clearer image on your device because the video that you load contains more [span className:"terms"]pixels[/span]. Pixels are units of visual information.
In the chart, the number in front of the "p" for quality refers to the height, in pixels, of the video.
This is why there are greater emissions for videos at 1080p than those at 360p: more pixels means more data loaded per frame.

Bitrate also [link text:"plays a role in video streaming" url:"https://www.bbc.co.uk/bitesize/guides/z7vc7ty/revision/6" target:"_blank"/].
Here, bitrate refers to the amount of visual and audio data loaded to your device over some timespan.
From the chart, it is clear that the "Old Town Road" music video has a higher bitrate than the 3Blue1Brown animation at both qualities. 
This difference could be attributed to a variety of factors, such as the frame rate, compression algorithm, and the producers' desired video fidelity.

In the examples provided, videos produced far more CO₂ than audio over the same time span.
This is especially apparent when comparing the emissions for the audio of "Old Town Road" and its corresponding music video.

[MediaPicker type:"bar"
  mediaTitle:`[
    "Lil Nas X: Old Town Road (song)",
    "Lil Nas X: Old Town Road (music video)"
  ]`
  data:mediaPickerData width:"100%" headers:false inline:true/]

[div className:"showLargeScreen"]
[Aside]
Ads are another common form of digital content.
Digital media providers often include advertisements as a method of generating revenue.
When the European version of the USA Today website removed ads and tracking scripts to comply with GDPR (General Data Protection Regulation), [link text:"the size of its website decreased from 5000 to 500 kilobytes" url:"https://newrepublic.com/article/155993/can-internet-survive-climate-change" target:"_blank"/] with no significant changes to the appearance of the site.
This means that the [span className:"emphasized"]updated, ad-free version of the website produced roughly 10 times less CO₂ on each load than the original[/span]. 
[/Aside]
[/div]

Not only does a video require loading both audio and visual data, but [span className:"emphasized"]visual data is also particularly heavy in information[/span].
Notice that loading the website Facebook produced the most emissions, likely a result of loading multiple videos and other heavy data.

[MediaPicker type:"bar"
  mediaTitle:`[
    "Facebook (newsfeed)",
  ]`
  data:mediaPickerData width:"100%" headers:false inline:true/]

[div className:"showSmallScreen"]
[Aside]
Ads are another common form of digital content.
Digital media providers often include advertisements as a method of generating revenue.
When the European version of the USA Today website removed ads and tracking scripts to comply with GDPR (General Data Protection Regulation), [link text:"the size of its website decreased from 5000 to 500 kilobytes" url:"https://newrepublic.com/article/155993/can-internet-survive-climate-change" target:"_blank"/] with no significant changes to the appearance of the site.
This means that the [span className:"emphasized"]updated, ad-free version of the website produced roughly 10 times less CO₂ on each load than the original[/span]. 
[/Aside]
[/div]

## Packets
[var name:"mediaPickerType" value:"bar"/]

[div className:"showLargeScreen"]
[LargeAside]
[ParametricGraphic hed:"Media Emissions by Medium" style:emissionGraphicPadding source:emissionSourceMinusEpa]
[MediaPicker type:mediaPickerType data:mediaPickerData width:"100%" headers:true shouldPreload:true/]
[caption][span className:"desktopInstructions"] Click[/span][span className:"touchscreenInstructions"] Tap[/span] on the "Show Timeline" button to see a timeline of loading packets for each type of media. [span className:"desktopInstructions"]Hover[/span][span className:"touchscreenInstructions"]Tap and drag[/span] to scrub through and view cumulative emissions.[/caption]
[Conditional if:`mediaPickerType === 'bar'`][button onClick:`mediaPickerType = 'timeline'`]Show Timeline[/button][/Conditional]
[Conditional if:`mediaPickerType === 'timeline'`][button onClick:`mediaPickerType = 'bar'`]Hide Timeline[/button][/Conditional]
[/ParametricGraphic]
[/LargeAside]
[/div]

In a lot of cases, when you stream content online, [span className:"emphasized"] you don’t receive all of the information for that content at once[/span].
Instead, your device loads incremental pieces of the data as you consume the media.
These pieces are called [span className:"terms"]packets[/span].
In each media emission visualization, we estimated emissions based on the size and quantity of the packets needed to load each type of media.

[span className:"desktopInstructions"] Click[/span][span className:"touchscreenInstructions"] Tap[/span] the "Show Timeline" button at the bottom of the visualization [span className:"showLargeScreen"]to the right[/span][span className:"showSmallScreen"]below[/span] to see a timeline breakdown of each type of media.

[div className:"showSmallScreen"]
[LargeAside]
[ParametricGraphic hed:"Media Emissions by Medium" style:emissionGraphicPadding source:emissionSourceMinusEpa]
[MediaPicker type:mediaPickerType data:mediaPickerData width:"100%" headers:true shouldPreload:true/]
[caption][span className:"desktopInstructions"] Click[/span][span className:"touchscreenInstructions"] Tap[/span] on the "Show Timeline" button to see a timeline of loading packets for each type of media. [span className:"desktopInstructions"]Hover[/span][span className:"touchscreenInstructions"]Tap and drag[/span] to scrub through and view cumulative emissions.[/caption]
[Conditional if:`mediaPickerType === 'bar'`][button onClick:`mediaPickerType = 'timeline'`]Show Timeline[/button][/Conditional]
[Conditional if:`mediaPickerType === 'timeline'`][button onClick:`mediaPickerType = 'bar'`]Hide Timeline[/button][/Conditional]
[/ParametricGraphic]
[/LargeAside]
[/div]

In this timeline breakdown, we can see that the way in which packets arrive for video and audio differs from the pattern for websites.
When playing video and audio, packets tend to travel to your device at regular intervals.

[MediaPicker type:"timeline"
  mediaTitle:`[
    "Old Town Road (song)",
    "Old Town Road (music video)"
  ]`
  noAutoplayTimeline:true
  data:mediaPickerData width:"100%" headers:false inline:true/]

In contrast, the packets for websites are weighted more heavily towards the beginning of the timeline, but websites may make more requests for data as you scroll through and load more content.

[MediaPicker type:"timeline"
  mediaTitle:`[
    "Google (search result)",
    "Amazon (product page)"
  ]`
  noAutoplayTimeline:true
  data:mediaPickerData width:"100%" headers:false inline:true/]

## A Case Study: YouTube's Carbon Emissions
// Variables
[var name:"scrollerIndex" value:0 /]
[var name:"stepProgress" value:0/]
[var name:"scrollerStages" value:`['worldmap', 'worldmap2', 'datacenter', 'pop', 'cdn', 'datacentersall', 'spacer1', 'spacer2', 'internet', 'spacer3', 'residential', 'cellular', 'device', 'all', 'all2', 'all3', 'beforesimple', 'simple', 'compare', 'final', 'none', 'none2']`/]
[var name:"showPops" value:false /]
[var name:"showGgcs" value:false /]

We’ve just seen how digital streams are made up of packets of data sent over the internet. These packets aren’t delivered by magic.
[span className:"emphasized"] Every digital streaming platform relies on a system of computers and cables, each part consuming electricity and releasing carbon emissions.[/span]
When we understand how these platforms deliver their content, we can directly link our digital actions to the physical infrastructure that releases carbon dioxide.

Let's take a look at YouTube.
With its 2 billion + users, YouTube is the prototypical digital streaming service.
How does one of its videos arrive on your screen?

[CustomScroller currentStep:scrollerIndex progress:stepProgress offset:0.65]
[Graphic]
  [div id:"pipeline"]
    [Pipeline stage:`scrollerStages[scrollerIndex]` progress:`stepProgress * 100` showPops:showPops showGgcs:showGgcs /]
  [/div]
[/Graphic]

[Step]
Videos are stored on servers called [span className:"terms"] "data centers"[/span]: warehouses full of giant computers designed for storing and distributing videos.
For global services, these are often placed around the world.
[/Step]
[Step]
YouTube's parent company, Google, has [link text:"21 origin data centers" url:"https://www.google.com/about/datacenters/locations/" target:"_blank"/] strategically placed throughout 4 continents (North America, South America, Europe, and Asia).
[/Step]
[Step]
Let's take a closer look at one of these origin data centers. This one is in The Dalles, Oregon, on the West Coast of the United States.
[/Step]
[Step]
For information to get from this data center to you, it first goes through Google's own specialized data network to what they call [link text:"Edge Points of Presence" url:"https://peering.google.com/#/infrastructure" target:"_blank" /] (POPs for short), which bring data closer to high traffic areas.
There are three metro areas with POPs in this region: Seattle, San Francisco, and San Jose.
[Button onClick:`showPops = !showPops` style:`{opacity: scrollerStages[scrollerIndex] === 'pop' ? 1 : 0}`]
[Conditional if:`!showPops`]
Show worldwide POPs
[/Conditional]
[Conditional if:`showPops`]
Hide Globe
[/Conditional]
[/Button]
[/Step]
[Step]
From these POPs, data is routed through smaller data centers that form the "Google Global Cache" (GGC).
These data centers are responsible for [link text:"storing the more popular or recently watched videos" url:"https://peering.google.com/#/infrastructure" target:"_blank"/] for users in a given area, ensuring no single data center is overwhelmed and service stays zippy. There are 22 in the region shown on the map. 
A more general term for this collection of smaller data centers is a [span className:"terms"]Content Delivery Network[/span] (CDN for short). 

[Button onClick:`showGgcs = !showGgcs` style:`{opacity: scrollerStages[scrollerIndex] === "cdn" ? 1 : 0}`]
[Conditional if:`!showGgcs`]
Show entire GGC
[/Conditional]
[Conditional if:`showGgcs`]
Hide Globe
[/Conditional]
[/Button]
[/Step]
[Step]
In 2018, researchers from the University of Bristol used publically available data to estimate the energy consumption of each step of YouTube's pipeline in 2016.

Google does not disclose its data center energy consumption for YouTube traffic specifically. Therefore, Chris Preist, Daniel Schien and Paul Shabajee used the energy consumption numbers released for a similar service's (Netflix) data centers to estimate YouTube's data center energy consumption. They found that all data centers accounted for [span className:"emphasized"] less than 2% of YouTube's electricity use[/span] in 2016 [Cite
authors:"Chris Preist, Daniel Schien, and Paul Shabajee"
title:"Evaluating Sustainable Interaction Design of Digital Services: The Case of YouTube"
url:"https://dl.acm.org/doi/10.1145/3290605.3300627"
id:"youtube-paper"
venue:"CHI"
date:"2019"
/].
[/Step]
[Step]
Google doesn't have their own network for communication between POPs and the Google Global Cache.
For that, they use the internet.
[/Step]
[Step]
The internet is a global “highway of information” that allows packets of data to be transmitted as electrical impulses. 
[link text:"A packet is routed" url:"https://medium.com/@User3141592/how-does-the-internet-work-edc2e22e7eb8" target:"_blank"/] from a source computer, through cables and intermediary computers, before arriving at its destination.
In addition to the [span className:"emphasized"][link text:"550,000 miles of underwater cables" url:"https://www.businessinsider.com/how-internet-works-infrastructure-photos-2018-5#if-the-worlds-underwater-cables-were-laid-out-end-to-end-the-cables-could-extend-from-here-to-the-moon-and-back-again-and-then-wrap-around-the-earths-widest-point-almost-three-times-5" target:"_blank"/][/span] that form the backbone of the internet, regions have their own land-based networks.
[/Step]
[Step]
Here's (roughly) what the major internet lines of the west coast look like [Cite
  authors:"Ramakrishnan Durairajan, Paul Barford, Joel Sommers, and Walter Willinger"
  title:"InterTubes: A Study of the US Long-haul Fiber-optic Infrastructure"
  url:"https://dl.acm.org/doi/abs/10.1145/2785956.2787499"
  id:"internet-paper"
  venue:"SIGCOMM '15"
  date:"2015"
/].
Perhaps not surprisingly, it resembles our interstate highway system.
Preist et al. estimate that this infrastructure consumed approximately [span className:"emphasized"]1,900 Gigawatt-hours of electricity to serve YouTube videos in 2016 [Cite
id:"youtube-paper"
/], enough to power 170,000 homes in the United States for a year, [link text:"according to the EIA" url:"https://www.eia.gov/tools/faqs/faq.php?id=97&t=3" target:"_blank"/][/span].
[/Step]
[Step]
The packets traveling across this information highway need “off-ramps” to reach your screen.
The off-ramps that packets take are either [span className:"terms"]"fixed line" residential networks[/span] (wired connections from homes to the internet) or [span className:"terms"]cellular networks[/span] (wireless connections from cell phones to the internet).
The physical infrastructure making up these two types of networks differ, and therefore have distinct profiles of energy consumption and carbon emissions.
[/Step]
[Step]
An estimated 88% of YouTube's traffic went through fixed line networks (from your residential Cable, DSL, or Fiber-Optic providers), and this accounted for approximately [span className:"emphasized"]4,400 Gigawatt-hours of electricity usage [Cite
id:"youtube-paper"
/]—enough to power over 400,000 U.S. homes[/span].
[/Step]
[Step]
In comparison, only 12% of YouTube’s traffic went through cellular networks, but they were by far the most expensive part of YouTube’s content delivery pipeline, accounting for approximately [span className:"emphasized"]8,500 Gigawatt-hours of electricity usage—enough to power over 750,000 U.S. homes[/span] [Cite
id:"youtube-paper"
/].
At over 10 times the electricity usage per unit of traffic, the relative inefficiency of cellular transmission is clear.
[/Step]
[Step]
Eventually, the video data reaches your device for viewing.
While your device might not technically be part of YouTube’s content delivery pipeline, we can't overlook the cost of moving those pixels.
Devices accounted for an estimated [span className:"emphasized"]6,100 Gigawatt-hours of electricity usage[/span] [Cite
id:"youtube-paper"
/]:
that's over half a million U.S. homes worth of electricity.
[/Step]
[Step]
In total, Preist et al.'s research estimated that [span className:"emphasized"]YouTube traffic consumed 19.6 Tetrawatt-hours of electricity in 2016[/span] [Cite
id:"youtube-paper"
/].
Using the world emissions factor for electricity generation as reported by the International Energy Agency, they place the resulting [span className:"emphasized"]carbon emissions at 10.2 million metric tons of CO₂[/span] (offset to 10.1 after Google's renewable energy purchases for its data center activities).
[/Step]
[Step]
YouTube emitted nearly as much CO₂ as a metropolitan area like [link text: "Auckland, New Zealand" url:"https://knowledgeauckland.org.nz/media/1057/tr2019-002-aucklands-greenhouse-gas-inventory-to-2016.pdf"/] did in 2016.
Put in other words, 10.2 MtCO₂ is equivalent to the yearly footprint of approximately [link text:"2.2 million cars in the United States." url:"https://www.epa.gov/greenvehicles/greenhouse-gas-emissions-typical-passenger-vehicle" target:"_blank"/]
[/Step]
[Step]
YouTube's monthly active user count has since increased by a minimum of 33% since 2016, (from [link text:"1.5 billion a year later in 2017" url:"https://techcrunch.com/2017/06/22/youtube-has-1-5-billion-logged-in-monthly-users-watching-a-ton-of-mobile-video/" target:"_blank"/] to over [link text:"2 billion last year in 2019" url:"https://variety.com/2019/digital/news/youtube-2-billion-users-tv-screen-watch-time-hours-1203204267/" target:"_blank"/]). This means its current CO₂ emissions in 2020 could be even higher than Preist et al.’s 2016 prediction.
[/Step]
[Step]
If we assume that the emissions factor for electricity usage is similar for each part of the pipeline, we can get a rough idea of the carbon footprint profile of YouTube. However, it's important to note that this [span className:"emphasized"]breakdown isn't necessarily representative of the entirety of the information and technology sector[/span].
A 2018 study by McMaster University researchers Lotfi Belkhir and Ahmed Elmeligi paints a surprisingly different picture for the sector as a whole [Cite
id:"ict-global-emissions"/]. 
[/Step]
[Step]
To compare the two studies, we can group the "Internet", "Residential Network", and "Cellular Network" sections into an umbrella "Networks".

[div className:"footnote"]* For data center emissions: Preist et al. note that Google purchases renewable energy to match its data center electricity consumption.
[/div]

[div className:"footnote"]
† For device emissions: Preist et al. do not account for the energy required to manufacture devices.
[/div]
[/Step]
[Step]
Belkhir and Elmeligi provide emission estimates for both 2010 (retrospective) and 2020 (prospective).
Most surprising is the weight Data Centers and CDNs have in this breakdown.
We can speculate that the relatively high bandwidth required to transfer videos as a medium contributes at least partially to the overweightedness of the "Networks" category for YouTube.

[div className:"footnote"]Note that, unlike Preist et al, Belkhir and Elmeligi **do** account for the emissions generated from the production of devices.[/div]
[/Step]
[Step]
In the same study, Belkhir and Elmeligi created two models to project the ICT sector's emissions decades forward.
Even their "unrealistically conservative" linear model put tech at a [span className:"emphasized"]6–7% share of GHG emissions in 2040[/span], and their exponential model had tech reaching over 14%.
[/Step]
[Step]
What is being done about this? Aside from increasing the efficiency of each part of the pipeline and taking advantage of renewable energy, mindful design could also go a long way.
[/Step]
[Step]
In the context of digital streaming, Preist et al. point out that much of YouTube's traffic comes from music videos, and a good number of those "views" are likely just "listens".
If this "listen" to "view" ratio were even just 10%, [span className:"emphasized"]YouTube could have reduced its carbon footprint by about 117 thousand tons of CO₂ in 2016[/span], just by intelligently sending audio when no video is required.
That's over [link text:"2.2 million gallons of gasoline worth of CO₂ in savings." url:"https://www.eia.gov/environment/emissions/CO₂_vol_mass.php" target:"_blank"/]
[/Step]
[/CustomScroller]

## Tech's Inconvenient Truths

Digital streaming is not the only instance where environmentally harmful aspects of technology are outside of public consciousness. 
Tech is rarely perceived as environmentally toxic, but here's a surprising fact: [span className:"emphasized"]Santa Clara County, the heart of "Silicon Valley", has the most EPA classified "superfund" (highly polluted) sites in the nation[/span].
These [link text:"23 locations" url:"https://www.theatlantic.com/technology/archive/2019/09/silicon-valley-full-superfund-sites/598531/#:~:text=Santa%20Clara%20County%20has%2023,these%20chemicals%20may%20be%20impossible." target:"_blank"/] may be impossible to fully clean. Silicon Valley is primarily to blame.

[var name:"isTouchScreen" value:false /]
[TouchscreenDetector isTouchScreen:isTouchScreen /]

[derived name:"superfundCaption" value:`"The 23 superfund sites of Santa Clara County. " +
  (isTouchScreen ? 'Press the "Enable Touch" button, then tap each site for more information.' :
    'Hover over each one for more information.') +
  ' Larger Hazard Ranking System Scores are worse.'`/]

[data name:"superfundSitesGeoJSON" source:"dist/superfund-sites.json"/]
[var name:"zoomEnabled" value:false/]
[div className:"superfund-map" fullWidth:true]
[ParametricGraphic hed:"The Superfund Sites of Santa Clara Valley" subhed:superfundCaption source:`[{label:"EPA", url:"https://www.epa.gov/superfund/search-superfund-sites-where-you-live"}]`]
[SuperfundSites map:"homes" zoomEnabled:zoomEnabled geoJSON:superfundSitesGeoJSON /]
[/ParametricGraphic]
[/div]

In their book [i]The Silicon Valley of Dreams[/i], David Pellow & Lisa Park detail how Santa Clara County became “Silicon Valley” in the 1960s when companies in the area found ways to embed computer circuits into small wafers of silicon. 
[span className:"emphasized"]Manufacturing these chips is inherently toxic[/span], and its chemical runoff has been linked to increases in cancer and birth defect rates around the Bay.
However, [span className:"emphasized"]the industry avoided responsibility for decades[/span] and actually cultivated a clean image, publicizing their pristine campuses and white-collar workforce [Cite
authors:"David N. Pellow and Lisa Sun-Hee Park"
title:"The Silicon Valley of Dreams: Environmental Injustice, Immigrant Workers, and the High-Tech Global Economy"
id:"silicon-valley-book"
url:"https://nyupress.org/9780814767092/the-silicon-valley-of-dreams/"
date:"December 2002"
/].

[Aside]
  [Recirc slug:parametricSlug /]
[/Aside]

Change in Silicon Valley never came easily.
In the 1980s, residents around the Bay took matters into their own hands, mapping out patterns of illnesses around factories and formed environmental coalitions. 
[span className:"emphasized"]It wasn't until this public outcry that one of the largest Silicon Valley manufacturers closed down a particularly problematic factory[/span] [Cite id:"silicon-valley-book"/].
For the first time, a tech company admitted responsibility for manufacturing-related illnesses, blowing a hole in the industry’s clean facade.
In the decades that followed, factories began shutting down as manufacturing moved out of Silicon Valley, but their environmental impact is still being felt today, primarily by working-class communities of color.
A study done in 2014 by researchers from Santa Clara University found that “social vulnerability, cumulative environmental hazards, and environmental benefits exhibit distinct spatial patterns in \[Santa Clara County\]” [Cite
authors:"Iris T.Stewart, Christopher M.Bacon, and William D.Burke"
title:"The uneven distribution of environmental burdens and benefits in Silicon Valley's backyard"
url:"https://www.sciencedirect.com/science/article/abs/pii/S0143622814002264"
venue:"Applied Geography"
id:"spatial-inequity"
date:"December 2014"
/].

[span className:"emphasized"]The tech industry’s legacy of toxic mining and manufacturing [link text:"continues today" url:"https://increment.com/energy-environment/a-rare-and-toxic-age/" target:"_blank"/][/span]—with hardware giants like Apple taking the throne of Silicon Valley's founding industrialists—only now on a global scale.
In 2009, [link text:"China produced 95% of the world’s “rare earth” minerals" url:"https://www.livescience.com/38094-facts-about-rare-earth-elements-infographic.html" target:"_blank"/], with an estimated [link text:"70% coming from the Bayan Obo mining district" url:"https://www.mindat.org/loc-720.html" target:"_blank"/] in Inner Mongolia.
In her book [i]Rare Earth Frontiers[/i], Julie Klinger discusses how the process of mining “rare earth” minerals releases contaminated water into the Yellow River, Bayan Obo's local water supply.
The local population has suffered devastating effects, as farmland has turned to lakes of [link text:"toxic waste" url:"https://www.bbc.com/future/article/20150402-the-worst-place-on-earth" target:"_blank"/], and cancer related deaths have spiked. Since 2004, villagers have organized to compel the government to take action. While the state has not fully fixed the devastation it’s caused, it has responded to public pressure by removing some of the toxic waste and providing the people of Bayan Obo better access to healthcare [Cite
authors:"Julie Michelle Klinger"
title:"Rare Earth Frontiers: From Terrestrial Subsoils to Lunar Landscapes"
url:"https://www.jstor.org/stable/10.7591/j.ctt1w0dd6d.2"
id:"rare-earth"
date:"2017"/].

[div fullWidth:true]
  [img src:"static/images/bayan_aerial.jpg" className:"fullWidthImage"/]
[/div]
[caption]Satellite view of the Bayan Obo mining district. The open pit mines and lakes of toxic waste can be seen clearly even from sattelite images. (Credit: [a href:"https://earth.google.com/web/search/bayan+obo/@41.78380059,109.93209582,1610.25556031a,16680.42058981d,35y,-0h,0t,0r/data=CigiJgokCcTR8f5nsjhAEcLR8f5nsjjAGXUebCk3qkVAIUpA2yYEDVDA" target:"_blank"]Google, Maxar Technologies[/a])[/caption]

Tech's complicity in environmental destruction is not just limited to toxic manufacturing waste and a large carbon footprint. 
[span className:"emphasized"]Companies also have a large influence on the climate crisis in the context of policy, the broader economy, and the flow of information[/span].
Reports in 2019 revealed that Google has made significant contributions to [link text:"climate denialist groups" url:"https://www.theguardian.com/environment/2019/oct/11/google-contributions-climate-change-deniers" target:"_blank"/], including the Competitive Enterprise Institute (CEI), which helped convince the Trump administration to withdraw from the Paris Agreement in 2017. 
[link text:"Facebook has come under fire" url:"https://www.scientificamerican.com/article/climate-denial-spreads-on-facebook-as-scientists-face-restrictions/" target:"_blank"/] for lax action against climate change denial on their platform, where disinformation can easily spread without diligent fact-checking. 
[link text:"Google, Microsoft, and Amazon" url:"https://www.vox.com/recode/2020/1/3/21030688/google-amazon-ai-oil-gas" target:"_blank"/] have partnered with large oil companies to build machine learning tools that streamline oil production.
In fact, the [link text:"oil industry invested an estimated $1.75 billion in 2018 into machine learning tools" url:"https://www.globenewswire.com/news-release/2019/07/18/1884499/0/en/Global-AI-In-Oil-and-Gas-Market-Will-Reach-to-USD-4-01-Billion-By-2025-Zion-Market-Research.html" target:"_blank"/], which is projected to grow to $4.01 billion by 2025. 


[Aside]
  [img src:"static/images/amazon_protest_lilian.jpg"/]
  [caption]Amazon employees joined the global "climate strike" in September 2019 to protest their company's complicity in the climate crisis. Someone holds a sign that says "No AWS For Gas And Oil". (Credit: Lilian Liang)[/caption]
[/Aside]

Our purpose in throwing light onto this destructive behavior is not simply to paint a bleak future.
[span className:"emphasized"] Through public pressure and the collective organizing of tech employees, there has been media coverage of some success in holding the tech industry responsible for their environmental destruction.[/span]
In 2019, after the Amazon Employees for Climate Justice organization [link text:"led a large walkout" url:"https://www.latimes.com/business/story/2019-09-19/amazon-climate-change" target:"_blank"/] in support of the global climate strike, Amazon pledged to reach net-zero carbon emissions by 2040.
Similarly, in response to [link text:"employee pressure" url:"https://medium.com/@googworkersac/ruth-porat-497bbb841b52" target:"_blank"/], Google pledged to [link text:"stop funding climate change deniers in 2020" url:"https://agreenergoogle.com/" target:"_blank"/].
Google also promised to [link text:"rescind all future contracts" url:"https://www.cnbc.com/2020/05/20/google-ai-greenpeace-oil-gas.html" target:"_blank"/] with oil companies in response to a [link text:"Greenpeace report" url:"https://www.greenpeace.org/usa/reports/oil-in-the-cloud/" target:"_blank"/] about tech’s oil contracts.

Regarding digital consumption and its associated emissions, many tech companies have set varying carbon footprint targets for 2030 and beyond. Some of these companies will [link text:"rely on offsetting carbon emissions" url:"https://sustainability.aboutamazon.com/about/the-climate-pledge" target:"_blank"/] (e.g., by [link text:"planting trees" url:"https://www.nationalgeographic.com/science/2019/12/what-are-carbon-offsets/" target:"_blank"/]). Others plan to run their operations [link text:"completely on carbon-free energy sources" url:"https://blog.google/outreach-initiatives/sustainability/our-third-decade-climate-action-realizing-carbon-free-future/" target:"_blank"/]. It's important to note, however, that these pledges often don't include emissions produced by usage from sources the company doesn't "own" but is ultimately responsible for. For example, Google pledging carbon-free doesn't mean the YouTube videos sent over the internet and viewed on devices will be emission-less.

[Aside]
[link text:"The efficacy of carbon offsets is debated" url:"https://www.wired.com/story/do-carbon-offsets-really-work-it-depends-on-the-details/" target:"_blank"/]: there are no guarantees that carbon offsets actually result in the desired amount of greenhouse gas removed from the atmosphere, and it is never instantaneous.
[/Aside]

These pledges are not enough to fully mitigate these companies' destructive practices. In our present climate crisis, [span className:"emphasized"]carbon reductions cannot come soon enough, and carbon free technology should be the ultimate goal[/span]. In addition, they don’t account for the climate impact tech has through the other avenues we've discussed. The good news is that public pressure and collective organizing have the power to raise the bar for climate accountability—change can only occur if we understand and acknowledge the hidden, yet harmful consequences of the technology we use. 

[div style:`{marginTop: '7em'}`][/div]

[AuthorBio][b][a href:"https://haldenl.com"]Halden Lin[/a][/b] is a visualization designer, developer, and researcher. He's currently creating visualization experiences at Apple.[/AuthorBio]

[AuthorBio]
[b][a href:"https://twitter.com/AishwaryaNirmal"]Aishwarya Nirmal[/a][/b] works as a software engineer at Airbnb, and enjoys visual design and writing in her spare time. 
[/AuthorBio]

[AuthorBio]
[b][a href:"https://twitter.com/ssshathiii"]Shobhit Hathi[/a][/b] works as an applied scientist at Microsoft focusing on natural language processing.
[/AuthorBio]

[AuthorBio]
[b][a href:"https://twitter.com/liang_lilian"]Lilian Liang[/a][/b] is a writer and software engineer at Apple.
[/AuthorBio]

// Edited by TK editor list.

[Cite.References/]

[NextArticle slug:parametricSlug fullWidth:true /]

[Footer fullWidth:true /]

// Uncomment before pub
[Analytics google:"UA-139053456-1" tag:parametricSlug /]