BerandaComputers and TechnologyThe Cost of Digital Consumption

The Cost of Digital Consumption

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.

Book

E-Book

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?

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.
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.

Book

E-Book

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?

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, the Information and Communications Technology sector is expected to account for 3–3.6% of global greenhouse gas emissions in 2020 [1].
That’s more than the fuel emissions for all air travel in 2019, which clocked in at 2.5%.
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 need 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.

Get updates from the Parametric Press

Each type of digital activity produces different levels of emissions. The amount of carbon dioxide emitted by a particular digital activity is a factor of the quantity of information that needs to be loaded. More specifically, we can estimate emissions using the following formula:

$n$

bytes

$times$

$X$

KwH/byte

$times$

$Y$

g CO₂/KwH

A byte is a unit for information.

$X = 6 times 10^{-11}$

is the global average for energy consumed for transmitting one byte of data in 2015, as calculated by Aslan et al. (2017) [2].

$Y = 707$

represents the EPA’s 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

Emissions of Websites

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. ClickTap each bar to show a preview clip of the scroll.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

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, according to the EPA). These emissions are a result of loading data for the text, graphics, and visualizations that are then rendered on your device.

The chart to the rightbelow 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. Clicking Tapping each bar in the chart will show a preview of that scroll, sped up to a 5 second clip.

Emissions of Websites

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. ClickTap a bar to show a preview clip of the scroll.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

Audio

Emissions of Audio

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 buffering.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

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 bitrate, which refers to the average amount of information in a unit of time.

Emissions of Audio

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 buffering.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

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

Emissions of Video

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.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

Videos are a particularly heavy digital medium.
The chart to the rightbelow shows emissions for streaming different YouTube videos at two different qualities—360p and 1080p, for 60 seconds each.

Emissions of Video

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.

SOURCE

Aslan et al. 2017 [2] , EPA

Parametric Press

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 pixels. 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 plays a role in video streaming.
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.

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), the size of its website decreased from 5000 to 500 kilobytes with no significant changes to the appearance of the site.
This means that the updated, ad-free version of the website produced roughly 10 times less CO₂ on each load than the original.

Not only does a video require loading both audio and visual data, but visual data is also particularly heavy in information.

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), the size of its website decreased from 5000 to 500 kilobytes with no significant changes to the appearance of the site.
This means that the updated, ad-free version of the website produced roughly 10 times less CO₂ on each load than the original.

Packets

Media Emissions by Medium

Click Tap on the “Show Timeline” button to see a timeline of loading packets for each type of media. HoverTap and drag to scrub through and view cumulative emissions.

SOURCE

Aslan et al. 2017 [2]

Parametric Press

In a lot of cases, when you stream content online, you don’t receive all of the information for that content at once.
These pieces are called packets.
In each media emission visualization, we estimated emissions based on the size and quantity of the packets needed to load each type of media.

Click Tap the “Show Timeline” button at the bottom of the visualization to the rightbelow to see a timeline breakdown of each type of media.

Media Emissions by Medium

Click Tap on the “Show Timeline” button to see a timeline of loading packets for each type of media. HoverTap and drag to scrub through and view cumulative emissions.

SOURCE

Aslan et al. 2017 [2]

Parametric Press

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.

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.

A Case Study: YouTube’s Carbon Emissions

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. Every digital streaming platform relies on a system of computers and cables, each part consuming electricity and releasing carbon emissions.
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?

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: Santa Clara County, the heart of “Silicon Valley”, has the most EPA classified “superfund” (highly polluted) sites in the nation.
These 23 locations may be impossible to fully clean. Silicon Valley is primarily to blame.

In their book The Silicon Valley of Dreams, 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. Manufacturing these chips is inherently toxic, and its chemical runoff has been linked to increases in cancer and birth defect rates around the Bay.
However, the industry avoided responsibility for decades and actually cultivated a clean image, publicizing their pristine campuses and white-collar workforce [5].

More from the Parametric Press

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. It wasn’t until this public outcry that one of the largest Silicon Valley manufacturers closed down a particularly problematic factory [5].
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]” [6].

The tech industry’s legacy of toxic mining and manufacturing continues today—with hardware giants like Apple taking the throne of Silicon Valley’s founding industrialists—only now on a global scale.
In 2009, China produced 95% of the world’s “rare earth” minerals, with an estimated 70% coming from the Bayan Obo mining district in Inner Mongolia.
In her book Rare Earth Frontiers, 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 toxic waste, 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 [7].

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: Google, Maxar Technologies)

Tech’s complicity in environmental destruction is not just limited to toxic manufacturing waste and a large carbon footprint. Companies also have a large influence on the climate crisis in the context of policy, the broader economy, and the flow of information.
Reports in 2019 revealed that Google has made significant contributions to climate denialist groups, including the Competitive Enterprise Institute (CEI), which helped convince the Trump administration to withdraw from the Paris Agreement in 2017. Facebook has come under fire for lax action against climate change denial on their platform, where disinformation can easily spread without diligent fact-checking. Google, Microsoft, and Amazon have partnered with large oil companies to build machine learning tools that streamline oil production.
In fact, the oil industry invested an estimated \$1.75 billion in 2018 into machine learning tools, which is projected to grow to \$4.01 billion by 2025.

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)

Our purpose in throwing light onto this destructive behavior is not simply to paint a bleak future. 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.
In 2019, after the Amazon Employees for Climate Justice organization led a large walkout in support of the global climate strike, Amazon pledged to reach net-zero carbon emissions by 2040.
Similarly, in response to employee pressure, Google pledged to stop funding climate change deniers in 2020.
Google also promised to rescind all future contracts with oil companies in response to a Greenpeace report 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 rely on offsetting carbon emissions (e.g., by planting trees). Others plan to run their operations completely on carbon-free energy sources. 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.

The efficacy of carbon offsets is debated: 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.

These pledges are not enough to fully mitigate these companies’ destructive practices. In our present climate crisis, carbon reductions cannot come soon enough, and carbon free technology should be the ultimate goal. 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.

Halden Lin is a visualization designer, developer, and researcher. He’s currently creating visualization experiences at Apple.

Aishwarya Nirmal works as a software engineer at Airbnb, and enjoys visual design and writing in her spare time.

Shobhit Hathi works as an applied scientist at Microsoft focusing on natural language processing.

Lilian Liang is a writer and software engineer at Apple.

1. Assessing ICT global emissions footprint: Trends to 2040 & recommendations, Lotfi Belkhir and Ahmed Elmelgi. Journal of Cleaner Production. March 2018.
2. Electricity Intensity of Internet Data Transmission: Untangling the Estimates, Joshua Aslan, Kieren Mayers, Jonathan G. Koomey, and Chris France. Journal of Industrial Ecology. August 2017.
3. Evaluating Sustainable Interaction Design of Digital Services: The Case of YouTube, Chris Preist, Daniel Schien, and Paul Shabajee. CHI. 2019.
4. InterTubes: A Study of the US Long-haul Fiber-optic Infrastructure, Ramakrishnan Durairajan, Paul Barford, Joel Sommers, and Walter Willinger. SIGCOMM ’15. 2015.
5. The Silicon Valley of Dreams: Environmental Injustice, Immigrant Workers, and the High-Tech Global Economy, David N. Pellow and Lisa Sun-Hee Park. December 2002.
6. The uneven distribution of environmental burdens and benefits in Silicon Valley’s backyard, Iris T.Stewart, Christopher M.Bacon, and William D.Burke. Applied Geography. December 2014.
7. Rare Earth Frontiers: From Terrestrial Subsoils to Lunar Landscapes, Julie Michelle Klinger. 2017.

RELATED ARTICLES