Zoom, video conferencing, energy, and emissions

I was recently invited to discuss the topic of IT energy and the lack of transparency in an episode of the documentary series, Dispatches, which was broadcast on Channel 4 in the UK at 8pm on 16 November 2020 (available online here). One of the topics I was asked about was video conferencing, and Zoom in particular.

Although claims about how green video conferencing is compared to travelling to a meeting are not new, the sudden growth of online meetings caused by the Coronavirus pandemic has brought this into the mainstream.

It intuitively makes sense that speaking to someone over a Zoom video conference is better for the environment than flying to meet in person. The problem with intuitive assumptions is that they often work out to be more complex than you think.

For example, charging shoppers for plastic bags has been very successful in reducing the number of plastic bags used. However, whether that is better for the environment depends on what they are replaced with and how many times that replacement is used.

A lifecycle analysis from the Danish government shows that compared to a standard plastic bag, an unbleached paper bag is better when considering impacts on climate change, but it would need to be reused 43 times to have a positive impact on all environmental factors. This is because “environment” means many things: pollution, ozone depletion, causes of human cancer, acidification of freshwater, resource depletion, and others. The worst thing you can do is replace a plastic bag with organic cotton which requires 20,000 reuses before it is better for the environment.

In the context of video conferencing, it is safe to assume that arranging a Zoom call is better than flying business class from London to New York, but is it better than both participants walking to a cafe in the city they both already live in? Maybe someone was going to drive their EV to the office. What about if you have many participants all over the world? How about if some connect via 4G vs others on a laptop?

To go a bit deeper than is possible in a TV documentary, this post attempts some rough calculations to try and get an answer to these questions.

What does Zoom say about their emissions? #

Not much. They treat their environmental impact as marketing.

One blog post from 2019 explains how an example Zoom call would generate 4kgCO2 vs 687kgCO2 flying from New York to San Francisco for the equivalent meeting (the actual video meeting was only 0.0045kgCO2, the rest being commuting to the office).

Although the blog post mentions “University of Pennsylvania” as the source of this, I could not find how they calculated that figure. The other source mentioned is a 2009 report from the WWF which does not provide a specific number, and is 10 years out of date.

The other blog post states that Zoom’s top 10 customers avoided over 685k metric tons of CO2 in 90 days. This is useless unless you know what they actually emitted.

If Zoom is so great for the environment vs the alternatives, you might expect them to provide numbers to prove it.

Calculating Zoom’s emissions #

In the absence of Zoom providing any data about the emissions of a video call, we can perform some basic calculations ourselves.

Zoom recommends the following bandwidth requirements:

• For 1:1 video calling:
  • 600kbps (up/down) for high-quality video
  • 1.2Mbps (up/down) for 720p HD video
  • Receiving 1080p HD video requires 1.8Mbps (up/down)
  • Sending 1080p HD video requires 1.8Mbps (up/down)
• For group video calling:
  • 800kbps/1.0 Mbps (up/down) for high-quality video
  • For gallery view and/or 720p HD video: 1.5Mbps/1.5Mbps (up/down)
  • Receiving 1080p HD video requires 2.5Mbps (up/down)
  • Sending 1080p HD video requires 3.0Mbps (up/down)

For 2019, the electricity intensity of fixed-line internet data transmission is 0.015kWh/GB (extrapolating the 2015 figure from Aslan et al, 2018 to 2019), so we need to convert these figures from bits into gigabytes:

• For 1:1 video calling for 1 hour:
  • HQ = 0.000075GB/s = 0.27GB/hr * 2 (up/down) = 0.54GB.
  • HD 720p = 0.00015GB/s = 0.54GB/hr * 2 (up/down) = 1.08GB.
  • HD 1080p = 0.000225GB/s = 0.81GB/hr * 2 (up/down) = 1.62GB.
• For group video calling for 1 hour:
  • HQ = 0.0001GB/s + 0.000125GB/s = 0.36GB/hr + 0.45GB/hr = 0.81GB.
  • Gallery view and/or HD 720p = 0.0001875GB/s = 0.675/hr * 2 (up/down) = 1.35GB.
  • HD 1080p = 0.0003125GB/s + 0.000375GB/s = 1.125GB/hr + 1.35GB/hr = 2.475GB.

These figures are per participant which means that a 1 hour 1:1 conversation could generate 1.08 – 3.24GB of network traffic using 0.0162 – 0.0486 kWh of electricity.

For a group call between 6 people, that could generate 4.86 – 14.85GB of traffic and use 0.0729 – 0.22275 kWh of electricity.

At the upper limit, Zoom can support up to 1,000 participants (and has gone up to 100,000 in special situations), which could generate a range of 810 – 2,475GB of network traffic using 12.15 – 37.125 kWh of electricity.

Now we know the electricity consumed, we can calculate the emissions. This is where it starts to get more complex because the emissions factor depends on the grid mix in each location. A very simple calculation could assume that all network transmission remains within a single country.

For example, a 1:1 HD 1080p video meeting of 1 hour between two people would require 3.24GB of bandwidth, consuming 0.0486 kWh of electricity. The 2019 UK electricity emissions factor is 0.25358 kgCO2 per kWh, so the CO2 emissions for this call are 0.012 kgCO2.

If this happened in the US between two people in New York, the emissions factor is similar to the UK at 0.28839 kgCO2 per kWh but if it was between two people in Chicago, that would be 0.56191 kgCO2 per kWh. Location matters.

Big assumption – are data transfer and network energy correlated? #

Update July 2022: The main assumption from the above calculation is that network energy and data transfer are correlated i.e. the more data transferred, the greater the energy consumption. A lot of academic literature uses this model, which is why you can find kWh/GB figures in the likes of Aslan et al, 2018 and the Carbon Trust Carbon impact of video streaming report. But is this accurate? Simple extrapolations generally result in absurd numbers.

Since writing this blog post in 2020, a different model of assessing energy intensity of applications which use the network has emerged depending on whether you are measuring the instantaneous increase in power consumption caused by that consumption, or whether you are allocating energy for reporting purposes.

The bottom line is that you can use averages to calculate energy consumption for past usage if you know the total energy and total data transferred. With these values, you can calculate an average intensity and then allocate it to a specific application. That’s fine for reporting, but doesn’t accurately describe the real situation. You can’t use energy intensity figures to make assessments of the present or make projections for the future. These approaches are discussed in detail in the Carbon Trust Carbon report and at a high level in a 2021 article in Joule. I wrote also a post about it. More work is needed here.

What is missing from the calculation? #

Note how we are only talking about the network energy consumption between participants in a single location. This is the most basic situation, and things quickly become more complex:

• What about people in different countries? You need to apply a local emissions factor for each person.
• What about the type of network they are connected to? The figure of 0.015kWh/GB applies only to fixed-line networks. Mobile connectivity in 2020 uses 0.1kWh/GB (Pihkola et al., 2018).
• What about their devices? We have only calculated the energy of the network. The energy used by the computer, the display, and the phone or router they are connected to is excluded
• What about the data center? Zoom does not publish any figures about its own infrastructure and how much energy that consumes. Zoom primarily uses Amazon Web Services for their infrastructure, but they also use Oracle and 9 Equinix data centers. Oracle reports that Zoom are transferring up to 7 petabytes of data per day. Using the 0.015kWh/GB figure above suggests that would consume 105,000 kWh of electricity every day just for a small part of their network traffic.
• What about all the servers? Zoom said they were adding 5000 – 6000 new servers per day on AWS to help during COVID. That is a lot. We don’t know where those servers are, how long they run for, or what spec they are.

How does this compare to travelling? #

As usual, it depends. In the UK, a medium sized battery electric vehicle generates 0.05275 kgCO2 per km travelled compared to a medium petrol car which generates 0.19158 kgCO2 per km travelled. A round trip economy flight from London Heathrow to New York JFK generates 670.9 kgCO2.

We can easily find an example where there are significantly more emissions than a video call, but we could just as easily pick an example to show the opposite e.g. by walking to a local cafe to meet someone who also walked there.

If the travel distances from our case study are halved, such that the domestic participant travels 500 km and the international participant travels 2500 km, the overall energy contribution from the in-person meeting will fall by half to give 10.6 GJ (plane), 8.7 GJ (train) and 10 GJ (car). The travel distances will need to be shortened 15-fold (e.g. to a miniscule 67 km for the domestic participant and 333 km for the international participant) to reduce the lifecycle energy cost of the in-person meeting such that it matches the cost of videoconferencing. However, these results show that there are certain conditions in which having an in-person meeting is still better than videoconferencing environmentally, especially for high-end telepresences that have a higher energy profile.

Ong, Moors & Sivaraman, 2014.

The form of transport being compared makes a big difference:

The potential of teleworking for reducing transportation energy use and GHG emissions also depends on the form of transportation it is replacing. In regions with a high proportion of walking, cycling, and public transportation, the benefit of teleworking is significantly less than in sprawling cities where commuters primarily rely on personal automobiles and with bad congestion.

O’Brien & Yazdani Aliabadi, 2020.

You would need to have over 55,000 hours of 1:1 HD 1080p video meetings to equal the same emissions generated as a return flight from London to New York (670.9 kgCO2 for the flight / 0.012 kgCO2 for the video call).

But if you and your colleague drove your EVs 10km into a local office (10 km * 2 to/from journey* 2 people * 0.05275 kgCO2 = 2.11 kgCO2) and worked together for an 8 hour day, this would be equivalent to 21 hours of HD 1080p video meetings (2.11 kgCO2 for the journey / (0.012 kgCO2 for 1 hour video call * 8 hours ).

These numbers get us to the point where the uncertainty over the remainder of non-network-transmission emissions from a Zoom call could make a difference. When you add in Zoom’s data center and the electricity consumption from your computer or phone, it could be that using the London Underground to commute into the office for a day of meetings is better (or similar to) using Zoom.

Understanding the total impact is even more challenging. Rebound effects of more frequently working from home mean an impact assessment is difficult. O’Brien & Yazdani Aliabadi (2020) discusses effects such as:

• The increased frequency and/or distance of non-work trips.
• Many commuters carry out errands during their commute, making them more efficient.
• Telecommuters can avoid rush hour so when they do travel, they may travel further because it is more enjoyable.
• Tendency to live further from amenities means that those trips are more likely to be taken by personal car, rather than more sustainable means.
• Teleworkers tend to drive alone rather than take transit on non-telecommuting days.
• Telecommuting makes vehicles more available to other household members to use.
• A reduction in commuting costs (e.g., multiple household cars) frees up funds for other potentially energy-intensive purchases (e.g., larger house, international vacations).

These factors are of course outside the control of Zoom, but highlight the complex factors that make such claims more challenging to accurately assess.

What should Zoom do about this? #

It is obvious that video conferencing is better for the environment in general than flying. It is probably better than driving a petrol or diesel car, may be better than an EV, but might not be better than public transport. However, we can’t say much more than that. Zoom could help in the following ways:

1. Provide annual reports on the total amount of energy consumed across the entire operations of the company.
2. Provide users with reporting on the total amount of data transmitted by their Zoom client (ideally broken down by client type e.g. x GB from the desktop client and y GB from the mobile client). If participants are members of a company account, this could be aggregated for all users.
3. Follow the same approach as Google and Microsoft in working towards offsets + 100% renewable matching + 100% renewable sourcing for their data centers.
   1. If they do this, users only need be concerned with the emissions from their own data transfer because Zoom will be mitigating the data center emissions.
   2. If they do not do this, Zoom should provide an energy intensity figure in kWh/GB of data transmitted (or minute of video streaming) to consider all the electricity consumed in the network and their data centers. Users (or their organisation on their behalf) can then calculate the emissions from their video streaming, and mitigate it themselves.

Nothing has zero emissions. Zoom compares favourably to the alternatives such as flying. That is great progress but it isn’t sufficient. Where they should be embarrassed is that they don’t release enough for us prove what they claim. Otherwise it all just looks like greenwashing.