Final report, evidence brief, and best practices

Tacking the Carbon Footprint of Streaming Media completed our research this summer. You can find our final report, evidence brief, and best practices recommendations here
We can now corroborate The Shift Project’s analysis that streaming video is responsible for over 1% of greenhouse gas emissions worldwide. Exacerbated by new habits established during the Covid-19 pandemic, that figure is currently estimated at 1.2% to 1.4% and rising fast (Sandvine 2020). We triangulated among ICT engineers’ calculations of streaming video’s electricity intensity, the electricity needed to move one gigabyte of data, and also did Fermi calculations. The research remains contentious, but we are confident that our general findings are correct.
Our literature review played out like a detective story. Throughout the literature, the disparity between figures is enormous. Just like Maxime Efoui-Hess, main author of The Shift Project’s report, we “quickly realized that much of the literature on the subject used figures from previous documents, very often without cross-referencing them with others, and without taking precautions regarding the limits of their validity." This practice leads to error propagation, where the uncertainties of variables multiply.
We observed what appear to be political differences between at least two camps. There are those researchers who insist that ICT’s electricity consumption is remaining flat, while also urging that governments invest in research and infrastructure to deal with the rise when it happens.
We found a great degree of disagreement among ICT engineers about both figures and methods of their calculation. The dissensus largely centers around varying definitions of the system boundary of the Internet and of streaming; that is, whether devices, data centres, production, disposal and mining of metals should be included. The classic definition of the internet is confined to networks: long-haul, local, and consumer. However, in the case of streaming video and other data- and calculation-intensive applications, it is essential to include data centers and servers, including storage. We also side with engineers who argue that devices—smart televisions, TV-top devices, desktop and laptop computers, tablets, and phones—must be included in the system boundary of streaming video, and so must include the electricity required to produce devices and other ICT infrastructure.
Another key finding is that streaming video epitomizes the rebound effect (also known as the Jevons paradox), whereby increased energy efficiency leads to greater consumption of a resource. The engineering literature vigorously debates the rebound effect. Some argue that increases in ICT energy efficiency more than compensate for the acceleration of consumption, while others point to numerous case studies that show that when the production and use of data centers, networks, and devices become more electricity-efficient, demand by companies increases. They pass their savings on to consumers, for example in the form of cheap data plans and cheap devices, but in order to encourage them to consume more. So yes, streaming one video does consume less electricity than driving to the video store. However, the availability of online video has created new consumption patterns, driven by addictive design, which cancel out any energy savings. Streaming video exists within a market-driven feedback loop of infrastructural expansion and consumer demand.
We found that videoconferencing also poses the danger of rebound effects. Currently videoconferencing uses less electricity per minute than streaming video, but only because the companies like Zoom impose low resolution (Obringer et al. 2021) and a slow frame rate. However, as people replace phone calls with video calls and companies like Cisco and Peloton market “immersive” teleconferencing with large, high-resolution screens, new habits consume a lot more energy.
We found that redundancy, or the doubling of power supplies for data centres and networks in anticipation of spikes in demand, is one of the foundations of ICT’s disproportionate carbon footprint. Overpreparedness for worst-case scenarios—where the worst case is not, for example, the failure of the data center in a nuclear power plant, but, for example, the failure to deliver high-resolution streaming movies without lag time—is one of the foundations of ICT’s disproportionate carbon footprint. Data center and network security is predicated on redundancy, the doubling of power supplies (traditionally by diesel generators and battery packs), networks, and other equipment that runs in standby mode to prevent momentary blackouts or system failures (Schomaker, Janacek, and Schlitt, 2015). These dramatically amplify electricity consumption. Horner and Azevedo (2016) point out that, because of the priorities of uptime, reliability, and fulfillment of service agreements, data centers are generally built with extreme redundancy. As Tung-hui Hu suggests, infrastructure “converts an imagined crisis in the future into present capacity."