Energy consumption per transferred data is a bad metric

Mytton et al. (2024)

It is commonly assumed that data volume and network energy consumption are directly proportional, a notion perpetuated by numerous studies and media coverage. This paper challenges this assumption, offering a comprehensive examination of network operations to explain why the relationship between energy consumption and data volume is nonlinear. The power model approach is explored as an alternative methodology for calculating network energy consumption providing a more reliable representation of network energy use. The power model demonstrates that simple energy intensity calculations, expressed as kilowatt hours per gigabyte of data, are insufficient for accurately estimating real-world network energy consumption.

Mytton, D., Lundén, D., & Malmodin, J. (2024). Network energy use not directly proportional to data volume: The power model approach for more reliable network energy consumption calculations. Journal of Industrial Ecology. https://doi.org/10.1111/jiec.13512

There is no correlation

IEA 2019

IEA 2019 – Global trends in internet traffic, data centre workloads and data centre energy use, 2015-2021-20240430084919512.webp

@Lunden.etal.2022.ElectricityConsumptionOperational

lunden.etal.2022.electricityconsumptionoperational (pg. 16)

As illustrated in figure 12, there is a substantial growth in processed and transported data and, hence, increased capacity over time. However, this has not been followed by a corresponding increase in electricity consumption.

lunden.etal.2022.electricityconsumptionoperational (image) (pg. 16)

Figure 12. Electricity consumption and data traffic for the reporting ETNO operators of 2015–2018 (ETNO Total data set), also including less granular data reported for a more limited number of operators for the period of 2010–2015, as reported in Reference [10]. The electricity consumption and the data traffic for the full period have been indexed in relation to the 2015 level.

lunden.etal.2022.electricityconsumptionoperational (pg. 15)

Figure 12 shows the reporting network operator’s electricity consumption (indexed) compared to the indexed data traffic for the ETNO Operators of 2015–2018 (ETNO Total data set). In addition, the reported data set for the period 2010–2015 [10] has been included to derive the trend over a longer time interval. Data traffic was reported to a more limited extent for the period 2010–2015 and only for a few years and by a limited number of operators; thus, only data points for 2010 and 2015 have been included in the figure. The electricity consumption shown in figure 12 represents the overall electricity consumption for the reporting ETNO operators, hence including the operators’ overhead, network, and data center operations (details for each part were outlined in figure 6).

Rudolf van der Berg

Rudolf van der Berg – Misconceptions about the Internet hurt the planet

Energy consumption of the Internet (PDF)

Some examples of good studies:

Sources of data center energy estimates: A comprehensive review, Mytton D., Ashtine M. (2022) Joule, 6 (9) , pp. 2032-2056.
Lundén, Dag, Jens Malmodin, Pernilla Bergmark, and Nina Lövehagen. 2022. "Electricity Consumption and Operational Carbon Emissions of European Telecom Network Operators" Sustainability 14, no. 5: 2637. https://doi.org/10.3390/su14052637 → @Lunden.etal.2022.ElectricityConsumptionOperational
The carbon footprint of streaming video: fact-checking the headlines, Kamiya, IEA, 2020 https://www.iea.org/commentaries/the-carbon-footprint-of-streaming-video-fact-checking-the-headlines
Koot, Martijn & Wijnhoven, Fons, 2021. "Usage impact on data center electricity needs: A system dynamic forecasting model," Applied Energy, Elsevier, vol. 291(C).

Energy proportionality of networking devices

jacob.vanbever.2023.internettomorrowmust (pg. 2)

The energy consumed by today’s networking devices is essentially independent of their load, as illustrated in Figure 1; i.e., most of the cost comes from powering on the device (P0), regardless of its utilization.

jacob.vanbever.2023.internettomorrowmust (image) (pg. 2)

Attributional vs. Consequential Assessment

It can be useful to use a constant energy value per data transfer in an attributional context, allowing the organization to allocate responsibility of emissions to different parties. However, in a consequential context it should not be used! In other words, it should not be used as a multiplier to determine that a system's carbon footprint has been reduced simply by a reduction in its page size.

Source: https://www.techcarbonstandard.org/technology-categories/networks (Technology Carbon Standard)