A long road ahead: a review of the state of knowledge of the environmental effects of digitization

So far, most global assessments of the direct environmental effects of ICT have focused on electricity consumption (including manufacturing and use phases) and greenhouse gas (GHG) emissions (also called carbon footprint) — leaving out other major effects such as material or water footprint, but also pollutions.

The most prominent publications on this topic are Andrae and Edler [2], which is known to be outdated and superseded by Andrae [1], Malmodin and Lundén [32], and Belkhir and Elmeligi [4]. While they are based on the same breakdown of ICT equipment into data centers, telecommunication networks, and end-user devices, these studies have different perimeters. By unifying the scopes considered in Refs. [2,32,4], Freitag et al. [15] argue that GHG emissions from the ICT sector ranged from 2.1% to 3.9% of global emissions in 2020 (1.2–2.2 GtCO2e).

The wide range of ICT’s carbon footprint estimate, and most importantly the disagreement over the associated trend — Malmodin and Lundén [32] believe that it will stagnate in coming years, while Andrae [1] anticipates an increase — is explained by differences in the scope chosen, for example, whether to include Internet of Things (IoT), cryptocurrencies, and other new services, in assumptions (about energy efficiency, ICT

decarbonization, etc.) but also in extrapolation methods: Malmodin and Lundén project trends based on equipment sales, while Andrae and Edler’s projections are based on the evolution of data traffic. While the former is too narrow in scope to anticipate future uses and services, the latter is known to be inherently flawed because increasing traffic does not necessarily lead to a similar increase in footprint due to improved equipment efficiency.

In a recent paper, Pasek et al. [35] identify six key factors that make estimating the carbon footprint of ICT and its future trends complex or even impossible: access to industry data, bottom-up versus top-down assessments, system boundaries (scope), geographic averaging, functional units (e.g. kWh/GB or kWh/ subscriber), and energy efficiencies.

Recent work by Masanet et al. [33] estimates worldwide electricity consumption of data centers at 205 TWh in 2018, a modest 6% increase compared with 2010, despite a 10-fold increase in data traffic. In contrast, based on data sets from the European and German data center industry, Hintemann and Hinterholzer [22] estimated worldwide energy consumption of data centers at around 400 TWh in 2020. The difference between these two figures can be explained by differences in the estimated number of hyperscale data centers (which are particularly efficient), and “differences in scope (e.g. including or excluding cryptomining), methodologies, and assumptions” according to the International Energy Agency (IEA), which recently estimated that data center electricity consumption in 2021 was between 220 and 320 TWh [24] (without including cryptocurrencies).

The IEA estimates that energy consumption of networks was between 260 and 340 TWh in 2021 [24].

Coroama and Mattern [12] note that “the mechanisms behind rebound effects in general, and thus of digital rebound as well, are essentially nontechnical in nature. Their roots reside in economics and in human behavior.”

In its last report [36], the IPCC Working Group III mentions: “At present, the understanding of both the direct and indirect impacts of digitalization on energy use, carbon emissions and potential mitigation is limited.” Our brief review of the current state of knowledge is in line with this statement.

While ICT is at the core of most critical infrastructures and human activities, the role it can play to mitigate GHG emissions remains unclear. More research is needed to identify under which multifaceted conditions (taking into account territorial specificities) a digital solution can have positive effects, and under which it has negative effects, meaning it is better not to deploy it or even to ‘un-digitize’ it depending on the situation. Last but not least, one should acknowledge that our times require urgent action and that answers to many questions raised in this paper may not get scientifically sound answers quickly enough. One crucial question that remains is then: what do we need to know to make decisions?