Debate: The data threat to 2050 net zero—public administrations’ responsibility for the ‘data-scape’

Why it matters

The significance of the energy consumed by data to public administration cannot be overstated as climate change continues to represent a significant future crisis to humanity (Kobayashi & Omori, 2023). Signatories of the Paris Agreement are committed to a worldwide effort to achieve net zero emissions by 2050. Governments’ net zero strategies subsequently outline pathways to decarbonization within which ‘digital’ plays a crucial role in realizing this long-term goal. However, it has been recently observed that our current global trajectory will not meet this climate target (Škare & Porada-Rochoń, 2023).

One pressing challenge to governments’ net zero strategies is the new data landscape. With recent reports suggesting that the total size of the ‘data-scape’ is increasing dramatically, the need to understand the relationship between data and energy is extremely pressing. This is in part evidenced by the increasing media attention on the energy-intensity of data centres, which play a central role in the digital world (Mikkonen & Wilson, 2023). Understanding the intricate relationship between data and energy consumption is therefore an imperative component to reducing global greenhouse gas emissions, especially CO₂, and reaching net zero by 2050.

The energy cost of 1GB of data is a hotly-contested debate—it is a complex and multifaceted issue, riddled with various interpretations and ambiguities in both academic and grey literature. This complexity arises from the diverse considerations involved in calculating kilowatt-hours (kWh) per unit of data, including data storage, transmission, and computing processes. Moreover, a lack of transparency among key players in the data industry further muddles the clarity of this matter (Mytton & Ashtine, 2022). Nevertheless, mitigating climate change is dependent upon better and more co-ordination between different organizations and sectors (see, for example, Pollitt, 2015) and this need still holds in the case of the climate impact of data.

To contribute to this discussion and to provide more clarity on the matter for public policy development, in this article we provide a summary of how the kWh/GB has evolved over time and offer a further means by which the question of how much energy does 1GB consume might be determined.

At this juncture, you might wonder if it is even feasible to ascertain the average energy consumption associated with 1GB, considering the diverse paths it may traverse. This data could range from serving as the backbone for training a sophisticated Artificial Intelligence model, to being a YouTube video viewed by a handful or millions of people across the globe, or even residing in the depths of cold storage and largely forgotten. Nonetheless, it's important to recognize that every piece of data undertakes a unique energy consumption journey. By attempting to calculate an average consumption figure, public policy-makers gain a valuable tool for forecasting the potential environmental impact of data being generated, processed, and stored across industries and sectors.

Tracing the journey to date

In the quest to determine the energy consumption of 1GB, our journey leads us back to two pivotal sources that elucidated the intricate relationship between data size, energy consumption, and CO₂ emissions. In 2009, a seminal white paper by Weber, Koomey and Mathews, colloquially referred to as the ‘Carnegie report’, synthesized previous research to establish an approximate energy consumption rate of 7 kilowatt-hours per gigabyte (kWh/GB) within the realm of internet data flows. Notably, they forecasted a halving of this rate every two years. Furthermore, the paper unveiled the range of 0.005 to 0.009 kilowatt-hours per megabyte (kWh/mb) for data centre energy use specifically.

In the subsequent year, 2010, these authors reaffirmed their findings, confirming the 7 kWh/GB figure coupled with the foresight of a 50% reduction every two years (Weber et al., 2010). These foundational studies shed light on the enduring significance of these energy consumption patterns in the data landscape. Another significant advancement came in 2018 in an article published in the Journal of Industrial Ecology by Aslan and colleagues, who provide a revised kWh/GB figure of 0.06 for fixed line networks in 2015 (based on a BT network in the UK). The authors confirm the ongoing halving every two years based on projected estimates for average transmission network electricity intensity. They are clear about their focus being on data transmission and show the huge variability in what has been included in previous calculations of the kWh/GB figure by different studies (Aslan et al., 2018; table 2).

Recent publications have adopted 0.06 kWh/GB as the benchmark, adjusting the figure based on the suggested halving every two years from 2015 (for example Barnsley et al., 2022; Blenkinsop et al., 2021). There is a lack of clarity, however, in how the unit of measurement is being applied. As observed by Pachilakis et al. (2023) of the Norton Group, many studies refer to the total electricity consumption of 1GB. Consequently, such studies appear to exclude the ongoing storage/memory cost of data. Not specifying if storage is accounted for, either as a one-off energy cost or on a rolling basis for the life of the data, is problematic as the ramifications are significant.

Moving forward

With most studies focusing solely on data transmission, which is defined as the ‘electrical energy consumed per amount of data transmitted’ (Aslan et al., 2018, p. 786), there is a risk that only a part of the story is being captured. Every unit of data may consume varying amounts of energy throughout its lifecycle, including the time it spends traversing a network, the duration of processing, and its storage across different mediums. The question that arises, then, is how can we obtain an approximate average figure for one unit of data to enable the forecasting of the environmental impact of data?

One possible approach is to examine the overall energy usage of a data centre, but to do so we must rely on available open ‘evidence’, which itself is sparse and limited owing to the lack of transparency across the data industry (Mytton & Ashtine, 2022). In addition to this challenge for policy forecasting, how an average figure should be calculated from such evidence poses a further significant issue. To illustrate this point, we show two different calculations and outcomes in Figures 1 and 2 below. Figure 1 determines the power (kWh/GB) requirements to maintain the system and Figure 2 determines the kWh per GB.

To begin, Facebook estimate that it uses 532GWh a year (Krug et al., 2014). It is estimated that it is responsible for between 1.3% and 3% of internet traffic (Sandvine, 2014). We assume that the total internet traffic is 31339 PBytes a month, as given by Cisco (2013) for 2012. This implies of order 20 to 50 µJ per bit. To ensure the same starting point, both illustrations adopt the upper value of 50 µJ per bit.

Though it appears to be a straightforward calculation, the illustrations in Figures 1 and 2 yield two very different answers. A key reason for this is that one includes a temporal dimension, while the other does not. In Figure 1, the result, 0.001349104 kWh per GB is calculated per one hour. However, in Figure 2, the result is 0.119 kWh per GB, but without a specific time period attached to it. These two illustrations highlight how, without consensus and a standardized approach to both the ‘input data’ and calculation to be adopted, policy-makers are severely limited in their ability to develop appropriate data carbon policies in support of governments’ net zero strategies.

To achieve a thorough comprehension of data energy consumption, future research is imperative. Investigations should focus on comprehensive data lifecycle analysis, accounting for such components as data centre efficiency, storage costs and evolving technology trends (for a thorough discussion—see Mytton & Ashtine, 2022). ‘On the ground’ public managers can ensure that ‘data’ features as part of their organization’s sustainability strategies, which should drive and promote responsible knowledge management practices (for example data minimization, efficient storage and responsible data disposal). Additionally, increased transparency from industry leaders and standardized reporting practices driven by public administrations and policy bodies would greatly aid in developing a clearer picture of the environmental impact of our data-driven world as we progress towards 2050. Specifically, it is crucial to develop a straightforward and user-friendly method for comprehending scenario-based energy consumption throughout the entire lifecycle of data for policy development.

Note

The calculations and numerical data included in this have not been independently verified by PMM’s editorial team.

Acknowledgements

We are extremely grateful to several colleagues for spurring the debate as to how the energy consumption of 1GB might be calculated. Special thanks are recorded to Dr Andrew Watson.

Additional information

Notes on contributors

Thomas W. Jackson

Tom Jackson is a Professor at Loughborough Business School, UK. With over 20 years of experience in information and knowledge management, he is renowned for creating EMOTIVE, a fine-grained emotion detection system, and is a pioneer in digital decarbonization, information overload and workplace interrupts.

Ian R. Hodgkinson

Ian R. Hodgkinson is a Professor of Strategy at Loughborough Business School, UK. Ian’s research to date has involved international healthcare policy, public spending and service optimization, digital solutions to wicked problems, and net zero digital initiatives.