Environmental footprint of digital art: reading the figures

In 2021, a significant portion of the environmental debate surrounding digital art was reduced to a single figure: the energy attributed to a transaction on a blockchain. The comparison with household consumption was easy to communicate, yet it conflated network, transaction, sale, and work as if they were the same unit.

The issue did not disappear when Ethereum drastically reduced the energy consumption of its consensus mechanism. A digital work may depend on files, servers, networks, screens, projectors, computers, sensors, transport, climate control, equipment replacement, and long-term preservation. No single figure automatically encompasses all these components.

This is why two estimates may differ without either being fraudulent. The divergence may arise from different boundaries, incomplete data, incompatible models, or different calculation periods. It may also conceal an error. To distinguish between them, one must ask what activity is attributed to the work, for how long, with what data, and by what method.

Before calculating, one must define the object

“The footprint of a digital work” may refer to very different operations:

The loading of a page or the transfer of a file.
A complete visit to an online work.
Reproduction during an exhibition.
The annual operation of servers and external services.
The manufacture, use, and end-of-life of equipment.
The installation, transport, and climate control of the gallery.
The registration of provenance or ownership on a blockchain.
The preservation of the work over a specific period.

There is no single figure until those boundaries are fixed. An estimate per visit serves to compare versions of a website, but it does not, by itself, describe an installation exhibited for three months. The electricity consumption of a screen during opening hours does not automatically incorporate its manufacture, nor does it allow for the attribution of the entire piece of equipment to the project if it is reused in other exhibitions.

The functional unit also changes the result. Emissions can be calculated per visit, hour of exhibition, visitor, transaction, year of conservation, or life cycle. Each unit answers a different question. Comparing them as if they measured the same thing produces apparent differences that, in reality, stem from the design of the calculation.

What CO2.js estimates

CO2.js, developed by the Green Web Foundation, is an open library for estimating emissions related to the use of websites, applications, and other digital services. It does not directly measure the electricity consumed by a work. It converts transferred bytes into an estimation using a declared model.

Its current default model is version 4 of the Sustainable Web Design Model. It takes data transfer as an indirect indicator of resource use and distributes the estimation among data centres, networks, and user devices. Unlike previous versions, it includes operational components and embodied emissions associated with the manufacturing of the infrastructure for those three segments.

The improvement corrects a frequent assertion: it is inaccurate to state that these estimations always exclude the manufacturing of devices. Version 4 incorporates a modelled allocation of embodied emissions. What it does not do is provide a specific analysis of the monitor, projector, or computer used in a particular exhibition. For that, one requires data on the equipment, its service life, its degree of utilisation, and a defensible allocation rule.

The result from CO2.js depends, among other factors, on the bytes transferred, the chosen model and version, whether it is a first or subsequent visit, the use of cache, the hosting, and the carbon intensity applied. The Green Web Foundation further warns that the change between versions 3 and 4 of the model can reduce comparable estimations by approximately two-thirds without the website having changed. Publishing a value without indicating the version prevents its reconstruction and may turn a methodological update into an apparent environmental improvement.

CO2.js is useful when the question is well-defined: how much the estimation changes when reducing page weight, avoiding transfers, introducing cache, or modifying hosting. It is less suitable for converting the load of a website into the complete footprint of a work, an exhibition, or a collection.

Ethereum reduced a part, not the whole

Ethereum executed The Merge on 15 September 2022 and replaced proof-of-work with proof-of-stake. The platform itself estimates the reduction in the network's energy consumption associated with the change in consensus mechanism at approximately 99.95%. Its current estimate places the annual consumption of the entire network at around 0.0026 TWh, although the value varies with the number and configuration of nodes and the method used.

The change invalidates a large portion of the figures prior to September 2022 when applied to the current Ethereum. It also necessitates specifying the network and the date: it is inappropriate to carry over the consumption of the former proof-of-work to a sale after The Merge or to extend Ethereum's reduction to other chains with different mechanisms.

The 99.95% does not equate to an identical reduction in the total footprint of a work. It refers to the network's energy consumption upon replacing its consensus mechanism. It does not, by itself, include the production of the file, hosting, the access website, off-chain storage, display devices, the exhibition space, or the manufacture and replacement of equipment.

Nor is it advisable to simply divide the annual consumption of the network by the number of transactions and present the quotient as the causal impact of each operation. Infrastructures share consumption and do not always vary linearly with an additional transaction. That distribution may serve as an attribution estimate if explained, but it does not demonstrate how much energy the network would have ceased to use had that transaction not occurred.

Exhibition and equipment change the scale

For a work displayed in a gallery, data transfer may be secondary to operating time, equipment power, space conditioning, and device manufacture. It cannot be stated as a general rule that embodied emissions always exceed those of use. The relationship depends on the equipment, the electricity used, the hours of operation, the service life, and the number of projects among which it is distributed.

Equipment purchased for a single installation and retired shortly after will bear a different allocation than one used for years. Similarly, a low-consumption screen that remains on for thousands of hours may accumulate significant operational emissions. The methodological decision must explicitly state whether the entire manufacture is attributed, a fraction based on service life, or only the usage during the exhibition.

The Gallery Climate Coalition calculator is designed for arts organisations and projects. It prioritises travel, artwork transport, and building energy, and incorporates additional categories for accommodation, materials, and digital activity. It is suitable for identifying significant focus areas within an organisation or project. It does not replace a life-cycle analysis specific to each screen, projector, computer, or electronic component.

This distinction prevents the aggregation of incompatible results. The estimation for a webpage, the electricity measured in a gallery, and the footprint declared by a manufacturer may possess different boundaries, units, and factors. Before aggregating them, one must verify that infrastructures are not duplicated, that the period is consistent, and that all figures are expressed in the same unit of carbon dioxide equivalent.

How to publish a figure that can be discussed

A useful estimation does not need to feign absolute precision. It must render the procedure visible. At a minimum, it should indicate:

Which work, activity, or service is being calculated.
Which components remain within and outside the boundary.
Which period and functional unit are utilised.
Which data are measured and which are estimated.
Which tool, model, and version are applied.
Which electrical and geographical factors are employed.
How cache, repeat visits, and the number of users are treated.
How the manufacturing and service life of equipment are allocated.
Which uncertainties may most significantly alter the result.
Whether the calculation represents an accounting attribution or a change caused by a specific decision.

It is also advisable to publish a range or several scenarios when uncertainty is broad. An exhibition may be calculated using different equipment service lives, electricity mixes, or attendance figures. The variation demonstrates which assumptions dominate the result and where a decision may produce a genuine reduction.

Data must lead to action. Reducing unnecessary transfers, compressing files without destroying relevant qualities, switching off equipment outside opening hours, extending its service life, reusing it, choosing appropriate infrastructures, and avoiding replacements without artistic benefit may be more defensible measures than offsetting an estimated figure at the end of the process. The priority depends on the specific inventory.

Responsibility does not rest solely with the artist. Platforms, data centres, manufacturers, institutions, curators, and exhibition designers determine a significant portion of the available options. As I argued in “Mass technology and ethical consumption”, many individual decisions are conditioned by technical and commercial systems. The guide on how to estimate the environmental footprint of a digital artwork develops a procedure applicable to a specific piece.

Discrepancies between figures should neither be celebrated nor concealed. They may reveal a legitimate difference in method, but also poorly defined boundaries, double counting, or outdated assumptions. An environmental figure acquires value when it allows for the reconstruction of the calculation, the comparison of compatible scenarios, and the determination of what to change. Without that information, it offers an appearance of accuracy, not a verifiable measure of the work.

About the open conversation

This text separates the data, the methodological boundaries, and my interpretation. It does not constitute legal advice or an environmental verification. If you work in cultural sustainability, digital infrastructures, or conservation and wish to clarify a datum or discuss this reading, this notebook remains open: write to me.

The author's relationship to the cases

There is no financial, professional or collection relationship with the individuals, companies and institutions mentioned.

Environmental footprint of digital art: why the figures differ