Skip to content

Sustainability is the single most important societal challenge of the 21st century, and computing turns out to be both a part of the problem and a part of the solution.

What does it mean to be sustainable?

by Koen De Bosschere and Patrick Blouet

When IT devices are on display in shops, they look shiny, attractive, clean. At the time of buying, very few people think about the impact of mining raw materials, the purification of the materials before they can be used in a fab, the energy the device will consume during its entire lifetime, and where it will eventually end up after they stop using it. Most users of the devices might not even be aware where and how the device they are about to buy has been produced. In order to become more sustainable, consumers have to be more knowledgeable about the environmental impact of computing devices, and this knowledge should be based on solid and universally accepted facts instead of beliefs. As long as some people disagree on the scientific facts and evidence, they might not take the right decision to reduce their carbon footprint, and they will have difficulties to assess the sustainability claims made by device producers.

Key insights

  • Determining the sustainability of computing solutions turns out to be extremely difficult, especially if one wants to take into account the full life cycle of products, and the secondary effects in the sectors where they are applied.

  • There are several sustainability beliefs that may hold at a local scale, but are not valid at the global scale, like efficiency = sustainability (does not take into account the rebound effect) or computing makes other sectors more sustainable (the so-called enablement, for which there is little evidence over the last 50 years).

  • Sustainability is a complex problem. The effective solutions must be based on universally accepted sustainability models, and not on popular beliefs.

Key recommendations

  • There is an urgent need for practical and universally accepted life cycle models for computing devices and services. The models should be used to steer decisions in the design and operation of computing systems, placing it on par with other design objectives such as cost, power consumption, and security.

  • There is a need to educate the computing community on sustainability and to encourage it to start working on it. The only antidote against greenwashing are consumers that have a basic understanding on what true sustainability actually means.


It was Jean-Baptiste Colbert, one of the main ministers of the French king Louis XIV, who, 350 years ago, saw the economic potential of the production of marine timber in the French forests. Without the ordinance of 1669 drawn up under his leadership, the majestic French-style oak groves like those of Tronçais or Bercé in France, managed by several generations of foresters, would not be available to us today.

According to (Grober, 2007), the term “sustainability” was introduced in the 18th century and it referred to managing forests to ensure that future generations had access to enough wood for fuel and for timber. In the 20th century it was broadened beyond forests. The first report of the club of Rome, “The Limits to Growth” (D.H. Meadows et al. , 1972), made very clear that exponential growth of resource consumption could not be sustainable on a finite planet, and that one day natural resources would be depleted. The same report already warned about the exponential increase of CO2 in the atmosphere in 1972. Later, the definition of sustainability was broadened with social and economic aspects into the formulation of the “triple bottom line” consisting of people, planet and profit (Triple bottom line, 2008). Today’s generally accepted definition of sustainability are the 17 sustainable development goals defined by the United Nations in 2015 (The 17 Goals, 2015).

Of all 17 sustainability goals, Goal 13: Climate Action is probably the best known. It states that “To limit warming to 1.5°C, global greenhouse gas emissions must fall by 45% from 2010 levels by 2030 and to zero by 2050”. Already in 2018, the UN Secretary-General António Guterres warned the world that “Climate change is moving faster than we are”; since then, greenhouse gas emissions have only increased. Today, the projected greenhouse gas emissions will be 10% higher in 2030 than in 2010. In order to reach the goals of 2030, we have to reduce the emissions by 42% in the next seven years (United Nations Environment Programme, 2023). This will be next to impossible to realize. With the current efforts, the climate will warm up between 2.5-2.9°C by the end of the century. Such a temperature increase will have catastrophic consequences for the whole planet.

But why is it so difficult to decarbonize the world? Some countries do not care about climate change because e.g. selling fossil fuels is a major source of income, or they simply do not have the resources to do it (e.g. in developing countries). Individuals are reluctant to cut down the use of fossil fuels because it requires investments (isolation, purchase of a heat pump or an electric car), or it requires lifestyle changes like moving to a smaller house, travel less, consume less, eat less meat, … Some industries are reluctant to make the energy transition as it will impact their bottom line of the next quarter; unless forced to change, they prefer business-as-usual as long as possible. The consequence is that fossil fuel consumption is expected to further increase until around 2030, and that the gap between the required reduction and the actual consumption will increase until 2040 (SEI, Climate Analytics, E3G, IISD, and UNEP, 2023). In order words, that we are not moving in the right direction fast enough.

One of the reasons why it is very difficult to decarbonize the world, is that the world is literally built from carbon (Vaclav Smil, 2022). The built environment primarily depends on cement and steel, both of which require lots of energy to produce, energy which has been fossil fuel based for the last 200 years. Many of the objects that surround us are made from oil in the form of plastics: clothing, furniture, toys, packaging, bottles, … The biomass that we grow (food, feed, wood, …) needs fertilizer, most of which is made from natural gas. Hence, we do not only need fossil fuels because they generate hot air, but also because we need their molecules in the chemical industry.

Given the current trends, reducing the global greenhouse gas emissions by 42% by 2030 won’t be realistic anymore as illustrated in Figure 1. Even the global lockdown in the year 2020 created only a small dent in the emissions.

Figure . Total Green House gas emissions 1990-2022 (United Nations Environment Programme, 2023)

We simply cannot decarbonize a fossil fuel based global industrial ecosystem that took 200 years to develop into a carbon-neutral one in few decades, especially because there are not yet proven alternatives for several types of emissions (e.g. air travel), and fossil fuel capital investments made today are expected to last for decades. For electricity production, there are several alternatives for fossil fuels (wind, solar, nuclear, hydro, biomass, …), but given the investment rate, the targets for 2030 are practically unreachable.

In the meantime, climate change is accelerating (Figure 2), and the effects of the warming become noticeable in our daily lives in the form of more extreme weather events.

Experts warn that the use of fossil fuel not only emits CO2, but also other particles like sulphur dioxide and soot that make the clouds more reflective and reflect up to 10% of the incoming radiation back into space. Phasing out the use of fossil fuels will automatically reduce the associated pollution (which is good), but also reduce global dimming, and thus increase the global temperature (which is not good).

Figure 2. Temperature evolution 1880-2023 (NASA SVS | NASA Summer 2023 Temperature Media Resources)

So, even if we would stop burning fossil fuels tomorrow, the climate will continue to warm for some time due to the reduction of global dimming. No matter what we do to fight global warming, we will also have to adapt to the changing climate, and therefore, we need a two-pronged approach: reducing emissions as fast as we can, while preparing for climate change. The worst effects of climate change are the more violent weather phenomena (storms, droughts, floods), and their impact on food production in the northern hemisphere (where most of the global food production happens). E.g. a reduced runoff in the Colorado river due to less snowfall in the Rocky Mountains would have a serious impact on the agriculture in southern California, one of the food baskets of the USA. Similar scenarios apply to the whole Mediterranean basin. A reduction of 10% in yield during a couple of years will lead to global famine, large scale migrations and political instability which does not help to increase food production (see Ukrainian war). Given the fact that agriculture is responsible for more than 25% of the global CO2e emissions, it is paradoxical that in order to slow down global warming, the world should cut down on agriculture to avoid global famine in the future.

In any case, it is in our interest to find ways to adapt to climate change (e.g. by selecting crops that are better resistant to the new climatological situation and compatible with the local ecosystem, and to manage the rainfall in a more effective way by buffering it for future use or to let it infiltrate and replenish the water table, …). It is mandatory for water to have several lives and to abandon the current model of single use.

In this article, we discuss some basic sustainability concepts.

Emissions by the IT industry: embodied vs. operational

Sustainability in the IT sector has two components. First of all, there are the resources needed to manufacture the electronic devices, and secondly, there are the resources needed to actively use the device, in practice it is primarily electrical energy and water. Although a single transistor or a single memory cell requires an almost infinitesimal amount of energy to switch, once multiplied by billions per device, and switching billions of times per second leads to measurable power consumption, ranging from less than 10 mW of operational energy for very low power devices to MWs of power for data centers. This multiplied by a few billion active devices on the planet leads to an industry consuming around 9.3% of the global electricity production and rising.

Concerning the minerals, modern IT devices are very complex devices that need complex supply chains to bring tens of semiconductor grade (= ultrapure) minerals from all continents together in one place, in order to integrate them into one device via a very sophisticated and energy and water consuming production process. The energy required and the emissions caused by this process are called the embodied energy and the embodied emissions, and they both contribute to global warming.

In order to contribute its share to the climate efforts, the IT industry should reduce its greenhouse gas emissions by 50% between now and 2030. However, according to Schneider Electric (Vincent Petit, 2021), the total energy consumption (used here as a proxy for the emissions) of the digital economy is projected to grow by 50% by 2030. So, instead of halving its energy footprint, it will double it!

Figure . Evolution of IT energy demand

A major source of growth is device manufacturing (the electricity part of the embodied energy), mobile networks (5G), and compute power (AI, blockchain, …). The growth rate (2020-2030) in the operational energy will be more than 15% for storage, mobile networks and IoT devices. Where the share of energy consumption of the IT sector was 9.3% in 2023 (including the energy to manufacture devices) will grow to 11.5% in 2030. Unfortunately, the energy required for dismantling and end of life management is not taken into account in this study.

The carbon footprint of the IT industry has been growing by 1.8% per year during the last decades, and this is faster than the growth of the global emissions. Its carbon footprint in 2020 is 2.1-3.9% of the global greenhouse gas emissions, and 30% of it are embodied emissions.

Carbon neutral, net-zero, and carbon-negative

When reading sustainability announcements by companies it is important to carefully read the wording used in the announcement.

Carbon neutral means that an organization measures the CO2 emissions caused by its operations, reducing them as far as possible, and offsetting the part that cannot be avoided (e.g. by reforesting, investing in renewable energy, or capturing CO2 from the air). Any company or organization can fairly easily become carbon neutral by replacing fossil energy sources by renewable ones where possible and paying for the remaining CO2 emissions. Even without sustainability investments, it is possible to become carbon neutral by offsetting more. For individuals, buying renewable electricity, driving an electric vehicle, installing a heat pump, and avoiding or offsetting air travel is a good start to become carbon neutral for the operational emissions. Making whole industries carbon neutral is however a totally different challenge.

The gold standard for organizations is however net-zero. This is more challenging as it requires that not only the carbon emissions are taken into account, but also other greenhouse gasses like methane, nitrous oxide, and fluorinated gas emissions. Some fluorinated gasses trap 1000 times more heat than CO2. It is a common practice to convert all greenhouse gasses into CO2 equivalents (CO2e) to simplify the math.

Net-zero emissions do not only apply to the business operations themselves, but over the complete supply chain. This might be feasible in the IT industry, but much harder in e.g. agriculture, which is a major source of methane and nitrous oxide emissions.

Carbon-negative means that a particular device or process extracts more carbon from the air than it emits. Trees are carbon-negative which explains why they are used to offset carbon emissions.

Figure 3 shows that the IT sector will continue to grow, and to use more operational energy in the coming years (despite the fact that devices will continue to be more power efficient). This growth is faster than the rate at which the energy generation is decarbonized globally. This means that the IT related emissions will continue to increase until 2030.

In order to limit global warming to 1.5°C, we need carbon neutrality by 2050, and net-zero in 2070. This will require that all stakeholders work on their own targets, without exceptions. The intermediate goals in 2030 are an instrument to assess whether we are on track or not. According to (United Nations Environment Programme, 2023), we are not at all on track.

Clean electricity to the rescue?

As illustrated with the case of the Apple products, most of the emissions are electricity for (i) manufacturing the devices and (ii) for running the devices. On paper, such emissions can easily be avoided by using clean energy for the production and for the use of the devices. Unfortunately, reality is more complex. Unless the source of the energy is non-intermittent like hydro-electric energy, biomass or nuclear, there is the need to somehow store the energy to bridge the periods when the electricity production is low. Currently, the most reliable storage is the grid, which means that one sells the excess local renewable energy to the grid and buys the electricity back when the local production is too low (like during the night for solar). Obviously, this only works well if the grid company has storage capacity (batteries, pumped storage, …) or can provide fossil-free electricity (nuclear, biomass, hydro-electric). If that is not the case, it will sell fossil-based electricity.

Hence, even if a facility like a data center comes with a renewable energy production system that produces (more than) the total yearly energy consumption, it does not mean that the consumption is 100% clean. It will most of the time be a mix of renewable and non-renewable energy sources. While this energy mix will be less carbon-intensive, it will not achieve carbon neutrality. The effect is even aggravated by the fact that the renewable production on the entire grid might at some point be higher than the consumption, leading to a situation in which the producer has to pay to inject energy into the grid, or that the grid company decides to shut down some installations to protect the grid, meaning that the renewable production system does not run at full capacity. Hence, unless the manufacturing and charging would be limited to the periods when there is enough clean electricity, there will always be a carbon footprint that will be equal to the carbon footprint of the mix of energy sources that are used to power the grid at that time. Hence, replacing all electricity by clean electricity is challenging in practice.

Applying these clean energy insights rigorously may lead to inconvenient conclusions, such as the simplest way to reduce the carbon footprint of data centers being to relocate them to regions with the lowest average carbon intensity for electricity, which is likely in countries with abundant nuclear and hydroelectric power.

Carbon offsetting to the rescue?

The current practice to make products carbon neutral is to first reduce the emissions, and then to offset the emissions that cannot be avoided. A popular way to offset is investing in reforestation projects. A mature tree absorbs between 10-40 kg of CO2 per year. In order to absorb 1000 kg of CO2, a tree has to grow on average 40 years. Hence, offsetting by reforestation means that the CO2 produced today (e.g. a flight) will be absorbed from the air during the next 40 years, i.e., by 2063. This means that the majority of the CO2 offsetted today will keep contributing to global warming during the next 20 years (2043) until the trees are mature enough to fully absorb it. Hence, reforestation does not help to slow down global warming in the next decade, but it does in the future. This assumes that the trees do not die during that period (e.g. by storm, by wildfire, by drought, or by disease). Eventually the trees will die, or be cut, which means that at that moment the stored carbon will be released again. In order to avoid the release of the carbon, a tree that dies or is cut should be replaced by another tree to store the released carbon again. If not, carbon offsetting is only a way to temporarily store the carbon. Hence, carbon offsetting is at best a temporary solution until we find a scalable and affordable technical means to permanently store the captured carbon because the space for reforestation on the planet is finite and it has to compete with agriculture. The most important insight is that offsetting has no immediate effect on the release of CO2 and cannot even guarantee that the CO2 will be removed from the atmosphere in the future. Offsetting is basically a right to pollute and does not stop global warming in the short term. The only way to (immediately) neutralize emissions is to capture them and store them permanently (capture can happen at the source, or from the atmosphere).

Furthermore, companies that try to offset CO2e emissions in order to reach net-zero are sometimes misleading themselves. Fluorinated gasses stay decades to centuries in the atmosphere and are removed by photodissociation in the stratosphere. There is no way to offset them by reforestation. Hence, converting them into CO2e, and subsequently offsetting the corresponding CO2e by reforestation will not help to reach net-zero for these gasses. There is only one effective measure and that is keeping them out of the atmosphere.

Jevons paradox or the rebound effect

Jevons paradox states that more efficient use of resources will not only lead to a smaller ecological footprint, but also to a falling cost of products and services which might in turn lead to an increase in the demand, undoing the benefits of the efficiency gains. This is very noticeable in consumer goods. Due to globalization and technological improvements, consumer goods have become much cheaper than e.g. 40 years ago. In the assumption that the disposable income stays the same, consumption will increase, leading to a growing ecological footprint, and in some cases even bigger than the original one. Where in 1985 people could afford one basic desktop computer of 2000 euro, with the equivalent amount of 5700 euro in 2023, they can buy each member of a family of four a basic laptop, a tablet, a smartphone and a smart watch, inevitably leading to more emissions. Hence, efficiency is not a valid proxy for sustainability.

Another example is the increase in performance thanks to Moore’s law. Instead of reducing the power consumption and the cost per chip in the new technology node, the industry has preferred to produce more powerful chips within the original power envelope and for about the same price. The extra compute power has been used by the software industry to add nice-to-have features that have subsequently increased the requirements for the hardware (more memory, more storage, more cores). In the last decade we have seen a transition from hard disks to SSDs, which consume less power, but have an embodied carbon footprint that is 8 times bigger than that of hard disks. Hence, Moore’s law has substantially increased the carbon footprint of computing up to the point that most consumer devices are today too powerful for the workload they have to run. Similarly, supercomputers seldom have an efficiency of more than 5% because the standard software stack adds a lot of overhead.

Enablement to the rescue

In its SMARTer 2030 report (Global e-sustainability initiative, 2020), the Global eSustainability Initiative states that ICT could save 12 GtCO2e in 2030 in other major industries like manufacturing, agriculture, construction and transport. Compared to the estimated 1.25 GtCO2e emitted by the ICT-industry itself, it would mean that ICT is net carbon negative and that an unconstrained growth of the ICT industry is actually good for the climate: for every additional kg of CO2 emitted by the ICT industry, another 10 kg CO2 will be saved in the rest of the economy, which means that a datacenter of a supercomputer virtually extracts carbon from the air.

This is obviously a flawed reasoning as it counts the emission reduction twice: once in the sector that realizes the reduction, and a second time in the ICT industry.

Furthermore, there is no evidence that enablement leads to carbon-negative emissions. Over the last 40 years, the ICT industry has grown and contributed to efficiency gains in all sectors of the economy, but this has not resulted in reduced carbon emissions at all (the proof if that the emissions are still increasing, despite the massive use of computing technology). The higher efficiency has led to more consumption and increased revenues, but also more pollution and faster depletion of natural resources. Historically, there is no proof that enablement works for reducing overall emissions because of the rebound effect in the sectors that implement digital technologies (Charlotte Freitag, 2021). Still, enablement is widely used to justify digital exceptionalism, which means that the computing industry can be exempted from carbon reduction (in comparison to e.g. the cement and steel industry), because they realize their reduction through enablement. A similar argument was used in the 20th century to exempt airlines from paying fuel taxes because air travel was good to better understand other cultures and thus for world peace.

Until the day that there is proof that enablement actively reduces emissions, the best way for the IT industry to contribute to the net-zero targets in 2070 is to become net-zero itself. If they would lobby for an exception to become net-zero in 2070 based on the enablement they claim to create, they will first have to show that enablement exists for their activities, and how big it is. Without hard evidence, they are lobbying for digital exceptionalism.

Then the question is: how can the IT industry reach net-zero by 2070. We refer to the other papers in the sustainable section for inspiration.

ICT for green

The European commission rightfully states that the green transition will require ICT to be successful. This is a no-brainer because optimization of resource consumption is at the basis of the green transition and advanced optimization requires computing. The statement that the green transition requires ICT is however not equivalent with the statement that ICT is good for sustainability (which is the assumption behind enablement). More concretely, it is not because some use cases in sectors like transport or energy can be made more sustainable with technologies like IoT, 5G or AI, that a widespread adoption of these technologies in the economy will have the same beneficial impact at a global scale. There is no proof of that.

Sustainability benefits of ICT will have to be proven use case per use case by comparing the scenario in which the technology is deployed, with the scenario in which the technology is not deployed or replaced by alternatives. The study should take into account the full life cycle, including all wanted and unwanted consequences of the technology used. An example is the self-driving car. The full optimization of the driving process will definitely have direct sustainability benefits: systematic eco-driving, congestion avoidance, less accidents. Combined with new business models like mobility as a service, it could also reduce the number of registered cars (which is good for the embodied energy, and requires less parking space, which could lead to more green spots in cities). Unwanted consequences could be that also children, very old people, and people without driving license can start using a car, leading to more car-km per year, and hence more energy consumption, or that people start working during their commute, and do no longer care about the duration of the trip (and hence about the jams). Before claiming that a self-driving car is more sustainable than a classical car, this analysis should be made.

Obviously, decisions are not only made based on sustainability criteria – there might be other criteria (safety, inclusion, increased productivity) that can justify the additional emissions. Based on such an analysis, it will be hard to prove that bitcoin calculations reduce the global carbon emissions (this seems obvious), or that the energy-hungry large language models will help reducing the global carbon emissions (this would be surprising, except for a few use cases). Other technologies like IoT and 5G might help reducing the carbon emissions in more use cases. Other cases are unclear: videoconferencing is always more sustainable than in-person meetings that require travel, but the convenience of videoconferencing has also replaced traditional phone calls and some intra-building business meetings leading to increased energy use (rebound effect).

Without additional legislation, a growth of the IT industry will not automatically lead to sustainability benefits. Sustainability benefits have to be proven case by case. Approaches like carbon accounting and Environmental, Social, Governance (ESG) reporting can help companies to understand their environmental impact, how to reduce it, and to prove that their IT solutions help in greening their company. It also guarantees that the reporting is standardized and can be used for benchmarking purposes. Eventually, this information could also be made available via the digital product passport (DPP) that might play a pivotal role in ending greenwashing practices within the industry. This information could further be translated into sustainability labels designed to assist consumers in making informed choices and increasing awareness about sustainability. These labels should be designed to be easily understandable and comparable to labels commonly used on food products or the euro-norms for car emissions.

According to ISS Insights (Aastha Agarwal, 2023), 57% of the European companies pledged net-zero commitment by 2050. In countries with a strong political net-zero commitment, more companies pledge to become net-zero by 2050, which proves that public policies help in making the industry more sustainable. Remarkably, the Information technology sector represents only a very small share of the companies that pledge net-zero (Figure 6).

Figure . Companies with net-zero commitments, by sector. Source: (Aastha Agarwal, 2023)

According to IDC, the most common investments in sustainability are:

  1. Investment into technologies empowering sustainable hybrid work environment (VPN, video conferencing, advanced security, etc.);

  2. Use of recycled/recyclable materials in IT equipment/use of closed loop materials in IT;

  3. IT infrastructure efficiency assessment and investments (energy-efficient devices);

  4. Smart building energy efficiency related technologies and tools;

  5. Carbon footprint/emissions tracking, data sourcing and data analysis;

  6. Flexible consumption technology deployment models to eliminate resources overprovisioning.

The concrete order varies among sectors and among countries. According to the same study, the sustainability ambitions are greater than the sustainability actions. Only 50% of the companies have actually achieved the goals set forward for the last two years. There are multiple explanations: gap between strategic planning in the boardroom and execution on the work floor, difficulties implementing the ESG commitments, macroeconomic challenges (energy prices, inflation, political instability, …), and the difficulty to present a clear business case with a reasonable return on investment for sustainability investments. In many companies, the sustainability transition requires a substantial change to stay in business. This is why sustainability is perceived as a risk management issue and compliance cost instead of business opportunity.

Proving the business case for sustainability is critical for widespread adoption, and ultimately the success of the European Union Green Deal. When sustainability is seen as a driver for business growth, it will be a door opener, getting traction from the top to the bottom of an organization. The EU or national governments could help by refocusing macroeconomic success away from GDP growth towards non-financial metrics to assess the well-being of the country.

Companies are struggling with preparing the workforce for a net-zero economy. Whilst it is true that the economic transition is creating tremendous new job opportunities, it also creates huge challenges in the short term. E.g., during the transition, transportation companies will have to service both and existing fuel-based fleet + a more recent electrical fleet (which requires a different skill set, different tools, often a different workshop).

Often forgotten: the importance of water

There is no electrical energy that is 100% emissions-free, but there are big differences between energy sources (Figure 7). Obviously, the use of fossil fuels causes most emissions, but even reservoir lakes have greenhouse gas emissions too due to decomposing organic material that ends up in the lake (e.g. methane).

Figure . CO2e emissions different electricity sources (in g per kWh)

In order to produce electricity, one does not only need fuel, but also water to (i) create the infrastructure (embodied water), and (ii) to close the thermodynamic cycle. Figure 8 gives an idea of the water footprint of the different types of energy generation. The Water Footprint is defined as the amount of water that is no longer available as drinking, irrigation, or process water after it has left de power plant. In most cases it is water that evaporates from cooling towers, or from reservoir lakes.

A nuclear power plant of 1 GW produces around 4TJ of energy per hour and needs 610x4=2440 m³ of water per hour, which is about the capacity of an Olympic swimming pool per hour (2500 m³). Coal and gas need 404x4=1760 m³ per hour. Firewood has a similar water footprint for the power plant but needs and enormous amount of water to grow the trees (Fuel supply). The high operational water consumption of hydropower might seem logical but the water that leaves the turbine is not ‘consumed’, it can be used downstream for irrigation, drinking or process water. The 15,100 m³ is the amount of water per TJ that evaporates in the lake between the time it is collected (spring) and the time it is used to generate electricity. A lake evaporates around 1 m water per year (up two 2 m in deserts) which amounts to an average of 15,100 m² per TJ generated energy. The water footprint of solar is mostly determined by its embodied water. This is a consequence of the fact that one panel represents a lot of embodied water but produces during its lifetime only 5-10 MWh. A nuclear power plant produces the same amount of energy in 18-36s, and a windmill in 1h. The preparation of uranium, oil and coal requires lots of water too. The embodied water in the infrastructure and the fuel supply is extracted at the time and place of construction, the embodied water of the fuel is extracted near the mining site or in the refining factories. Only the operational water is spent at generation time and place.

One thing is clear: there is no electricity without water. Furthermore, the water consumption for electricity production is projected to double between 2010 and 2030 due to the increased electricity production, but also by investments in renewable energy sources like hydropower and biomass (firewood, energy crops, …) that have a larger water footprint. Extended periods of drought might in the future limit the options to produce electricity in some parts of the world.

Data centers do not only need electricity to power them, but also need large amounts of water to cool them (and in the absence of water, they need electricity). In 2023 several data centers did not obtain a permit – because either the grid could not guarantee the power that was needed or because the water needed to cool the facility was not available. The water discussion is especially important in arid areas: Microsoft could not obtain a permit to use water cooling in his new data center in Phoenix Arizona in the middle of the desert.

Not only data centers, but also fabs use lots of water. Today, a semiconductor fab uses around 8 l/cm² of ultrapure water (Wang, 2023), which means that a fab producing 20 000 wafers of 200 mm per month, needs more than one Olympic pool of process water per day. TSMC had to abandon a plan to build a 2 nm fab in a science park in the north of Taiwan after protest from the local residents about the space needed, and the amount of water it was going to consume (Mann, 2023).

Figure 8. Water footprint of electricity generation (Mesfin M. Mekonnen, 2015)

Ireland is data center capital of Europe (O'Halloran, 2021) with 70 data centers in operation in 2021 (consuming 11% of the national electricity production). There will be more than 100 by 2025, and they are projected to consume 27% of the electricity production of the island. Adding one datacenter is like adding an extra town to the grid. Countries have to reduce their carbon emissions by 42% by 2030. It is clear that adding data centers is not helping them to reach their targets, and that governments prefer to invest in electricity generation for their citizens (to power their cars and heat pumps) rather than power data centers from large hyperscalers. Similar discussions take place in The Netherlands and Singapore.

What do to with old devices?

The prevailing business models are rooted in the linear economic paradigm, which follows the take-make-consume-throw away cycle. Under this model, the more products a company produces and sells, the more successful it is perceived to be, benefiting the economy and society. This approach heavily relies on the availability of abundant and affordable raw materials and energy. However, as resources become scarcer, costs rise, and the environmental impact of waste becomes unsustainable (as exemplified by single-use plastics), there arises a pressing need to reduce the consumption of raw materials and energy. Additionally, waste should be viewed as a potential source of raw materials. Transitioning away from this linear model necessitates the development of new business models, based on a waste hierarchy. A waste hierarchy is a framework to reason about the end-of-life processing of devices. Europe has a Waste Framework Directive which is currently being revised. It consists of five levels (Figure 9).

Figure . Waste hierarchy

Prevention means that products are designed in ways to reduce waste: less pollution, less energy consumption, less harmful products or processes, less packaging, easier end-of-life processing. The European standardization of chargers is a good example of prevention.

Re-use can mean many things: devices can be sold as second-hand devices – even at a large scale by leasing companies; devices can be repaired – provided that they are repairable and spare parts are available; devices can be refurbished or remanufactured, i.e. updated to the current standards, or they can be repurposed, i.e., given a second life in a less demanding application context.

Recycling means that the materials of which the device is made are reinjected as raw materials in the industrial supply chains to create new products. Some fractions of e-waste are straightforward to recycle: glass, metals, plastics. Others like silver, gold, palladium and copper are more challenging but still economically viable. The semiconductors consist mostly of silicon with very small amounts of minerals deposited on it. Consequently, the silicon becomes 'contaminated' with these minerals, and their concentration is too low for economic (and ecological) extraction. Therefore, a die is an example of a non-circular product—it cannot be recycled at the end of its life. Hence, currently, the components that cause the majority of the emissions unfortunately cannot be recycled (which means that they cannot be part of the circular economy). This is not the only problem in this stage: globally, only a small fraction (17.4%) of the e-waste is currently being collected, and hence not recycled at all.

Recovery involves extracting energy from the fraction of the device that cannot be recycled (some plastics, glue, …) by e.g. incinerating or gasifying it, and using the heat or gasses in a useful application. Incineration can be challenging if the smoke contains toxic components that could pollute the environment, but it has the advantage that it also reduces the volume of the waste, and the ashes could be used in some applications (e.g. in construction).

Disposal involves storing the output of the recovery phase permanently into a landfill.

Currently, the world produces 53 million tonnes of electronic waste, or e-waste, and this will increase to 75 million tonnes by 2030. In 2019, only 17.4% of it was processed in recycling facilities. The remaining e-waste was either shipped to low or middle-income countries (where it often pollutes the environment), was thrown in the garbage bin, or kept at home for potential future use. In order to be sustainable, devices should be returned as soon as they are no longer used just like cars are returned when a new car is bought. This increases the chance on re-use and guarantees that the device will be properly recycled.

The most challenging level in the waste hierarchy, and the Achilles heel of the circular economy is recycling. Reuse can fairly easily be done by creating a market for second hand products (which is the norm for vehicles, houses, and other expensive goods), and recovery is standard in many western countries (incineration). Recycling is challenging however, not only because only a small percentage is being recycled, but also because the recycling technology for e-waste can only extract a limited number of raw materials from the e-waste. Ideally, there raw materials should be used to create new products. Unfortunately, we are far from this in the electronics industry. Very often the extracted materials do not have the same chemical properties to be used again in the electronics industry where ultrapure materials are required. They are therefore used in different sectors, just like recycled paper often ends up in cardboard. The fact that they are used in other industries is of course excellent and must be encouraged, but it makes the electronics industry non-circular because it remains dependent on freshly mined materials, some of which are in limited supply, and are used as geopolitical weapons by some countries. Hence, in order to become more sustainable, the electronics industry should invest in more effective recycling and try to become more circular.

Sustainability and time

Sustainability is now everywhere. As soon as an organization reduces its water or electricity consumption it calls itself sustainable. This is obviously incorrect. It is only sustainable if all the water and electricity consumption is sustainable. Even the expression that the organization has become ‘more sustainable’ is misleading as it suggests that it was already sustainable and now has become even ‘more’ sustainable. It would be better to state that the environmental impact or footprint has been reduced (which is very valuable in itself, but it is not synonymous with being sustainable, which can only be the ultimate end goal).

The ultimate goal for carbon emissions is to become carbon neutral by 2050 (and net-zero by 2070), and thereafter remain sustainable and net-zero forever. There is an intermediate checkpoint in 2030 to verify whether the joint global efforts are on track. Some people seem to believe that these deadlines are a kind of payment deadline by which a sum of money has to be paid back, and as soon as the debt has been paid, they can continue business as usual. This is obviously not the case. Now we have to work on a reduction of 43% by 2030, and after that we will have to work on the remaining 57% (which will probably be the hardest part).

Hence, becoming sustainable is a transition process in which the environmental footprint is gradually reduced until the moment that it is within the natural absorption capacity of the planet. For CO2, this requires that we will have to stop emitting CO2 from burning fossil fuels (either by stop using them, or capturing the emissions, and use them or store them). Hence, reducing the emissions of a process that emits 1000 kg CO2 per day with 200 kg (per day) by adding a computing device with a carbon footprint of 20 kg per day is definitely good because it structurally reduces the total emissions from 1000 kg to 820 kg CO2 from then on. The next question is how the remaining 820 kg can be further reduced to eventually be low enough that the remaining emissions are absorbed by natural processes. Hence, becoming sustainable is a continuous process.


The computing industry has an environmental footprint which will be growing during the next decade. In order to make the industry carbon neutral, and eventually net-zero, more efforts will be needed. For the time being the two most important recommendations are to work on practical and universally accepted life cycle models for computing devices and services. Without such models, it is impossible to explore the design space for the most sustainable solution. In parallel, there is a need to educate the computing community on sustainability and to encourage it to start working on it and to make real progress.


The authors would like to thank Katharina Grimme and Angela Salmeron of IDC for their market insights.


Koen De Bosschere is a professor in the electronics department of Ghent University, Ghent, Belgium.

Patrick Blouet has been research and development program manager at STMicroelectronics with responsibility for European projects. For more than 25 years, he has been interested in environmental issues as a personal concern.


[1]: Aastha Agarwal, S. S. (2023, 5 16). Net Zero Pledges in Europe. Retrieved from ISS Insights:
[2]: Charlotte Freitag, M. B.-L. (2021, September). The real climate and transformative impact of ICT: A critique of estimates, trends, and regulations. Patterns, 1-18.
[3]: D.H. Meadows et al. . (1972, March 2). The Limits to Growth. Retrieved from
[4]: Global e-sustainability initiative. (2020). ICT Solutions for 21st Century Challenges. Global e-sustainability initiative (GeSI).
[5]: Grober, U. (2007). Deep roots – A conceptual history of ‘sustainable development’ (Nachhaltigkeit). Retrieved November 27, 2022, from
[6]: Mann, T. (2023, October 17). TSMC abandons plans for 2nm chip plant after Taiwanese locals protest. Retrieved from The Register:
[7]: Mesfin M. Mekonnen, P. W.-L. (2015, 3 9). The consumptive water footprint of electricity and heat: a global assesment. Environmental Science: Water Research & Technology, 1, 285-297.
[8]: O'Halloran, M. (2021, September 29). Ireland should pause development of data centres. The Irish Times.
[9]: SEI, Climate Analytics, E3G, IISD, and UNEP. (2023). The Production Gap: Phasing down or phasing up? Top fossil fuel producers plan even more extraction despite climate promises. . Stockholm Environment Institute.
[10]: The 17 Goals. (2015). (United Nations) Retrieved November 27, 2022, from
[11]: Triple bottom line. (2008, November 17). (The Economist) Retrieved November 26, 2022, from
[12]: United Nations Environment Programme. (2023). Emissions Gap Report 2023: Broken Record – Temperatures hit new highs, yet world fails to cut emissions (again). Nairobi. doi:
[13]: Vaclav Smil. (2022). How the world really works. Penguin books.
[14]: Vincent Petit, S. C. (2021). Digital economy and climate impact. A bottom-up forecast of the IT sector energy consumption and carbon footprint to 2030. Schneider Electric Sustainability Institute.
[15]: Wang, Q. (2023). Environmental data and facts in the semiconductor manufacturing industry: An unexpected high water and energy consumption situation. Water Cycle(4), 47-54.

The HiPEAC project has received funding from the European Union's Horizon Europe research and innovation funding programme under grant agreement number 101069836. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them.