Green Computing - Can Technology Help Save the Planet?

Here is an uncomfortable truth that the technology industry has been slow to confront.

The devices and digital services that feel clean, invisible, weightless, frictionless, are not. Every Google search, every streamed video, every cloud-stored photograph, every AI-generated response requires physical infrastructure: servers that run hot and consume electricity, cooling systems that consume more, transmission networks that span continents, and manufacturing processes that extract rare materials from the earth in ways that leave lasting scars.

The internet, as a whole, currently accounts for approximately 3 to 4% of global greenhouse gas emissions, roughly equivalent to the aviation industry (Freitag et al., 2021). That number is not falling. The explosion of AI, streaming, cloud computing, and connected devices is driving it upward. And the infrastructure required to power the AI revolution in particular, data centers consuming gigawatts of electricity, training runs that emit hundreds of tons of carbon dioxide, have become one of the most significant and least discussed environmental stories of the decade.

This is the honest starting point for a conversation about green computing. Not that technology is the enemy of sustainability, it is not. Technology is, simultaneously, one of the most significant contributors to our environmental challenge and one of the most powerful tools we have for addressing it. The question is whether the latter can outrun the former, and what it will take to make that happen.

This article examines that question from every angle, what green computing is, where the environmental costs of technology actually lie, what the industry is doing about it, what genuine progress looks like, and what individuals and organizations can do to participate in a more sustainable digital future.

What Green Computing Actually Means

Green computing, also called green IT or sustainable computing, refers to the design, manufacture, use, and disposal of computing resources in ways that minimize environmental impact. The term encompasses a wide range of practices and technologies, from the chip architecture inside your laptop to the energy source powering the data center processing your emails to the recycling programme that handles your old smartphone.

It is useful to think of green computing as operating at four levels.

Hardware design, engineering devices and components that consume less energy, use less environmentally harmful materials in their manufacture, and are designed for longevity, repairability, and eventual recycling rather than planned obsolescence.

Data center operations, powering and cooling the physical infrastructure of the internet and cloud computing as efficiently and sustainably as possible, ideally using renewable energy sources.

Software efficiency, writing code and designing systems that accomplish their tasks with the minimum necessary computational resources, rather than treating compute as effectively free and limitless.

Circular economy practices, managing the full lifecycle of electronic devices, from responsible sourcing of raw materials through extended product life through end-of-life recycling, in ways that recover valuable materials and prevent toxic waste.

Each of these levels matters. And each is currently a mixture of genuine progress, inadequate effort, and significant remaining challenge.

The Environmental Cost of Technology: The Full Picture

Before examining solutions, the problems need to be clearly understood. The environmental footprint of information technology is larger, more complex, and more unevenly distributed than most people realize.

The Energy Hunger of Data Centers

Data centers, the physical facilities that house the servers, storage systems, and networking equipment powering the internet and cloud services, are prodigious consumers of electricity. Globally, data centers consume approximately 200 to 250 terawatt-hours of electricity per year, representing roughly 1% of global electricity demand (IEA, 2024).

That figure has remained relatively stable over the past decade, despite exponential growth in data traffic and computing workloads, a testament to significant efficiency improvements in data center design and operation. But the AI boom is now straining stability severely.

The AI training runs that produce large language models are extraordinarily energy intensive. Training GPT-3, OpenAI's 2020 model, was estimated to consume approximately 1,287 megawatt-hours of electricity and produce approximately 552 tons of CO₂ equivalent (Patterson et al., 2021). The training runs for more recent, larger models are believed to be significantly more energy-intensive, though the major AI labs have become less forthcoming about publishing this data.

AI inference, running the trained models to respond to user queries, may ultimately represent a larger total energy burden than training, simply because inference happens billions of times per day across millions of users, while training happens once. Microsoft's own environmental report disclosed that its global water consumption increased by 34% between 2021 and 2022, largely attributable to the cooling requirements of AI infrastructure (Microsoft, 2023). Google similarly disclosed a 48% increase in greenhouse gas emissions between 2019 and 2023, explicitly attributing the increase to data center energy consumption driven by AI workloads (Google, 2024).

These are not small numbers, and the trajectory is upward.

The Carbon Cost of Manufacturing

The environmental cost of computing is not limited to operational energy consumption. Manufacturing hardware, extracting raw materials, fabricating chips, assembling devices, and shipping products globally, carries a substantial carbon cost that is often omitted from sustainability discussions focused primarily on energy use.

For many consumer devices, the manufacturing carbon footprint is larger than the operational footprint over the device's lifetime. Apple has disclosed that manufacturing accounts for approximately 70% of the total lifecycle carbon footprint of an iPhone (Apple, 2023). This means that buying a new phone has a significantly larger carbon impact than any reasonable amount of phone use, which has profound implications for how we think about sustainable technology consumption.

The manufacturing of semiconductor chips, the processing units at the heart of all computing devices, is particularly resource intensive. Chip fabrication requires ultrapure water in enormous quantities, specialized chemicals, rare earth elements, and enormous amounts of energy. TSMC, the world's largest chip manufacturer, consumed approximately 22 billion litres of water in 2022 (TSMC, 2023).

The extraction of the rare earth elements and critical minerals that electronics require, lithium, cobalt, neodymium, tantalum, and others, involves mining operations with significant local environmental impacts, and supply chains with serious human rights concerns in some regions.

The Electronic Waste Crisis

At the end of their lives, electronic devices become waste, and electronic waste, or e-waste, is the world's fastest-growing waste stream. The Global E-waste Monitor reported that 62 million tons of e-waste were generated globally in 2022, of which less than 23% was formally collected and recycled (Forti et al., 2023).

The remaining 77%, containing toxic materials including lead, mercury, cadmium, and brominated flame retardants, end up in landfill, is incinerated, or is shipped to low-income countries where informal recycling exposes workers and communities to serious health hazards while the valuable materials are incompletely recovered.

At the same time, e-waste contains significant concentrations of valuable metals, gold, silver, copper, and palladium, that are worth recovering. One ton of circuit boards contains more gold than one ton of gold ore. The failure to recover these materials is both an environmental failure and an economic one.

What the Industry Is Actually Doing

The response from the technology industry to its environmental footprint ranges from genuine and significant to greenwashed and inadequate. Distinguishing between the two requires looking beyond headline commitments to specific, verifiable actions.

The Renewable Energy Transition

The most significant and well-documented environmental improvement in data center operations has been the shift toward renewable energy. The major hyperscale data center operators, Google, Microsoft, Amazon (AWS), Meta, and Apple, have collectively made substantial investments in renewable energy procurement, through a combination of on-site generation, power purchase agreements with renewable energy projects, and renewable energy certificates.

Google has been purchasing enough renewable energy to match 100% of its global electricity consumption since 2017 and has committed to operating carbon-free energy 24/7 by 2030, meaning that at every hour of every day, the electricity powering its facilities comes from carbon-free sources (Google, 2024). This is a significantly more rigorous commitment than simply purchasing annual renewable energy credits, which can mask the reality that data centers often run on fossil fuel power at night or during periods of low renewable supply.

Microsoft has committed to being carbon negative by 2030 and to removing all the carbon it has ever emitted by 2050 (Microsoft, 2023). Amazon has committed to 100% renewable energy by 2025 and net-zero carbon by 2040 (Amazon, 2023).

These commitments are meaningful. They are also conditional on credible implementation and genuine accounting, and the disclosure of material increases in emissions alongside these commitments raises legitimate questions about whether stated goals are achievable on the announced timelines.

For smaller companies and organizations without the leverage to negotiate power purchase agreements with renewable energy developers, the renewable energy transition is slower and more dependent on the decarbonization of national electricity grids, a process that is proceeding at different rates in different countries.

Data Centre Efficiency: The Power Usage Effectiveness Story

The efficiency of data centers, measured by a metric called Power Usage Effectiveness (PUE), which compares total facility energy consumption to the energy consumed by the computing equipment itself, has improved dramatically over the past two decades.

A PUE of 1.0 would mean perfect efficiency: every watt entering the facility is used for computing. A PUE of 2.0 means that for every watt used for computing, another watt is used for cooling, lighting, and other overhead. The global average PUE for data centers in 2007 was approximately 2.5. By 2023, the average for hyperscale data centers operated by major cloud providers had fallen to approximately 1.2 to 1.3 (Uptime Institute, 2023).

This improvement has been achieved through advances in cooling technology, server density, hot-aisle and cold-aisle containment, waste heat recovery, and the shift from less efficient on-premises data centers to more efficient hyperscale facilities. Google's most efficient data centers have achieved PUE values approaching 1.06, meaning that overhead energy consumption is only 6% of computing energy consumption (Google, 2024).

The efficiency improvements are genuine and significant. They have, however, been largely absorbed by the growth in computing demand rather than translating into reduced total energy consumption, a pattern sometimes called the Jevons paradox, where efficiency improvements enable increased consumption that offsets or exceeds the efficiency gains.

The Software Efficiency Frontier

One of the most significant and least discussed levers for reducing the environmental impact of computing is software efficiency, writing code that accomplishes its purpose with the minimum necessary computational resources.

The computing industry has spent decades subsidizing software inefficiency with cheap hardware. When a poorly optimized piece of software makes a programme run twice as slow, the solution has typically been to buy faster hardware rather than optimize the code. This approach has hidden the environmental cost of software sloppiness behind the falling price of compute.

As environmental costs become more salient, whether through carbon pricing, energy costs, or reputational pressure, the incentives for software efficiency are changing. Research has demonstrated that the energy consumption of different software implementations of the same function can vary by orders of magnitude, factors of 10, 100, or more, depending on the programming language, the algorithm, and the implementation choices (Pereira et al., 2017).

Efficient algorithms that accomplish the same result with fewer computations, programming languages and runtimes optimized for energy efficiency, machine learning model architectures that achieve equivalent accuracy with fewer parameters and less training compute, these are all areas of active research and development with significant potential environmental impact.

Microsoft's research into "carbon-aware computing", scheduling computationally intensive workloads to run when and where renewable energy is most abundant, is one practical example of how software can be designed to minimize carbon impact rather than simply minimizing latency (Wiesner et al., 2021).

The Right to Repair and Longevity Movement

One of the most structurally significant changes in the green computing landscape is the growing legal recognition of consumers' right to repair their own devices, and the associated pressure on manufacturers to design products that last longer and can be maintained and repaired rather than replaced.

The European Union's Ecodesign Regulation, which came into force progressively from 2021, requires manufacturers of electronics including smartphones, laptops, and tablets to make spare parts available, provide repair information, and design products that can be opened and repaired without specialized tools. The EU's Right to Repair Directive, adopted in 2024, extended these requirements and gave consumers the right to have their devices repaired at reasonable cost (European Parliament, 2024).

Frameworks Software pioneer Fairphone has demonstrated that designing a modular, repairable smartphone, with user-replaceable components including the battery, screen, camera, and charging port, is commercially viable, with their devices achieving average lifespans significantly longer than conventional smartphones (Fairphone, 2023).

The Framework laptop, a modular, repairable laptop designed for component-level upgradability and repair, has achieved significant commercial success and critical acclaim, demonstrating market appetite for sustainable hardware design beyond the small ethical consumer niche.

Extending the average life of a device by even one year reduces its total lifecycle carbon footprint significantly, because manufacturing, as noted above, dominates the lifecycle impact. A phone used for four years instead of two approximately half of the annualized manufacturing carbon footprint.

E-Waste: The Circular Economy Response

The formal recycling and circular economy response to e-waste is slowly improving, driven by a combination of extended producer responsibility legislation, industry take-back programmers, and growing market value for recovered materials.

Apple's Daisy robot, an automated disassembly line capable of processing up to 1.2 million iPhones per year to recover materials including aluminum, rare earth elements, and cobalt, is one of the most visible examples of industry investment in circular economy practices (Apple, 2023). Apple has stated a goal of eventually manufacturing products entirely from recycled or renewable materials.

The EU's updated Waste Electrical and Electronic Equipment (WEEE) Directive, and equivalent legislation in an increasing number of countries, places responsibility on manufacturers for the end-of-life management of their products, creating financial incentives to design for recyclability and fund take-back programmes.

The challenges remain significant. The informal e-waste sector, handling the majority of global e-waste, operates outside formal regulatory frameworks, typically in low-income countries, with serious health consequences for the workers involved. International coordination on e-waste governance is improving but remains insufficient given the scale of the problem.

Technology as a Climate Solution: The Other Side of the Equation

Having examined technology's environmental costs honestly, the other side of the ledger deserves equal attention. Technology is not only a consumer of resources and emitter of carbon, but also one of the most powerful tools humanity has for addressing the climate challenge.

AI-Accelerated Climate Science

Machine learning is transforming climate science into ways that have direct implications for our ability to understand, predict, and respond to climate change.

The most striking example is DeepMind's GraphCast, a weather prediction model that produces ten-day global weather forecasts in under sixty seconds, with accuracy superior to conventional numerical weather prediction models that require hours of supercomputer time (Lam et al., 2023). More accurate weather prediction has direct economic value and enables better management of renewable energy systems whose output depends on weather conditions.

AI models are also being applied to climate modelling at larger timescales, improving the accuracy and resolution of climate projections that inform policy decisions about emissions reductions and adaptation investments. The ability to run higher-resolution climate models faster enables a more granular understanding of regional climate impacts that is essential for effective adaptation planning.

Optimizing Renewable Energy Systems

The integration of large amounts of renewable energy, solar and wind, into electricity grids is significantly more challenging than managing the steady output of conventional power plants, because renewable generation is variable and not perfectly predictable. AI and machine learning play an increasingly important role in grid management, forecasting renewable output, optimizing energy dispatch, and matching supply with demand in real time.

Google's DeepMind demonstrated a 30% reduction in the energy used for cooling at Google data centers through AI-optimized control of cooling systems, and the same approach is being applied to electricity grid management more broadly (Evans and Gao, 2016). The potential for AI to improve the efficiency of renewable energy integration at scale is significant and represents one of the clearest cases where the environmental cost of AI infrastructure is likely to be outweighed by the climate benefit it enables.

Precision Agriculture and Land Use

Agriculture accounts for approximately 18 to 20% of global greenhouse gas emissions, including methane from livestock, nitrous oxide from fertilizers, and carbon dioxide from land clearing (IPCC, 2022). AI and satellite technology are enabling precision agriculture approaches that significantly reduce the resource inputs required to produce food.

Satellite imagery combined with machine learning can identify crop stress, soil moisture levels, and nutrient deficiencies at field level, enabling targeted application of fertilizer and water rather than blanket application across entire fields. This reduces fertilizer use, and the associated nitrous oxide emissions, which have a global warming potential nearly 300 times that of CO₂, while maintaining or improving yields.

The potential for smart agriculture to reduce the environmental footprint of food production is substantial. A comprehensive analysis estimated that digital agriculture technologies could reduce agricultural greenhouse gas emissions by 20 to 30% by 2030 if deployed at scale (Shepherd et al., 2020).

Materials Discovery and Clean Energy Technology

The development of new materials, for better batteries, more efficient solar cells, superconductors, and carbon capture systems, is one of the most important bottlenecks in the clean energy transition. Traditional materials discovery is slow, expensive, and dependent on human expertise. AI is dramatically accelerating it.

Google DeepMind's GNoME model, announced in 2023, discovered approximately 2.2 million new crystal structures, of which 380,000 were identified as stable and potentially useful, expanding the known universe of stable inorganic materials by nearly tenfold (Merchant et al., 2023). The implications for materials science relevant to clean energy, battery materials, photovoltaics, superconductors, are potentially transformative.

Microsoft and others are using quantum computing and AI to simulate molecular and atomic interactions at scales that conventional computing cannot manage, with the goal of discovering materials for hydrogen fuel cells, carbon capture, and next-generation batteries that could transform the economics of clean energy storage.

What Organizations Can Do

For businesses and organizations seeking to reduce the environmental footprint of their technology operations, a practical framework involves several interconnected decisions.

Measure before you manage. The carbon footprint of IT operations, including Scope 3 emissions from cloud services and hardware manufacturing, is invisible until it is measured. Tools including the Green Software Foundation's Software Carbon Intensity specification and cloud provider carbon dashboards (available from AWS, Azure, and Google Cloud) enable organizations to quantify and track their digital carbon footprint.

Choose cloud over on premises for most workloads. Hyperscale cloud data centers are significantly more energy-efficient than the typical on-premises server room and have substantially higher renewable energy procurement. For most organizations, migrating workloads to cloud infrastructure reduces rather than increases carbon footprint, despite the common assumption that "the cloud" is environmentally problematic.

Select cloud regions powered by renewable energy. Not all cloud regions are equally green. Google, AWS, and Azure all publish carbon intensity data for their regions. Scheduling workloads in regions with lower carbon intensity or enabling carbon-aware scheduling tools that automatically route workloads to cleaner regions, can reduce the carbon footprint of cloud computing significantly.

Extend hardware lifecycles. The environmental case for extending the life of computing hardware is strong. Establishing laptop lifecycle policies for four or more years rather than three, refurbishing rather than replacing devices where possible, and choosing hardware from manufacturers with strong sustainability credentials and repairability commitments all contribute meaningfully.

Design software for efficiency. Organizations that develop software should incorporate energy efficiency into their engineering standards alongside performance and reliability. Choosing efficient algorithms, avoiding unnecessary computation, and optimizing resource utilization reduces both cloud costs and carbon footprint simultaneously, making efficiency a business case as well as an environmental one.

Implement responsible e-waste disposal. Establishing certified e-waste recycling channels for all retired hardware, rather than allowing devices to enter general waste streams, ensures that valuable materials are recovered and toxic materials are safely managed.

What Individuals Can Do

Individual choices, aggregated across billions of people, matter, even if systemic change matters more.

Keep your devices longer. The most significant individual action for reducing your personal technology carbon footprint is simply using your devices for longer before replacing them. A smartphone is used for four years instead of two roughly half the annualized manufacturing emissions. A laptop used for six years instead of three does the same.

Choose reputable refurbished devices. Buying a well-refurbished device from a reputable supplier carries a fraction of the carbon footprint of buying new. The refurbished electronics market has matured significantly and now offers reliable devices with warranty coverage across most product categories.

Use streaming more efficiently. Streaming video is a significant personal data footprint driver. Reducing video resolution when high resolution is not necessary, background music does not need 4K video and downloading content for offline viewing rather than repeatedly streaming reduces the energy demand of your media consumption.

Enable energy-saving modes. Power management settings on devices, sleep modes, display brightness, processor performance limits, have real effects on energy consumption. Enabling aggressive power management on devices you are not actively using costs nothing and saves meaningfully on scale.

Search for greener AI options. Not all AI services have the same carbon footprint. Some AI providers are more transparent about their energy sourcing and more committed to renewable energy procurement than others. For AI-intensive workflows, choosing services from providers with credible sustainability commitments matters.

Participate in and advocate for take-back programmes. Using manufacturer and retailer take-back programmes for old devices, advocating for right to repair policies, and supporting extended producer responsibility legislation are the most impactful ways individuals can influence systemic change.

The Honest Assessment: Are We Winning or Losing?

The honest answer to whether technology is helping save the planet is: both, simultaneously, with the outcome uncertain.

On the cost side: the energy consumption of AI infrastructure is growing faster than the renewable energy transition is proceeding, at least in the near term. The e-waste problem is worsening. The manufacturing footprint of consumer electronics remains enormous and is only partially offset by recycling and longevity improvements. The gap between corporate sustainability commitments and verified corporate sustainability outcomes is real and significant.

On the benefit side: the renewable energy transition in data centers is genuine and accelerating. AI is making meaningful contributions to climate science, renewable energy optimization, materials discovery, and precision agriculture that could, in aggregate, deliver climate benefits that substantially outweigh the emissions associated with AI infrastructure. Efficiency improvements in hardware and software continue. Regulatory pressure on product longevity, repairability, and e-waste is strengthening globally.

The International Energy Agency's assessment is that digital technology could deliver energy savings equivalent to 15% of current global energy demand by 2030, if deployed at scale and accompanied by appropriate policy frameworks (IEA, 2024). That is a significant positive. It is also conditional on choices, by industry, by policymakers, and by consumers, that are not yet guaranteed.

The technology industry has more power than almost any other sector to shape its own environmental trajectory. It is also, for the first time, under sufficient scrutiny that the gap between stated commitments and actual performance is becoming visible. That visibility, uncomfortable as it is for industry, is a necessary precondition for accountability.

The Bottom Line

Technology is not inherently green. It is not inherently destructive either. It is a set of tools and systems whose environmental impact is determined by the choices made at every level of the stack, from how a chip is designed to how a data center is powered to how long a consumer keeps their phone to how software is written and deployed.

The case for optimism is real. The renewable energy transition in major data centers is happening. AI is genuinely accelerating clean energy and climate science in ways that matter. Regulatory pressure on hardware sustainability is increasing. And the growing public and investor scrutiny of technology companies' environmental claims is creating accountability where previously there was largely self-reporting.

The case for urgency is equally real. The AI infrastructure buildout represents a material risk to climate commitments if it is powered by fossil fuels during a critical decade for global emissions. The e-waste crisis is not improving at the rate required. And the gap between what is technically and economically possible in sustainable computing and what is being done remains large.

Green computing is not a niche concern for environmental specialists within the technology industry. It is a central challenge for an industry whose products and infrastructure now underpin virtually every dimension of modern life, and whose choices about energy, materials, longevity, and efficiency will be a significant factor in whether humanity meets its climate commitments in the decades ahead.

Can technology help save the planet? Yes, demonstrably, credibly, meaningfully. But only if the technology industry first confronts the environmental cost of technology itself with the same rigor and ambition it applies to the products it builds.

Cover image by Freepik [https://www.freepik.com]

References

Amazon (2023) Amazon sustainability report 2022. Seattle, WA: Amazon.com Inc. Available at: https://sustainability.aboutamazon.com (Accessed: 15 June 2026).

Apple Inc. (2023) Apple environmental progress report 2023. Cupertino, CA: Apple Inc. Available at: https://www.apple.com/environment/pdf/Apple_Environmental_Progress_Report_2023.pdf (Accessed: 15 June 2026).

Evans, R. and Gao, J. (2016) DeepMind AI reduces Google data center cooling bill by 40%. London: Google DeepMind. Available at: https://deepmind.com/blog/article/deepmind-ai-reduces-google-data-centre-cooling-bill (Accessed: 15 June 2026).

European Parliament and Council of the European Union (2024) Directive (EU) 2024/1799 of the European Parliament and of the Council of 13 June 2024 on common rules promoting the repair of goods (Right to Repair Directive). Official Journal of the European Union. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202401799 (Accessed: 15 June 2026).

Fairphone (2023) Fairphone impact report 2022. Amsterdam: Fairphone B.V. Available at: https://www.fairphone.com/en/impact/impact-report/ (Accessed: 16 June 2026).

Forti, V., Baldé, C.P., Kuehr, R. and Bel, G. (2023) The global e-waste monitor 2023. Bonn: United Nations University and United Nations Institute for Training and Research. Available at: https://www.itu.int/en/ITU-D/Environment/Pages/Spotlight/Global-Ewaste-Monitor-2020.aspx (Accessed: 15 June 2026).

Freitag, C., Berners-Lee, M., Widdicks, K., Knowles, B., Blair, G.S. and Friday, A. (2021) 'The real climate and transformative impact of ICT: a critique of estimates, trends, and regulations', Patterns, 2(9), article 100340. doi:10.1016/j.patter.2021.100340.

Google (2024) Google environmental report 2024. Mountain View, CA: Alphabet Inc. Available at: https://sustainability.google/reports/google-2024-environmental-report/ (Accessed: 15 June 2026).

IEA (International Energy Agency) (2024) Electricity 2024: analysis and forecast to 2026. Paris: IEA. Available at: https://www.iea.org/reports/electricity-2024 (Accessed: 15 June 2026).

IPCC (Intergovernmental Panel on Climate Change) (2022) Climate change 2022: mitigation of climate change. Contribution of working group III to the sixth assessment report of the Intergovernmental Panel on Climate Change. Cambridge: Cambridge University Press. doi:10.1017/9781009157926.

Lam, R., Sanchez-Gonzalez, A., Willson, M., Wirnsberger, P., Fortunato, M., Alet, F., Ravuri, S., Ewalds, T., Eaton-Rosen, Z., Hu, W., Merose, A., Hoyer, S., Holland, G., Vinyals, O., Stott, J., Pritzel, A., Mohamed, S. and Battaglia, P. (2023) 'Learning skillful medium-range global weather forecasting', Science, 382(6677), pp. 1416–1421. doi:10.1126/science.adi2336.

Merchant, A., Batzner, S., Schoenholz, S.S., Aykol, M., Cheon, G., Cubuk, E.D. and Cubuk, E.D. (2023) 'Scaling deep learning for materials discovery', Nature, 624(7990), pp. 80–85. doi:10.1038/s41586-023-06735-9.

Microsoft (2023) Microsoft environmental sustainability report 2023. Redmond, WA: Microsoft Corporation. Available at: https://query.prod.cms.rt.microsoft.com/cms/api/am/binary/RW15mgm (Accessed: 15 June 2026).

Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L.M., Rothchild, D., So, D., Texier, M. and Dean, J. (2021) 'Carbon and the machine learning cloud', Communications of the ACM, 65(6), pp. 52–58. doi:10.1145/3434204.

Pereira, R., Couto, M., Ribeiro, F., Rua, R., Cunha, J., Fernandes, J.P. and Saraiva, J. (2017) 'Energy efficiency across programming languages: how do energy, time, and memory relate?', in Proceedings of the 10th ACM SIGPLAN International Conference on Software Language Engineering, pp. 256–267. doi:10.1145/3136014.3136031.

Shepherd, M., Turner, J.A., Small, B. and Wheeler, D. (2020) 'Priorities for science to overcome hurdles thwarting the full promise of the "digital agriculture" revolution', Journal of the Science of Food and Agriculture, 100(14), pp. 5083–5092. doi:10.1002/jsfa.9346.

TSMC (Taiwan Semiconductor Manufacturing Company) (2023) TSMC sustainability report 2022. Hsinchu: Taiwan Semiconductor Manufacturing Company Limited. Available at: https://esg.tsmc.com/en/update/sustainabilityReport/reportAndDownload.html (Accessed: 15 June 2026).

Uptime Institute (2023) Global data centre survey results 2023. New York: Uptime Institute LLC. Available at: https://uptimeinstitute.com/2023-data-center-industry-survey-results (Accessed: 15 June 2026).

Wiesner, P., Behnke, I., Scheinert, D., Gontarska, K. and Thamsen, L. (2021) 'Let's wait awhile: how temporal workload shifting can reduce carbon emissions in the cloud', in Proceedings of the 22nd International Middleware Conference, pp. 260–272. doi:10.1145/3464298.3493399.