What is the real environmental cost of AI?

As generative and agentic AI become part of everyday life, explore their growing material impact on the planet – from energy and water consumption to data centres, critical minerals and global supply chains.

---

 

In October 2026, Calvium is exhibiting at the Blue Earth Summit, where people and organisations will be exploring impact-driven solutions for the global environment, societies and future economies.

One topic we are particularly interested in is the growing material impact of generative AI and agentic AI on the planet. As their use becomes more widespread, there is huge potential for AI to contribute to positive change, but the technology also comes with sizeable environmental and social costs. From the energy and water needed to power data centres to the materials and infrastructure behind them, these impacts raise important questions about how we develop and use AI responsibly.

There is still a lot we need to understand, which is why we are looking forward to learning from the many experts, organisations and different perspectives at the conference. How can the impacts of AI be better understood and mitigated? How do we make sure digital innovation genuinely benefits people, place and planet?

One useful way to consider those questions is through the lens of the Sustainable Development Goals (SDGs), and the progress the world has made towards them, which is what this article seeks to do.

UN Sustainable Development Goals

Just over 10 years ago, United Nation member states agreed to adopt a set of 17 interconnected global social, economic and environmental goals united by the aim to “leave no one behind”. The Sustainable Development Goals (SDGs) officially came into force on 1 January 2016, with a target to achieve these by 2030. 169 targets and 230 indicators were implemented to measure progress, which felt potentially achievable at the time. As we approach that target, it is feeling more and more unlikely.

While ‘net zero’ is not a specific SDG, it is inextricably linked to the environmental goals and a central concept of the 2015 Paris Agreement, which is also aiming for a 2030 target. There has been mixed and insufficient progress to-date, with just 35% of SDG targets on track or showing moderate progress. 18% of targets, meanwhile, have regressed from the 2015 baseline.

Given this, it is clear that new mindsets, models and ways of working are needed if we are to reach those goals, for which digital technologies have been positioned as part of the solution. Artificial Intelligence (as an umbrella term for many and diverse technologies) has a role to play in accelerating progress, but it also brings challenges that risk undermining the SDGs – environmentally and socially. It begs the question: if digital innovation is seen as a solution to many global challenges, but also emerging as a new source of environmental strain, how can we innovate for the genuine good of people, place and planet?

AI through the lens of the SDGs

We have previously written about how digital tech can support the SDGs, and while those points still stand, a lot has changed over the past four years. It feels more urgent today that we work together to stem the rapid trajectory towards irreversible climate catastrophe. The responsible use of AI’s data capabilities is already being used to make a positive contribution. Environmentally, it is enhancing climate modelling, agricultural optimisation and land use analysis. There are also growing examples of its use in healthcare diagnostics, revolutionising the way we treat and manage some of the world’s most prevalent illnesses.

The trouble is, to create the technologies and infrastructure that enables this is coming at a cost. While it remains difficult to measure, it is widely acknowledged that many examples of AI and their associated supply chains are causing problems in conflict with the SDGs – from consuming vast amounts of energy and water, to mining minerals, producing toxic waste and causing habitat destruction as a result. 

According to the UN’s chief digital officer, Golestan Sally Radwan: “You need to break it down along many dimensions across the entire environmental lifecycle, which starts from the raw materials that go into manufacturing all these graphics processing units (GPUs), the construction of data centres, energy consumption, emissions, water use, e-waste generation, land degradation, and the rise in unsustainable consumption and production owing to the granular marketing that AI helps us do.” 

The globalisation of AI means these problems are only set to get worse. We need to understand the bigger picture – the end-to-end material impact of AI.

The energy cost of AI

The rapid adoption of the technology, compounded by the mainstream use of GenAI, requires enormous computing demand, thus increasing reliance on energy-intensive data centres. Training large language models (LLMs) is particularly energy intensive as they rely on specialised hardware processors (GPUs and TPUs), which work around the clock and consume energy that may not come from renewable sources.

AI needs power every time it generates a response, and the figures surrounding this are quite shocking. According to the IEA, data centres consumed around 485 TWh of electricity in 2025, which is 1.5% of the total global consumption and set to double by 2030.

On a macro scale, an AI data centre consumes as much electricity as 100,000 households. At a household level, some experts estimate one 10-second AI video uses the same power as watching 5.5 hours of Netflix. Around 2.5 billion prompts are sent every day on ChatGPT alone, which is thought to use enough energy to charge eight million phones. 

AI’s carbon footprint is growing as a result, with researchers estimating training GPT-3 emitted roughly 500 metric tonnes of CO2. Some may argue that AI’s carbon footprint is relatively small compared to aviation or cement production; the problem is the rate at which it is accelerating.

Carbon market specialists, Climate Impact Partners, highlight why it is so important that we strike the right balance: “If we pull back on computing resources to save energy, we risk slowing innovation. But if we keep scaling AI without regard for its carbon footprint, we risk undermining the very future those innovations are supposed to protect.”

Materials and supply chains

Another major concern is the damage being caused by supply chains and mining raw minerals to build the physical infrastructure needed to power AI. There is particular demand for critical minerals such as lithium, cobalt and copper; many of these play an essential role in the transition to renewable energy due to their use in electric vehicle batteries and wind turbines, for example, but AI is adding new pressures. 

Giant quantities of copper are needed for chip-level wiring, server hardware and power distribution, for example. So much so that AI-related copper demand could rise to by 50% by 2040, creating global shortages. Others are even used to polish semi-conductor chips.

Meanwhile, extraction in biodiverse and protected areas is causing long-term habitat damage – soil degradation, water contamination, toxic waste – which is explicitly at odds with SDG 15 (Life on Land). High-performance AI data centres also need to be cooled, which creates a huge water footprint.

According to one piece of research, generating 10 to 50 AI prompts consumes about 500ml of water, which amounts to one standard bottle. Another report by the UK Government predicts AI will increase global water usage from 1.1bn to 6.6bn cubic meters by 2027 – equivalent to more than half of the UK’s total water usage. Problematically, many of the world’s data centres are in river basins with high risk of water pollution, which means much of the local water supply may be unsafe for use. This is compounded by the fact that just 0.5% of the Earth’s water is available freshwater, so AI is increasingly causing water stress and scarcity. 

Mitigating the impact

The reality is becoming increasingly concerning and the chance of meeting those 2030 targets without urgent intervention is feeling less and less likely. 

Fundamentally, the infrastructure must be rethought and rebuilt. Green data centres that are renewable energy-powered and more efficient should be seen as a priority. There is movement happening here, with the Greening AI Data Centres Coalition (GADCC) launched in April 2026. The global initiative has pledged to set credible standards for sustainable data centre development and define what “green” genuinely means.

“Greening AI data centres is no longer a choice; it is an imperative for a sustainable digital future,” said Mr K S Venkatagiri of the Indian Green Building Council, one of the founding members. “As AI accelerates demand, sustainability must accelerate innovation.”

Greener software practices are needed too. For example, green coding principles, efficient model design, optimising code to use less CPU power. But it is also about thoughtful automation and design; integrating legacy systems and existing hardware, and ensuring infrastructure is adaptable to reduce the need for constant updates. These are things individual businesses can look to implement internally – as we have written about in detail here. The Circular Electronics Partnership also highlights how circular strategies can help businesses to reduce their carbon footprint. Small changes combine to have a big impact.

Globally, there needs to be more transparency along the supply chain. If AI providers share data on the impact of the technology, we can start to create reporting standards and frameworks so that impact and progress can be measured, like the GADCC is striving to do. There is something to be said about localisation too; hosting AI tools locally would help to reduce data transfer and give greater control over environmental footprint.

That said, location itself needs careful consideration. Concentrating too many data centres close to large populations can create competition for local water supplies, particularly because of the significant amounts of water needed for cooling. In areas where water is already under pressure, such as the South East of England, growing demand from data centres could add to water stress and put further strain on supplies needed by communities and other local users. This means decisions about where AI infrastructure is built need to consider not just efficiency and connectivity, but the availability of local resources and the wider impact on the places and people around them.

Clearly, whilst just skirting the surface, we can see that collaboration across sectors and countries is essential. Global governments, tech companies, designers and engineers must look for solutions that align AI development with the SDGs. With mindful design, transparency and cooperation, AI can be more of a part of the solution, and less a part of the problem.

Continuing the conversation at Blue Earth Summit

There are no simple answers to the challenges outlined here. The impacts of AI are complex, interconnected and, in many cases and for many reasons, very difficult to measure. What is clear is that addressing them is critical and urgently needed – non-negotiable.

That is one of the reasons why Team Calvium is looking forward to being part of the Blue Earth Summit. As well as sharing our own expertise of designing and developing digital systems, we are particularly interested in learning from those around us. A starter for 10: How can AI’s environmental and social impacts be mitigated? What work is already happening, and what can we learn from it? Where does responsibility lie? And what do these lessons mean for the way digital products and services are designed, developed and deployed?

If generative and agentic AI are going to be part of the solution to global challenges, rather than adding massively to them, we need to think about their impact now. That means asking difficult questions about when and where AI is appropriate, understanding the infrastructure and resources sitting behind them, and designing technology with people, place and planet in mind.

We hope that Blue Earth Summit will provide an opportunity to learn, challenge our thinking and continue that conversation with others working towards the same goal. If you are attending, come and find Calvium at our exhibition stand, we would love to meet you!

 

Calvium is a digital consultancy and product partner for purpose-led organisations. We help clients navigate complexity, create trusted digital services, and accelerate change through responsible digital innovation.

We collaborate with our clients to understand their challenges, bringing together strategic thinking, inclusive design and technical expertise to create solutions that are built around the needs of the people who use them. Every project is designed to integrate with existing systems and processes, helping organisations improve services and adapt as their needs evolve.

Calvium is a certified B Corp and winner of the National Technology Awards 2026 Responsible and Ethical AI Award.