Source: Getty Images montage

.

For at least a decade, universities around the world have been declaring their determination to get to net-zero emissions. Then, in 2019, a slew of universities around the world went a stage further, declaring climate “emergencies” that often came with accelerated targets to eliminate net institutional carbon emissions.

Announcing Cardiff University’s climate emergency in 2019, for instance, its then vice-chancellor Colin Riordan said the university had already fully divested from fossil fuels and must “lead by example and accelerate our plans to reduce our carbon emissions, energy and water use and overhaul our operational activities”.

And in 2021, Universities UK’s Confronting the climate emergency report committed all UK universities to setting targets for reducing the carbon footprint of sources they control directly (known as source 1 emissions) and of the energy they use (known as source 2). It also promised that institutions would “set a target” for reducing scope 3 emissions, caused by the production of products and services they use – or, failing that, “commit to a programme of work to set targets as soon as possible”. They would publish these targets on their websites and “set out how progress against these targets will be reported in a transparent, consistent, and understandable way”.

But then, late the following year, ChatGPT was unleashed, and institutional attention turned towards the adoption of AI. There were concerns about it, of course, but they were centred around its corrosive effect on academic integrity. And notwithstanding those concerns, many universities have committed to making AI tools available to all staff and students and embedding technology into the curriculum – on the grounds that the technology is here to stay and students need to be fluent in its use.

In 2025, for instance, California State University rolled out ChatGPT Edu – OpenAI’s customised education version of its large language model tailored to higher education institutions – to more than 460,000 students and more than 63,000 staff, making it the biggest educational roll-out of ChatGPT in the world. And later that year, the University of Oxford became the first UK institution to roll out ChatGPT Edu to students, while earlier this year the University of Manchester announced a “world-first” partnership with Microsoft to provide access to its AI tool, 365 Copilot, for all staff and students.

Yet amid all the debates about employability, pedagogy and the integrity of student assessment, the potential conflict of mass AI roll-outs with net-zero targets has been rather overlooked.

A June report by UK IT body Jisc and the Environmental Association for Universities and Colleges (EAUC) warned that pressure for universities to adopt AI has “outpaced clarity about what responsible action looks like in practice”.

The report cites statistics from the International Energy Agency, which show that global data centres’ energy consumption could more than double by 2030 – or even, according to Greenpeace Germany, increase elevenfold. And data centres already account for 6 per cent of all electricity consumption in the US and the UK, according to an industry body – and a much higher proportion in some other countries.

“As institutions with public commitments to net zero, tackling climate change, and broader environmental sustainability goals…it is imperative that we recognise how the adoption of these technologies is contributing to both our individual and collective environmental footprint across multiple dimensions,” the Jisc/EAUC report read.

Water consumption is also a major concern, as large amounts of water are needed to cool data centres. Research published in the journal NPG Clean Water in 2021 found that even a relatively small 1 MW data centre can consume about 25 million litres of water per year, while the UK water company Affinity Water told UK MPs in May that one recent proposal for a data centre had estimated that its daily water need would be equivalent to that of 147,000 people.

And Jonatan Pinkse, research director at the Centre for Sustainable Business at King’s College London, noted that Google’s recently released sustainability report revealed an 18 per cent year-on-year increase in carbon emissions as it expands its AI operations, and an 81 per cent increase in emissions from 2019, despite having a 2030 net-zero target.

“The negative [of AI use] has become so clear so fast that there’s no going around it any more,” he said.

Michael Draper, professor in legal education at the University of Swansea and director of the university’s academic regulations and student cases board, said: “If you actually design your assessment so that AI [must be] used by the student, and some students are saying, ‘Well, we shouldn’t be doing this because of the impact on the environment, and some of the ethical concerns’, there are real concerns around that, and that conversation isn’t really had…It’s all around academic integrity. The point I raise…is that institutions often have a commitment to UN sustainability goals. Well, where does that commitment fit with the widescale adoption of artificial intelligence when it’s using all this energy and water?”

One reason that such conversations rarely occur is that the environmental impact of AI is not universally known. Cal Innes, a digital sustainability specialist at Jisc and co-author of the Jisc/EAUC report, told Times Higher Education that “a lot of people are going into [AI usage] as they do with a lot of aspects of digital behaviours: not realising that there’s an environmental footprint associated with it”.

But even university managers who are aware of the problem and determined to address it face an uphill struggle since AI companies rarely disclose information regarding individual institutions’ AI use, the Jisc/EAUC report says, making it almost impossible for universities to produce reliable estimates of the environmental cost of their AI use.

THE asked several universities that have announced major AI roll-outs about how they are tracking their emissions data in relation to AI use. However, out of California State, Oxford, Manchester, Arizona State, the Massachusetts Institute of Technology and the universities of Cambridge and Surrey, only Surrey, Manchester and MIT provided responses.

A spokesperson for Surrey, which recently announced that AI is being incorporated into curricula for all subjects, explained that it had “robust monitoring processes” in place as part of its sustainability agenda, and added it is “applying those same standards to how we procure AI tools for our framework”. It estimates its indirect emissions and reports them in both its annual sustainability report and its institutional annual report.

“Scope 3 emissions are harder to pin down than direct emissions, as they depend on suppliers reporting their own carbon data back to us,” the spokesperson conceded. However, “we’ve recently updated our procurement policy to require this, and we’re now starting to collect that data, including from external AI vendors”.

A spokesperson for Manchester, meanwhile, argued that it is important that staff and students have “equitable” access to AI tools, and for them to be equipped with “necessary skills for the workplace”, including learning to use AI “responsibly”. The university is working “closely” with Microsoft to ensure transparency around AI’s environmental impacts, the spokesperson said, adding it has initiated “groundbreaking research in partnership with Microsoft to model the impacts of our Copilot usage and to inform the actions the university will take to manage and minimise these impacts. We are exploring how best to understand and monitor these impacts as the university-wide roll-out progresses.”

Alex de Vries-Gao, founder of Digiconomist, which examines the impact of technology trends on the environment, is concerned by AI’s typical absence from universities’ sustainability statements: he would expect organisations to reference it even if merely to acknowledge “how difficult it may be to obtain the right information”.

“If you’re not capable of getting the numbers, at least talk about it and show that you’re thinking about this because if you’re not discussing it, you’re probably not thinking about it,” he said.

A spokesperson from OpenAI said the company gives considerable thought to the best use of its computing power and said it supports its partners to meet their sustainability and water-consumption goals. And they pointed out that the Jisc/EAUC report did not take into account modern closed-loop water cooling systems, which are more efficient than traditional cooling towers that allow water to evaporate away, adding that it is currently developing data centres in Norway that run entirely on renewable energy.

OpenAI believes AI will be instrumental in tackling climate change by optimising energy systems and accelerating research, the spokesperson added, noting that the company is partnering with leading universities to accelerate these efforts, such as Oxford and the US National Laboratories.

Microsoft declined to comment but its 2026 Environmental Sustainability Report said that in the 2025 financial year it had “replenished more water than we withdrew​” and “achieved our milestone to match 100% of our annual electricity consumption with renewable energy”.​ It added that the company is “scaling clean, reliable energy to meet growing demand from cloud and AI, exploring sources from nuclear to fusion, and investing in the grid infrastructure and technologies needed to get there​”.

.

Universities are also exploring how AI can help them reduce their own scope 1 emissions. While MIT was unable to provide details on its generative AI (GenAI) use and how it is tracking such data, it outlined that it is currently expanding a pilot launched in 2023 that uses machine learning to optimise its energy systems and increase efficiency in heating and cooling its buildings. It said the programme had reduced the pilot building’s energy use by up to 40 per cent annually and was now being expanded to additional buildings and updated so MIT systems can automatically respond to live data.

But Charlotte Bonner, chief executive of EAUC and co-author of the report with Jisc, noted that traditional machine learning is distinct from GenAI tools. And De Vries-Gao, who is also completing a PhD at the Vrije Universiteit Amsterdam Institute for Environmental Studies, said machine learning tools do not have as significant an impact as GenAI and are more likely to have been developed in-house, which provides institutions with greater transparency over their energy use.

That lack of transparency about AI’s environmental cost is the “ultimate problem”, he said, adding that institutions are also “not capable of really dissecting what is the benefit of each individual application [of AI] either. So you’re missing both sides of the equation.”

Examples of such applications are also few and far between. Bonner and Innes had wanted to include more positive examples of universities using AI to address the environmental impact, but “once we really started to look into the research available, we found that where claims were being made about how AI use was helping drive positive environmental change, it was nearly always as a result of traditional machine learning rather than GenAI,” Bonner said. “We can’t really find at this moment in time any substantiated evidence that generative AI is helping drive positive environmental change.”

And even though universities’ AI use may lead to research breakthroughs that ameliorate climate change, King’s’ Pinkse said this is far from guaranteed.

“We hope that it will of course lead to fantastic breakthroughs,” he said. “But you can never really promise it because…the whole point of research and development [is] that we don’t always know what the outcome is going to be.”

.

So what are universities to do? None of the experts that THE spoke to said that abandoning AI was the way forward. Pinkse’s view is that universities are in an “almost impossible” dilemma given the ubiquity of AI.

“If a university were to say, ‘We’re becoming a non-AI university’, students would start saying, ‘We’re not too sure about this one’,” Pinkse said. “That puts them in a very vulnerable position because they are competing with other universities and other organisations. There are [also] demands that we deliver students who are AI literate and so forth. So I don’t think it’s that easy to say, ‘We should not be doing this’…It’s unfair as a demand on universities.”

Meelis Kitsing, rector of the Estonia Business School, thinks he has found the right balance by seeking to produce graduates who are not only AI-literate but also able to engage with wider ethical debates regarding technology. To that end, the school has embedded questions on AI throughout its curriculum and strategy, embracing the tech while also remaining critical of it.

The institution is also hosting its seventh “digitalisation and sustainability” summer course bringing together students, academics and policymakers to examine how AI is reshaping business models while simultaneously intensifying energy demands.

“We need to encourage people to be critical thinkers, rather than ideologues who believe that AI will solve all problems – or the opposite: that AI will only create problems for sustainability,” Kitsing said. “I think that truth is more likely to be characterised by different shades of grey, rather than be black and white.”

Ultimately, the “guilt” for AI use’s environmental impact should not fall on academics and students, Innes believes. He referred to a concept known as “greenshifting”, whereby the responsibility for environmental harm is projected by companies on to consumers, which he said “diverts away from institutional responsibility, where we have the greatest leverage”.

But do universities really have any meaningful leverage over BigTech firms?

Bonner suspects that they have more power than they think to demand greater transparency on AI’s water consumption and energy use, provided they act collectively – and, ideally, with other sectors, such as health systems. And she can “foresee a future” in which universities develop shared “sector-specific large language models” and data centres – in a similar way that numerous institutions have access to Isambard-AI, the University of Bristol-based supercomputer purpose-built for AI research.

That, as she and others pointed out, would allow them to better track the technology’s usage and energy consumption. “But I don’t think that we’ll ever get to a situation where we don’t use some of [the AI developed by] the household names,” she conceded.

Universities could also restrict academics and students to “acceptable” uses of AI, Innes suggested, noting that creating just eight seconds of video with AI can produce up to 2,000 times more carbon than a single text prompt. But he was wary of singling out AI firms when other forms of digital technology also have significant environmental impacts. While it is important to hold them to account, “we’re not routinely questioning the footprint of watching YouTube videos, doomscrolling social media, or leaving webcams on in online conferences”, he noted.

It is abundantly clear that AI is only going to become ever more closely integrated into university teaching, research and administration. Jisc itself, for instance, is currently running a trial of AI marking tools. But in the midst of another unprecedently hot summer across Europe and North America, the phrase “climate emergency” has never seemed more apt – and it would seem odd for institutions pledged to combating it to simply dismiss AI’s huge carbon footprint as a problem too difficult and poorly documented for them to engage with.

For her part, Bonner conceded that institutions can often feel “overwhelmed” when it comes to tackling the sustainability consequences of AI. But she urged them not to wait until they have “perfect data” on it before acting.

“We don’t want there to be a paralysis of action because the data quality isn’t there,” she said. Because if universities take that approach, “we’re never going to do anything”.