TL;DR
A June report from Jisc and the sustainability body EAUC finds that the pressure on the sector to take up AI has run ahead of any clear account of what responsible adoption means. The binding problem is measurement: AI vendors seldom release institution-level usage figures, so universities that have publicly committed to net zero cannot calculate what their rollouts cost in carbon or water.
The numbers behind the concern
The report draws on International Energy Agency projections that global data centre electricity demand could more than double by 2030 — Greenpeace Germany puts the possible increase at elevenfold. Industry figures already place data centres at around 6% of electricity consumption in both the UK and the US.
Water is the less-discussed half. Research in NPG Clean Water found a modest 1MW facility can get through roughly 25 million litres annually. Affinity Water’s evidence to MPs back in May described one proposal needing the daily volume used by 147,000 people.
Jonatan Pinkse of King’s College London points to Google’s latest sustainability disclosure: emissions up 18% year on year and 81% above 2019, against a 2030 net-zero commitment.
Nobody can show their working
Times Higher Education approached seven institutions with major AI deployments. Four — including Oxford and Cambridge — did not respond. Surrey said its procurement policy has been rewritten to require suppliers to report carbon data, and that it is now beginning to collect it from AI vendors. Manchester cited a Microsoft research partnership set up to model what its Copilot usage costs.
EAUC chief executive Charlotte Bonner notes that the positive examples her team went looking for mostly turned out to be conventional machine learning rather than generative tools, and that no substantiated evidence exists of generative AI driving environmental improvement. Alex de Vries-Gao of Digiconomist calls the transparency gap the fundamental problem — institutions can assess neither the cost nor the benefit of individual applications.
Why this reaches beyond campuses
The same disclosure gap applies to any UK organisation with a reported carbon target and an enterprise AI licence. Scope 3 emissions depend on suppliers reporting their own figures, and the major AI vendors currently do not break them down per customer. Surrey’s response — making it a procurement requirement — is the only mechanism available, and it only works if enough buyers insist.
Looking forward
This is the reporting-side twin of the planning fights now surrounding UK data centres. Communities are contesting the visible infrastructure while the institutions buying its output cannot say how much of it they use.