Editor’s note
Welcome to the 20th edition of the Digital Sustainability Newsletter.
If time is short, I’d encourage you to go straight to the link in the top news story and read the UN University report on AI’s energy, water and land footprint. It’s one of the most comprehensive assessments of its kind to date, and it makes a compelling case that we’ve been measuring AI’s environmental impact through far too narrow a lens. Carbon matters, but the water and land costs of the infrastructure powering AI are substantial, unevenly distributed, and almost entirely absent from current policy and governance frameworks. It’s essential reading.
That report, like several of the other news stories this month, underlines that we find ourselves a genuinely difficult moment in time. The tension between technological acceleration and sustainability commitments is real and growing, and the rollbacks are concrete: Microsoft quietly retreating from its hourly clean energy target, Scotland’s planning framework approving “green” data centres against criteria written before ChatGPT existed, over a hundred UK data centre projects seeking permanent gas connections. These aren’t hypothetical risks, they’re very real, as is the potential for the situation to get much, much worse if left unchecked.
And yet. The academic research featured this month, alongside the tools and resources we highlight each edition, still gives me genuine cause for optimism. The quality and breadth of work being done on the environmental footprint of the tech sector is improving rapidly, and the people doing it are asking the right questions.
I hope you find something useful in here. As ever, get in touch with stories, papers, or resources worth sharing, and subscribe for future editions of the Digital Sustainability Newsletter via the DIGITAL-SUSTAINABILITY JiscMail list if you haven’t already.
– Cal Innes, Digital Sustainability Specialist, Jisc (cal.innes@jisc.ac.uk)
Digital sustainability news
Here’s a quick roundup of this month’s biggest digital sustainability news headlines:
A new UN University report finds that by 2030, global data centres are projected to consume 945 terawatt-hours of electricity, with an associated water footprint equivalent to the annual domestic water needs of 1.3 billion people in Sub-Saharan Africa, and a land footprint exceeding 14,500 square kilometres. The report concludes that “low-carbon” is not automatically “low-water” or “low-land,” warning that evaluating AI sustainability through a single metric can hide trade-offs and shift environmental burdens onto regions already facing water or land stress.
A petition circulating across Europe is calling on governments to ban bitcoin on environmental grounds, arguing its annual electricity consumption of 138 to 172 terawatt-hours is unsustainable, though the proposal has prompted debate about whether cryptocurrency is the right target compared to other energy-intensive industries.
Microsoft is considering delaying or scrapping its 100/100/0 target, which committed to matching 100% of its electricity consumption with zero-carbon energy on an hour-by-hour basis by 2030, according to people with knowledge of the matter cited by Bloomberg. The company has been adding around a gigawatt of data centre capacity every three months and expects to spend $190 billion on infrastructure through the end of December, with mounting costs squeezing budgets in divisions responsible for reducing its carbon footprint.
More than 100 data centre projects in the UK are seeking permanent gas connections to generate their own electricity, after years-long waits to connect to the National Grid, with regulators warning the trend raises serious questions about the UK’s Clean Power 2030 targets.
The first studies to quantify the resource demands of large-scale quantum computing find that while electricity needs may be manageable, supplies of helium-3, a rare isotope needed to cool quantum systems, and water could become the binding constraints on growth. Efficiency gains alone will not reduce overall demand, and leading quantum nations have yet to address any of this in their national roadmaps.
Four of the world’s largest tech companies have backed a new membership-based programme that will fund and pilot sustainability technologies inside data centres, investing up to $5 million per project across areas including energy, water, materials, and circularity, with the aim of scaling solutions that can eventually be deployed across other sectors including hospitals and schools.
Brazil’s lower house has passed a bill that would prohibit environmental agencies from using satellite images alone to restrict commercial activity on illegally deforested land, requiring ground-level confirmation instead. IBAMA, Brazil’s federal environmental police, warns the measure could jeopardise around 70% of its enforcement actions in the Brazilian Amazon, where it has roughly 1,250 agents to patrol a forest roughly the size of Western Europe.
Scotland’s National Planning Framework 4 states that “green data centres” will have “overall negligible impact” on Scotland’s emissions reduction goals, but according to Action to Protect Rural Scotland, the greenhouse gas assessment underpinning this conclusion was published in October 2022, predates ChatGPT, and explicitly excluded hyperscale AI facilities, which APRS says were simply not considered when the framework was written. With no government definition of what a “green data centre” actually is, APRS is calling for an immediate moratorium on hyperscale data centre decisions until policy catches up.
The European Commission is introducing a mandatory labelling system for larger data centres as part of a revision to the Energy Efficiency Directive, requiring operators to display ratings covering power usage, water consumption, renewable energy share, and waste heat reuse on an A-to-G scale, with the first labels due to be generated by the European database in August 2027.
Academic papers
Now for a summary of some of the most important new digital sustainability research:
Streaming the future of sustainability
Authors: Ryan Hardesty Lewis & Alex Conway
A position paper that makes a counterintuitive but carefully argued case that the shift to cloud-based streaming, far from being an environmental problem, could actually be one of the most effective levers available for reducing the embodied carbon and e-waste associated with personal computing. The core argument is that by offloading computation to renewably-powered data centres and redesigning edge devices as lightweight streaming receivers rather than general-purpose computers, device lifespans could be dramatically extended.
From cradle to cloud: a life cycle review of AI’s environmental footprint
Authors: Katherine Lambert & Sasha Luccioni
A review of 61 papers on AI’s environmental impacts, organised through an eight-stage life cycle framework from hardware manufacturing to end-of-life disposal. The headline finding is that research clusters heavily around training and inference while most other stages remain largely unstudied. One of the most striking findings from this study is that none of the analysed papers explicitly modelled post-training adaptation as a distinct life cycle stage, despite it now being central to how AI models are actually developed and deployed. Reporting is also narrowly focused on energy and carbon, with water use, material extraction, and toxicity rarely featuring. The authors call for standardised, life cycle-aware reporting as a basic prerequisite for meaningful environmental governance of AI.
Authors: Yuhan Sun, Jiacong Sun, Xiaoling Yi & Marian Verhelst
Chiplet architectures, which split AI processors into smaller modular units rather than integrating everything on one die, have been proposed as a lower-carbon alternative to monolithic chip designs. Previous modelling suggested chiplets could cut embodied carbon by up to 30%, but those estimates ignored the physical overheads of connecting chiplets together. Testing against two real processors, this paper finds that packaging and interconnect overhead can account for up to 75% of total carbon footprint, and that chiplets only come out ahead when chip area is relatively large and chiplet count stays moderate. According to the findings from this research, for smaller designs a conventional single-chip approach is greener.
Resource corner
Each month, we share a digital sustainability report, tool, or resource that we think is worth your time.
Wholegrain Digital’s digital declutter checklist
This month’s digital sustainability resource is Wholegrain Digital’s free checklist, which covers the practical end of individual and organisational digital sustainability, with sections on hardware, email, cloud storage, video calls, streaming, and AI. The tips are specific enough to be useful: turn off camera during webinars, download on WiFi rather than streaming, link to files rather than attaching copies, close AI chat threads before starting new ones. Nothing here will surprise anyone already working in this space, but it’s a well-organised resource to share with colleagues who are newer to it.
Digital sustainability articles
Here is a selection of our favourite articles on digital sustainability from the last month. Click on the title link to be redirected to the full article:
Panthalassa’s floating, wave-powered data centre technology
Charlie King, writing in Sustainability Magazine, profiles Oregon-based start-up Panthalassa, which has raised $140m at a valuation of nearly $1bn for its wave-powered floating data centre technology, designed to generate electricity from open-ocean waves and deliver compute capacity via Starlink satellite with no land connection required. The concept sidesteps two of the biggest constraints on data centre expansion, grid connection and energy supply, by generating and consuming power in the same place.
AI and Sustainability Communication: Avoiding the Most Common Pitfalls
Fiona Leibundgut’s blog post argues that vague language around AI and sustainability is enabling greenwashing and obscuring real harm, and sets out four common pitfalls in public discourse that she says must be addressed before meaningful accountability is possible. Pitfalls include framing AI as a straightforward climate solution, treating AI as a monolith and ignoring the human cost of AI.
What it will take to make AI sustainable
WIRED’s interview with AI sustainability researcher Sasha Luccioni, who is leaving Hugging Face to co-found Sustainable AI Group with former Salesforce sustainability chief Boris Gamazaychikov, covers the growing pressure on companies to quantify their AI emissions, the structural incentives keeping big models dominant, and what meaningful transparency from the major AI providers would actually look like.
We optimised the hardware. We forgot the software. It’s time to build better by design
Luc Burnip, Architecture Lead at DXC, in a guest blog for techUK, argues that while hardware efficiency and renewable energy have dominated the conversation around data centre sustainability, software design remains a largely unregulated and underexamined source of energy waste, and that embedding carbon awareness into software development culture is one of the most significant untapped opportunities for meaningful carbon reduction in the AI era.
Tokens and greens: measuring the impacts of agentic AI
Navveen Balani sets out in a post on the Green Software Foundation website why agentic AI represents the next major and largely unacknowledged efficiency problem in the sector, arguing that the combination of planning loops, tool calls, reflection, and multi-agent debate can multiply token consumption by up to 1,000 times compared to equivalent single-call interactions, with direct consequences for energy, carbon, and water that most organisations deploying agents are not yet measuring
Are tokens the only data centre metric that matter in the age of AI?
Dan Swinhoe, in Data Centre Dynamics, examines whether token efficiency is displacing PUE and WUE as the primary metric by which data centre operators measure and optimise performance, and whether that shift risks pushing sustainability considerations further to the margins at exactly the moment they matter most. The piece draws on a recent Schneider Electric presentation and industry sources to make the case that while token efficiency is economically logical, it is environmentally incomplete.
Podcast pick
Green IO – AI & sustainability: follow the money!
Another fantastic episode of Green IO, hosted by Gaël Duez, featuring two guests who also appear elsewhere in this newsletter. Boris Gamazaychikov and Sasha Luccioni make the case that economics, not carbon metrics, is the key lens for understanding AI’s environmental trajectory, and that organisations are now gaining the bargaining power to demand accountability from hyperscalers.
Get involved
We want to hear from you! Share your comments, suggestions, and digital sustainability highlights. Contact our Subject Specialist for Digital Sustainability, Cal Innes, at cal.innes@jisc.ac.uk
And don’t forget to subscribe to our DIGITAL-SUSTAINABILITY JiscMail mailing list for future editions of the Digital Sustainability Newsletter.
