You’ve probably heard by now that AI has huge environmental costs. When potential benefits of AI come up, rarely do we hear them weighed up against the known environmental impacts. The story is often about more compute, more data centers, more (eventual) benefit.
To better ground our understanding of what value AI can add to the world, we hosted a Learning Lab on the implications of AI on the environment.
This session focused on the environmental costs of AI infrastructure and the global resistance to data centers. We covered the ongoing legal battles where local communities are working to prevent the development of data centers — which require large amounts of water and electricity — in areas that already struggle with drought and other forms of environmental degradation.
We also explored how AI infrastructure is politicised, and how AI hype is used as a springboard for colonial powers to exploit areas of the world that contain the precious minerals needed to build hardware.
Speakers:
Boxi Wu (The Oxford Internet Institute) — Currently completing their MsC in Social Science at Oxford, with a specific interest in the social and political impacts of AI infrastructure. They also organize for ESEA Green Lions in London.
Jenna Ruddock (Policy Counsel at Free Press) — Jenna’s policy work focuses on tech, privacy, digital civil rights, and competition, with an interest in how these policy areas intersect with environmental justice.
The intersections between the environment and AI are numerous. This session did not cover potential applications of AI to combat the climate crisis, like tracking environmental harms, or the use of AI to power disinformation about the climate crisis. If these topics are of interest, or you want help navigating any others, we’re open to covering them in further sessions.
Watch the recording
Password: AI
Data centres are energy vampires
This Learning Lab had two threads running through it that are key when discussing AI & climate:
‘The cloud’ is physical infrastructure, and is destructive: the narrative pushed by industry players is that the infrastructure necessary to run AI systems is lightweight and infinite — whereas ‘the cloud’ in fact exists as large, resource-hungry data centres.
The existence of data centres is invisible to most: the scale and pollutant power of these data centres is most obvious to effected communities, and barely visible to everyone else. Even climate activists are unaware of the environmental implications of AI.
With these threads in mind, both Boxi and Jenna made it clear that the infrastructure supporting AI systems is physically large, destructive to local communities and ecosystems, and heavily politicised.
AI systems need a large amount of compute power for model training, maintenance, and continual API access which supports ‘always-on’ consumer products (such as ChatGPT). This compute is provided by GPUs (or ‘AI chips’), the majority of which are manufactured by Nvidia — Boxi mentioned that Nvidia’s competitors can be counted on one hand — and housed in data centres. These centres resemble industrial farms and have measurable impacts on surrounding communities and ecosystems by taking up land, consuming water, and creating significant noise pollution.
The above impacts are central to community concerns: there are prime spots for data centres in places like Chile, where communities already struggle with drought and therefore cannot afford the additional strain on their water supply. Areas with data centre development also have severely increased energy demands, which benefits existing fossil fuel operations: Dominion Energy in Virginia have delayed the retirement of two coal plants, and are even building new gas plants, to accommodate the energy needs of data centres in the area.
It’s important to note that community push-back comes in different shapes and sizes, as does corporate and governmental resistance to this push-back. Jenna shared that, in 2021, The Oregonian, a local newspaper, was sued by The Dalles on behalf of Google for requesting water consumption data on their proposed data centre. Google sued on the grounds that water consumption data was a trade secret — demonstrating both outsized corporate power and a clear strategic need for these data centres to exist. The Dalles has since settled the lawsuit and will provide water consumption data to the public.
In England, the push-back is more about the protection of green spaces and local ecologies. Those protesting data centre development on the Green Belt — a green area surrounding London, protected to prevent urbanisation — are often mischaracterised as individualistic reactionaries who are mainly concerned with aesthetics, rather than environmental degradation. The development of data centres is also framed by corporate entities as a positive move, creating more jobs and painting local areas as new thriving ‘hubs’ for technological progress. This framing manifests a false trade-off between growth and sustainability, leaving communities feeling marginalised by the perceived benefits of AI.
Investing in infrastructure means investing in a very specific vision for the future. The continued development of AI infrastructure is part of long-term, resource-intensive plans, from which tech companies expect major returns. Equally, governments are now fighting to keep this infrastructure on-shore, as they are a conduit for political power and a means of escaping austerity.
There are coordinated efforts around the world to slow or stop extractive production of AI infrastructure, such as:
New zoning laws and ordinances: Chandler, Arizona have experienced an accumulation of data centres over the last two decades, and the energy and water consumption — as well as noise pollution — have become unmanageable. They passed two ‘first of their kind’ local laws: one for noise pollution, and another more stringent zoning law for new data centre development.
Strategic litigation: In Chile, which is in the middle of a 15-year drought, an environmental court partially reversed a permit for Google to build a new data centre, requesting that its cooling system be revised so that it wouldn’t take up so much water.
Better reporting on these issues: Industry partners are aligned on lobbying strategies, pushing narratives that cloud infrastructure is cost-effective and weightless. Providing counter-narratives is vital, and Karen Hao is currently running an initiative to train 1000 journalists to report more rigorously on AI (funded by The Ford Foundation).
Protests & the documentation of protests: Also funded by Ford is the documentation of protests, such as those organised in Northern Virginia by The Piedmont Environmental Council. Powerful images and videos shared online show the scale of data centre projects and allow protesting communities to tell their stories.
Plans for AI systems that are actually sustainable: Smaller models, open-source compute, and effective international collaboration ensure that communities outside of the Global North are not beholden to the strategic whims of just a handful of countries.
In this Learning Lab, we also heard briefly from Erin Simpson and Michael Brennan from The Ford Foundation. Both touched on their current grant-making efforts, which include policy development and orchestrating the narrative shifts necessary to facilitate it. Also mentioned was the lack of connective tissue between those working in tech and climate activists. Michael noted a conversation he had with an activist who wanted to use AI to help with their work yet had no idea about the environmental implications, highlighting the need for these groups to convene more regularly.