3 days ago
Europe's thirst for AI runs into local water worries
Europe's dream of expanding homegrown AI is hitting a resources problem – with push back against power hungry data-centres and their growing thirst for water necessary to keep racks of servers from over-heating.
That's a serious problem, given that expanding AI has become a top political priority across Europe. Those same water resources are needed by farms and power stations, intensifying the fight over water as droughts become more frequent.
For example, French President Emmanuel Macron, who hosted an AI Action Summit in Paris in February, wants the country to become a go-to destination for AI investment, which means luring more data centres onto French soil. France already hosts 322 data centres, and plans to ease permitting to host the more powerful data hubs underpinning the generative AI boom.
But the French proposal is facing pushback from opposition politicians and NGOs concerned about the environmental footprint of such projects, including their heavy use of electricity and water.
In May, French NGO la Quadrature du Net mobilised to call for a pause on new data centre permits, pending further public debate about resource usage – so far unsuccessfully. But anti-data centre movements are springing up across Europe.
Something similar happened in Germany when Elon Musk announced plans for a new Tesla electric vehicle plant in the lake-filled forests east of Berlin. In Dublin – a hub for overseas tech giants in Europe – NGOs have been calling for a moratorium on new data centres as their pull on the country's energy grid has increased.
Spain's drought-stricken Aragón region also made headlines this year after local activist group, Tu Nube Seca Mi Rio (your cloud dries up my river), fiercely challenged new data centre developments in a place where water is already so scarce.
A trickle of data
report in
Nor is th
Getting a comprehensive picture of the water use of data centres is challenging, as not all companiesthe same level of a single authority tracking this information.
To offer a rough idea of how thirsty data centres can be, in Google's latest Environmental Report (for 2024), the tech giant discloses that it consumed around 31 billion litres across its operations (data centres and offices) worldwide. That's equivalent to irrigating 54 golf courts for a year according to Google.
Google's figures for 2022 and 2023 also recorded a 17% increase in water consumption (reaching 23.1 billion litres two years ago) – at a time when the company's headcount shrank by several thousands, suggesting its data centres (operation and construction) were driving the increase.
Complicating the picture is Google's lack of any data on its water use at shared third-party data centres, which they do not operate themselves, a solution deployed by many Big Tech companies, but which makes accounting accurately for water use even harder.
North Holland provides another snapshot of water usage data after a Microsoft facility became operational in 2021. However the 75 million litres of drinking water reported for cooling the data centre – much more than the 12-20 million Microsoft had projected when applying for a permit – is disputed by the cloud giant which says it included water used for construction (another heavily water dependent industry).
Thirsty generative AIs
Dedicated information processing hubs designed to handle huge volumes of computation inevitably need water to cool the servers racked inside.
In colder countries, such facilities can pipe in outside air to reduce heat. But when temperatures exceed 25°C, as is often the case in Europe, air alone becomes insufficient. Making matters more complicated, AI is one of the most intensive processing activities going on in data centres today cranking up the need for coolant.
Quantifying the exact water usage associated with AI tools is also challenging. One 2023 study estimated that every 10 to 50 general queries fed to the GPT-3 version of OpenAI's chatbot require the equivalent of a 500ml water bottle – and that's in addition to all the cooling during the initial development of ChatGPT.
The EU's water industry association estimates that annual data centre water consumption will spike 52% by 2030 – rising from 62 million litres in 2024, to 94 million litres in 2030.
In areas where water is scarce, the biggest cloud providers, AWS, Google, and Microsoft, say they typically do not use water for cooling. But this comes with a cost trade-off; requiring more electricity to power air-based cooling.
Most cloud giants have also committed to reducing their overall impact – pledging to become "water-positive" by 2030, meaning they expect to re-inject more water into the environment than they use by then, such as by supporting wetland restoration projects.
Although, this extra water may not necessarily replenish supplies in the same areas where their operations are extracting water.
In parallel, data centre efficiency is increasing rapidly, and new technologies allow more recent AI models to be developed with much lower computing power requirements than their predecessors. Less power means less waste heat.
Ever-growing AI demand
Still, Europe's AI market is predicted to grow 26.3% yearly on average between 2025 and 2031, so as the systems get more efficient demand will boom. That's prompting climate activists to point out that gains will not make up for rising uptake of AI tools.
'Companies now offer more integrated high compute services [...] into everything", Matthias Spielkamp, founder of NGO AlgorithmWatch, told Euractiv, pointing as example to the "AI overviews" embedded in Google search results.
"You can't escape, you can't turn them off," he added, although Google denies its resource use has changed since it launched the AI overviews..
The European Commission also has big ambitions to ramp up AI development across the bloc.
According to its AI Continent action plan, the EU will fund five new AI "Gigafactories" to carry out the most resource-intensive AI training tasks on European soil. The initial call for interest in these massive data-processing facilities quickly received responses for more than 70 projects.
On top of that, the EU is exploring ways to simplify permitting for new data centres via the Cloud and AI Development Act, expected later this year. It has said this aims to ease access to local resources, as well as investing in innovations to make AI more resource-efficient.
The Commission also hopes to increase transparency into data centres' water usage to help local authorities anticipate shortages and improve their resource allocation.
As of May 2024, larger data centres must report water and energy use to the Commission under the EU's Energy Efficiency Directive (EED) – although no data has yet been made public. With its Water Resilience Strategy, the Commission also wants to improve data centres' sustainability ratings by including clearer water consumption criteria.
Critics, however, suggest the EED reports won't be detailed enough to properly support local decision-making around a finite – and increasingly precious – resource.
(nl, jp, cp)
