Every click in the browser, every photo sent in a family chat, every email consumes resources — not only on the device itself, but also in data centers that manage digital traffic. The EU has ruled that data centers must operate efficiently and make their consumption transparent. However, this reporting framework can be gamed to create a false impression of ecological progress, as a paper by researchers at HPI now shows.
Data centers are physical facilities powering almost all digital services, from social media posts to complex AI models. Many data centers for cloud-based and Large Generative Artificial Intelligence (AI) applications, such as Chat GPT, are increasingly owned and scaled across Europe by major Big Tech companies.
Between growth and sustainability
Under the recast Energy Efficiency Directive, operators are required to quantify and report the energy and water impact of their facilities. At the same time, the European Commission considers the expansion of data center capacity essential to making Europe more sovereign and competitive. This creates a conflict: wherever ever-larger data centers are built, entire communities may face water scarcity and pressure on electricity grids.
Against this backdrop, HPI postdoctoral researcher Daria Onitiu and HPI Professor Sandra Wachter from the Technology & Regulation group, together with Brent Mittelstadt from the University of Oxford, took a closer look at the criteria used to define efficient data centers.
The scaling paradox
They have now published their findings in a pre-print showing that operators can use the benchmarks endorsed by the recast EED — Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE) — to present “efficient” low scores. “At the heart of the problem is an efficiency paradox,“ says Daria Onitiu. “Data center operators can expand their facilities indefinitely while maintaining, or even reducing, their average scores across those facilities. This reporting framework risks obscuring or ignoring the resulting strain that larger data centers place on local and global electricity grids and water resources.“
This paradox operates on two levels:
- Operators can make their data centres more energy efficient by employing a range of “technical optimisation measures”, such as more efficiently utilising existing servers. However, second-order effects can quickly eliminate these gains, for example via rebounds where rising demand for AI services mitigates the benefits of more efficient servers.
- Data centre operators must scale aggressively to maintain their lower PUE and WUE scores, for example by increasing the number of servers or retrofitting cooling infrastructure, both of which pose additional environmental costs.
Ways out of the contradiction
The authors make three suggestions to make the recast EED scheme fit-for-purpose to increase sustainability:
- A new Delegated Regulation to reveal and certify efficiency improvements
- Reporting requirements for providers to reveal PUE and WUE trade-offs
- A bespoke mechanism for the EU Commission to track efficiency improvements, their diminishing returns and countereffects over time.
The full preprint of the paper is available here.
More news
Contact persons
