by Madhuri Karak & Cathy Richards

In a recent event hosted by MERLTech’s Climate + AI Working Group, Anastasia Tsougka from Environmental Defense Fund Europe and Hannah Lipstein from Data and Society discussed the regulatory landscape attempting to tackle the rising climate and socio-environmental impact of data centers in the US and the EU. 

1. Europe’s energy math is alarming 

In 2022, data centers accounted for nearly 3% of the EU’s total electricity demand. By 2024, they were responsible for roughly 60% of the growth in Europe’s electricity demand. Projections from the International Energy Agency suggest consumption could hit 150 terawatt-hours by 2030 — about 5% of total EU electricity demand.

The EU wants to triple its data center and AI capacity. If current numbers are already straining the grid, the implications of that ambition are hard to overstate.

2. Efficiency alone is a trap

Both speakers flagged the “overshoot effect.” When you make a system more efficient, you often create stronger incentives to expand it. More efficiency doesn’t mean less consumption — it can mean a rise in demand. 

Drawing on a report she wrote last year for the Environmental Coalition on Standards, Tsougka argued for a sufficiency approach: rather than making data centers run cleaner, Europe should ask whether it needs to build new ones at all. The existing infrastructure, she argued, is significantly underutilized.

3. EU regulation is fragmented in key areas

Reporting obligations for large data centers were instated in late 2024, but compliance was poor; this data isn’t truly public, and individual operators can’t be identified in aggregated results. A rating scheme is in development, and minimum performance standards are expected around 2026 — but both are efficiency-focused, not consumption-focused.

Meanwhile, simplification agendas are quietly stripping environmental requirements from AI and digital legislation. The 2025 EU AI Act, Tsougka noted, was “very poor in environmental content.”

4. Where data centers get built matters

Locating a data center in Greece or Spain — countries already dealing with 40°C summers — massively increases cooling demands compared to cooler northern climates. Yet EU-backed gigafactory plans include exactly these countries. 

Tsougka’s recommendation: factor in local climate, water stress, and community input before approvals, not after.

5. Fighting data centers one at a time is not enough 

Lipstein described the risk of “whack-a-mole” activism — defeat one data center, and it pops up somewhere else, often in a more vulnerable community with fewer resources to push back. Her answer: address demand upstream. Smaller, purpose-built AI models use dramatically less energy to train and run. If we need fewer computing cycles, we need fewer data centers.

She’s also working on a public procurement toolkit to help government agencies ask vendors the right environmental questions.

6. The question we need to ask more often

Perhaps the most important thread through the conversation was this: do the models powering AI now need to be as big as they are now? 

Lipstein was direct. The current wave of large language models operates on a “bigger is better, scale at all costs” logic that isn’t the only way to build AI. The assumed productivity gains are shakier than advertised — research suggests AI often redistributes labor rather than reducing it.

Resources you will want to explore:

• ECOS, Environmental Coalition on Standards: This is an international NGO with a network of members and experts advocating for environmentally friendly technical standards, policies, and laws. You can find their reports on how data centre expansion risks derailing climate goals here.
• EcoLogits Calculator: A web app to raise awareness on the environmental impacts and consequences of using GenAI for different use cases.
• The Green Software Foundation has compiled a comprehensive slate of technical tools to measure and reduce the environmental impact of software.
• The Green Web Foundation has developed its own open-source technical tools to manage the environmental impact of digital services. 
• The Institute of Electrical and Electronics Engineers (IEEE) has a resource guide including its own list of Green Software Measurement Tools.
• France’s AFNOR Spec General Framework for Frugal AI document offers developer-facing comprehensive guidance for the measurement and reduction of an AI product’s environmental impact. It also offers some evaluation metrics for potential procurers (see Table 1). Annex 1 of the document contains an inventory of tools and databases available to developers.
• Multiple projects compiling technical and non-technical resources on Frugal AI. See: Cambridge University’s Judge Business School’s Frugal AI Resource Hub or Rémy Marrone’s Frugal AI project. 

This post is based on a CoP event on April 14 2026, discussing data center evaluations in the United States and the European Union. The event speakers were Anastasia Tsougka (Environmental Defense Fund Europe) and Hannah Lipstein (Data and Society).