They are apparently trying to do AI as greenly as possible: https://blog.ecosia.org/ecosia-ai/

Reducing AI’s footprint isn’t enough — we’re here to make a positive impact. That’s why we generate more renewable energy than our AI features use, from 100% clean sources like solar and wind.

We’ve invested €18M in renewable energy projects — expanding solar parks and adding clean power to the grid. The energy we generate helps displace fossil fuels and accelerate the transition to renewable energy.

We use tools like the AI Energy Score and Ecologits to select efficient models and track their energy use — keeping our process transparent, and ourselves accountable.

https://support.ecosia.org/article/1006-ai-search#Model-training-and-adjustments-LOaKB

We use OpenAI’s GPT-4.1 mini because it strikes a strong balance between performance and efficiency. By choosing a smaller, mainstream model, we aim to deliver a high-quality experience while keeping our energy and water footprint low. We believe there is a need for greater transparency around AI’s environmental impact and continue to push for change across the tech industry.

Our AI Search is built using smaller, more efficient models less energy than larger ones, according to a recent study. We assessed the model’s impact using tools like EcoLogits, CodeCarbon or Hugging Face AI Energy Score. Additionally, we avoid energy-intensive functionalities like video generation or deep research.

We generate more clean energy from solar and wind than our AI features consume. The surplus goes into the electricity grid, helping to replace fossil fuels. While we can’t yet account for the emissions generated during the AI model’s training, using our AI Search features helps accelerate the energy transition towards renewables.