Ecosia’s Bet on Leaner, European AI Challenges Big Tech’s Power-Hungry Models

Berlin-based Ecosia has spent years turning web searches into forests. Now the not-for-profit search engine wants its artificial intelligence to pull its weight for the climate too. On May 22, 2026, the company announced it had shifted its AI almost entirely to a European model provider. The move promises lower energy use, stricter privacy rules and a path toward self-hosting. But not everyone buys the green narrative.

The latest blog post struck a hopeful tone. “What if AI could be greener? What if it could be better for the planet?” Ecosia asked. (Ecosia Blog). The answer, according to the company, lies in smaller models, renewable energy surplus and EU oversight. Queries don’t linger with third parties. The system runs on power from solar and wind that exceeds its own consumption. Surplus electricity flows back into the grid, the firm says, displacing fossil fuels.

Smaller Models, Measured Impact

Ecosia relies on Mistral Small 4 for its AI Chat and Mistral Small 3.2 for AI Overviews. These choices reflect a deliberate trade-off. Larger models demand more electricity and water. Smaller ones strike a balance. The company evaluates them with tools such as EcoLogits, CodeCarbon and the Hugging Face AI Energy Score. It skips video generation and other intensive features entirely. (Ecosia Support).

Christian Kroll, Ecosia’s founder and CEO, spelled out the logic in a 2024 interview. “We’re not training our own models from scratch,” he said. Training would consume far more energy than deploying an existing model. And Ecosia’s entire surplus goes to tree planting and climate projects. Diverting funds to massive compute clusters would contradict the mission. (Forbes).

Yet the numbers stay vague. Ecosia claims its AI uses “far fewer resources than mainstream models.” It generates more clean power than its features consume. Kroll once told Forbes the company aims to be carbon-negative, returning three times the clean energy it uses. Exact figures for current AI operations remain unpublished. Critics want proof.

Ketan Joshi, an energy and climate analyst, called the effort “false hope for corrosive tech” in December 2025. He argued that even efficient generative models still encourage plagiarism of human writing and distract from genuine search. Automatic summaries, he wrote, block users from original sources. (Ketan Joshi). Recent X discussions echo the divide. Some users praise the optional AI toggle. Others see any integration as a step backward for a tree-planting service.

And Ecosia does keep the feature optional. Users can disable AI Overviews in settings. AI Chat lives in its own tab. The company insists it only activates when a query clearly benefits from conversation. Auto-routing can send complex questions there, but users may turn that off after the first trigger. File uploads arrived recently. Users can feed PDFs, spreadsheets or images up to 5 MB each, limited to five per chat. Files vanish from servers within seven days.

Upcoming additions include memory, web-sourced answers and a guided learning mode. The latter breaks topics into steps instead of handing over final answers. A “Climate Intelligence” feature, sketched in earlier posts, would offer eco tips drawn from current science. The goal is a personal climate coach. But only if the user opts in. (Ecosia Blog, Aug 2025).

Privacy forms another pillar. As a European company, Ecosia falls under GDPR and the EU AI Act. Conversations stay in local browser storage unless users enable encrypted history. The firm collects minimal data. It runs no email, maps or payment services that could create user profiles. When queries reach Mistral, Ecosia forwards them without building long-term dossiers. Sensitive topics receive warnings. The system can still hallucinate facts or miss nuance. Feedback buttons let users flag problems.

Independence matters too. Ecosia once leaned on Bing and Google. In 2024 it formed a joint venture with French search engine Qwant to build a European index. That index, still developing, will let the company combine fresh web data with language models. Kroll explained the gap. “An LLM is generated maybe a few months ago or sometimes even a few years ago, and they don’t have the recent information. You need to combine an LLM with an index.” (Forbes).

The European provider shift announced in May 2026 advances that vision. It reduces exposure to U.S. political pressures and opens a route to self-hosting inside Europe. Data centers would sit closer to Ecosia’s values. Yet the company still depends on Mistral, a French outfit. Full independence lies ahead.

Supporters see a rare alternative. Most search giants push AI summaries by default. Ecosia lets users ignore them. Its not-for-profit status funnels revenue to reforestation. Over 250 million trees planted, the firm claims in recent updates. Every search theoretically contributes. But critics counter that AI still burns resources better spent on direct conservation or simpler tools.

Recent web chatter shows the tension. On X, users recommend Ecosia precisely because AI can stay off. Others dismiss the green claims as marketing. One developer thread suggested pairing Ecosia with plugins that avoid AI entirely. Adoption grows among those fleeing Google’s expanding AI Overviews, set for further rollout this week.

Ecosia measures its AI impact with third-party calculators. It admits it cannot yet tally emissions from initial model training. The focus stays on inference: the electricity and water used when users actually query the system. By choosing compact models and avoiding flashy extras, the company reports a lighter load. Surplus renewables, it says, accelerate the grid’s shift away from coal and gas.

So the experiment continues. Ecosia refines its models, listens to feedback at AI.feedback@ecosia.org and expands features only after testing. It publishes licensing deals with publishers to promote transparency. It experiments with citizen-led urban forestry in Berlin. The AI push fits a broader pattern: technology in service of the planet, not the other way around.

Whether that formula scales remains open. Mainstream models grow more capable and more expensive to run. Water shortages near data centers make headlines. Regulators tighten rules. In that context Ecosia’s restraint looks strategic. Smaller. Optional. Tied to measurable climate returns. But the proof sits in the data it has yet to publish in full.

Users face a choice. They can click the star icon for conversation or stick to plain results. They can upload a report for summary or read the original. They can trust the system’s sources or verify elsewhere. Ecosia bets enough people want an option that plants trees while answering questions. Recent updates suggest that bet is still live.