AI isn’t free: Google and Mistral finally put numbers on chatbot costs 💧

Cliquez ici pour lire en français

Who knew a single AI query could spark such an energy debate? This week, Google and French AI darling Mistral lifted the veil on just how much juice their chatbots actually burn. The numbers are in—and they’re either reassuring or alarming, depending on how you look at them.

The numbers are finally out 📊

Google now claims that a “typical” prompt to Gemini, its flagship AI model, consumes around 0.24 watt-hours—roughly the power needed to keep a small LED bulb lit for a few seconds. Water usage? About 0.26 milliliters—just five drops. The company even equates the electricity to about nine seconds of TV time (depending on the set, of course).

Mistral, on the other hand, went in a completely different direction. The French startup published a broader lifecycle analysis that includes training, usage, and chip production. For Mistral Large 2, the company estimates 1.14 grams of CO₂ and nearly 50 milliliters of water per query. That’s not just transparency—it’s a statement.

Dueling methodologies, dueling narratives ⚖️

The real fight isn’t just about numbers, but about how you count them. Google’s figures exclude user devices, model training, and even server recycling. The company’s approach focuses only on the live service, not the bigger picture—and its methodology hasn’t been vetted by ISO standards.

Mistral, meanwhile, leaned on independent experts to validate its model. Unsurprisingly, the two approaches produce vastly different estimates, highlighting just how murky AI’s environmental reporting still is.

Responsibility or reassurance? 🌍

Google insists it’s making progress: in the past year, it says Gemini’s energy usage dropped 33x and its carbon footprint 44x, thanks to datacenter efficiency and model optimization.

Mistral, while smaller, is pushing for something bigger: international standards. Its pitch is simple—if every AI provider measured the same way, users could finally compare apples to apples and hold companies accountable.

For now, what we’re seeing is the start of a conversation, not a finished framework.

A more transparent AI future ?

These first disclosures are less about hard answers and more about setting the stage. Should we trust the reassuring efficiency claims of a global giant, or the tough-love transparency of a challenger? Either way, as AI becomes more embedded in our daily lives, real accountability can’t stay optional.

👀 What do you think—are these numbers surprising? Can AI really go green? Share your thoughts in the comments and let’s keep the debate going.