We're building AI systems at breakneck speed, but we have no real idea what they're doing to the planet. That's the core argument from researcher Sasha Luccioni, who's been tracking AI's environmental footprint.
The problem isn't just that AI uses a lot of energy. It's that we don't have good data on how much energy different AI tasks actually consume. Training a large language model is obviously expensive, but what about all those API calls you're making every day? Nobody's really measuring that at scale.
Luccioni argues we need two things before we can fix this. First, better emissions tracking across the entire AI pipeline, from training to inference. Second, we need to understand how people are actually using these tools in practice.
That second point matters more than you might think. If most AI usage is people generating mediocre marketing copy or asking ChatGPT to write their emails, the environmental cost starts looking pretty hard to justify. But if AI is accelerating scientific research or making critical systems more efficient, the calculus changes.
For anyone building with AI tools, this is worth paying attention to. Sustainability concerns could shape which models are available, how much they cost, and whether regulators start putting guardrails around AI usage. The free-for-all phase won't last forever.
