Can AI frameworks be green? Here’s what we found

AI is powerful, but it’s also power-hungry

Every headline about LLMs seems to highlight the same problem: soaring compute costs and rising carbon footprints. For enterprises in Europe especially, this is becoming a boardroom and regulatory concern.

Here’s what we’ve learned while benchmarking frameworks:

Many orchestration frameworks add 5–7× CPU and memory overhead compared to what’s really needed.

That inefficiency isn’t just slower and more expensive, it’s wasteful.

We’ve been working on a framework that flips that equation: lighter on your servers, lighter on the planet.

To put it in perspective:

• If a typical enterprise AI workload consumes 100 kWh, efficient execution can bring that down to ~15–20 kWh.

• That’s a CO₂ reduction of ~19 kg per workload run.

• At just 1,000 runs per year, that’s ~19,000 kg CO₂ saved the equivalent of planting 850 trees or avoiding 80,000 km of car travel.

Sometimes the biggest impact isn’t about what AI can do, it’s about how sustainably we can run it.

Curious to hear from this community:

Do you see carbon footprint and compute efficiency becoming a deciding factor in AI adoption where you work?