Green AI
Green AI means designing and running AI to use less energy for each request, and measuring that energy with a method others can check. Our work on it builds on Prime Edge AI, which decides where each AI request is answered.
Last updated
Four ways to use less energy
A large share of the energy AI uses goes into answering requests, and it grows with every new user. So we start with the request.
-
Answer on the device where it can
Prime Edge AI answers routine requests on the phone or computer they come from, so they never travel to a data centre at all.
-
The smallest model that does the job
Each request goes to the smallest model that can answer it well. Small models, and larger ones made smaller by distillation and quantisation, use less computing than the largest models.
-
Less work per request
Shorter answers where they serve people better, and answers reused instead of worked out again.
-
Heavy jobs when the grid is cleaner
Work that can wait, such as training and large batch jobs, is scheduled for hours when the electricity grid runs on more renewable power.
How is the energy of an AI request measured?
A number is only worth publishing if you can check it. This is how we will measure ours.
- A stated baseline. We compare each request answered our way with the same request answered in the cloud alone.
- The whole journey. We count the energy used on the device and the network as well as in the data centre.
- A published method. We will measure with the Green Software Foundation's Software Carbon Intensity for AI (SCI for AI) specification, ratified in December 2025, and publish the method alongside the results.
- Results as ranges. We will publish what we find, on named workloads, before any figure appears in our marketing.
What we do not claim
We make no claim that our AI is carbon neutral or has no environmental impact. Answering on a device still uses energy, and we count it.
Why it matters now
- AI is driving demand for electricity. The International Energy Agency projects that data centres' electricity use will roughly double, from about 485 TWh in 2025 to about 950 TWh in 2030 (Key Questions on Energy and AI, April 2026).
- Buyers are starting to ask. Large organisations reporting their emissions want figures from their suppliers, down to the workload.
- It can cost less too. The same choices that save energy, smaller models and fewer trips to the cloud, can also cut cloud bills and make answers faster.
- A right-sizing review. Which of your AI requests could use a smaller model, or run on the device, and what that would save in cost and energy.
- Energy per request, reported. Measurement built into the AI we build and run for you, for your own reporting.
- Our patent applications. Prime Edge AI draws on our work on controlling where AI processing requests can go. See our patent portfolio