In short
Compare regions on grid carbon intensity and the share of carbon-free energy, using your provider's published region data or a source such as Electricity Maps, then weigh latency, data-residency rules and cost. Batch work such as training, fine-tuning and overnight processing is the easiest to move; user-facing inference usually needs to stay close to users.
Steps
- List which workloads can run anywhere, starting with training, fine-tuning and batch jobs.
- Compare candidate regions on grid carbon intensity and carbon-free energy share.
- Check data-residency, latency and cost constraints for each candidate.
- Allow for the energy and cost of moving data to the new region.
- Re-check the choice yearly, as grids and provider data change.
Region choice can change the carbon of the same job several times over, for no change in code. The electricity behind a kilowatt-hour differs widely between grids.
Where do I find region data?
- Google Cloud publishes, for each region, the hourly average share of carbon-free energy and the grid carbon intensity in gCO₂e/kWh. Its console marks regions "Low CO2" where carbon-free energy is at least 75% or, where that isn't available, grid intensity is at most 200 gCO₂e/kWh.
- Electricity Maps gives real-time, forecast and historical carbon intensity for more than 350 zones, useful for any provider.
- Other providers' sustainability pages and carbon dashboards; see your cloud provider's carbon data.
Which workloads should move?
The Green Software Foundation's pattern for AI workloads suggests running training, fine-tuning and batch work in low-carbon regions where latency is less critical. Microsoft's Azure Well-Architected guidance advises weighing carbon intensity, latency and resilience together when choosing regions.
Common mistakes
- Moving latency-sensitive inference far from users, which can hurt experience and add network traffic.
- Ignoring data movement. Copying large datasets between regions has its own cost.
- Relying on certificates. A provider buying renewable certificates doesn't make a high-carbon grid low-carbon; see checking whether your AI runs on renewable energy.
- Treating region choice as a substitute for efficiency. It cuts carbon, not energy.
More in hosting and data centres.
Sources
- Carbon free energy for Google Cloud regions, Google Cloud (2025)
- Electricity Maps data, Electricity Maps (2026)
- Use carbon-aware scheduling and region selection for AI workloads, Green Software Foundation, Green Software Patterns (2025)
- Reliability recommendations for sustainable workloads on Azure, Microsoft Azure Well-Architected Framework (2026)
Related
How do I schedule training when the grid is greener?
Run training and batch AI jobs when grid carbon intensity is low: use forecasts, make jobs pausable with checkpoints, and keep deadlines as hard limits.
How do I find the carbon intensity of my electricity grid?
Where to get grid carbon intensity data for AI footprints: regional, hourly sources such as the GB Carbon Intensity API and Electricity Maps.
How do I check whether my AI runs on renewable energy?
Tell hourly clean power from annual certificates: ask where your AI runs, the grid's carbon intensity and the hour-by-hour clean share for that region.
Green AI hosting and data centres
How to choose where AI runs: grid carbon intensity, PUE and WUE, renewable energy claims, water use and siting, with questions to ask hosts.