Carbon-Aware Computing — Vocabulary and Language

Carbon intensity (gCO2eq/kWh): the greenhouse gas emissions per kilowatt-hour of electricity. Low-carbon periods: when more renewables (solar, wind) are generating. High-carbon periods: when fossil fuel peaker plants are running. Carbon-aware software moves workloads to low-intensity times or regions. APIs like Electricity Maps provide real-time carbon intensity data.

Demand shifting (temporal): delay a batch training job until 3am when wind power is high and carbon intensity is low. The work still happens — just at a lower-carbon time. Examples: CI/CD builds, ML training, backup jobs, report generation. The Green Software Foundation calls this 'temporal demand shifting.'

Spatial shifting: if your workload can run in multiple cloud regions, choose the region with the lowest carbon intensity at the time of execution. Example: run a batch job in a region where it is currently windy (low carbon) rather than a region running on coal. AWS, Azure, and GCP now provide per-region carbon intensity data to enable this.

Energy proportionality (Barroso & Holzle, Google): a system with 10% utilization should use approximately 10% of its peak energy. In practice, servers at 10% utilization may consume 50%+ of their peak energy due to idle overhead. Better utilization (consolidation, workload scheduling) is often more carbon-efficient than buying more efficient hardware.

GHG Protocol scopes: Scope 1 (direct — burning fuel on-site), Scope 2 (indirect — purchased electricity), Scope 3 (all other value-chain emissions — hardware manufacturing accounts for a large portion of tech industry Scope 3). For software: the embodied carbon of manufacturing hardware is often larger than operational energy over the hardware's lifetime.