survey of carbon aware scheduling for ai workloads

Carbon-aware computing research indicates that flexible batch workloads can reduce emissions intensity by aligning compute windows with cleaner grid intervals (Google Carbon Intelligent Computing). AI training and large inference batches are prime candidates.

see also: power purchase agreements enter software roadmaps · europe power pricing reshapes data center siting

evidence stack

• Time-shifting non-urgent jobs lowers carbon intensity without hardware changes.
• Regional workload routing amplifies benefits when network constraints permit.
• Savings shrink when strict latency SLAs eliminate scheduling flexibility.

practical boundary

Carbon-aware routing succeeds only when teams expose workload classes by urgency and maintain visibility into energy mix signals.

my take

The main barrier is organizational, not technical. Most stacks can schedule cleaner compute, but few product teams classify workload urgency rigorously enough.

linkage

• [[power purchase agreements enter software roadmaps]]
• [[europe power pricing reshapes data center siting]]
• [[data center power futures rise in constrained regions]]

ending questions

which workload class should be moved first to maximize emissions reduction with minimal user impact?