Google shifted flexible compute across time and place to chase clean energy

The carbon emitted per kilowatt-hour varies widely by time of day and region, so a fleet of data centers running the same workload at a fixed time produces an avoidable carbon footprint.

Google's carbon-intelligent computing treats that as an optimization problem: it forecasts hourly carbon intensity, predicts demand, and uses risk-aware optimization to produce capacity curves that limit flexible workloads at high-carbon times and relocate them to cleaner hours and cleaner regions.

The approach lets Google run more computing where and when carbon-free energy is abundant without sacrificing the jobs that must run now. It was recognized as a finalist project for the INFORMS Franz Edelman Award, and Google has published the underlying method and its measured behavior.

Move the workload, not the grid, to greener power

Carbon intensity varies by hour and location, so when you compute matters as much as what you compute.

Some workloads are flexible, so they can wait or move without hurting users or service-level targets.

Risk-aware optimization keeps the work within a day, so delaying it does not turn into missing it.

The same system saves money by using less expensive, cleaner power at off-peak and sunnier times.

When energy is greener elsewhere or later, the cheaper move is often to relocate the work rather than buy cleaner power for a fixed schedule.

Google extended shifting across time to shifting across location, and made per-region hourly carbon-free-energy data public so customers could pick greener regions. The work is part of Google's goal to run on round-the-clock carbon-free energy everywhere by 2030, and it was named a Franz Edelman Award finalist.