EN
Back to the archive

The encyclopedia · Software & IT · Operational decision · 2020–2022

Google shifted non-urgent compute to sunny, windy hours to cut grid carbon.

Google reschedules flexible jobs around forecast hourly grid carbon, in time then location, cutting carbon without new hardware.

Google

the move

Google matched all its energy with renewable purchases by 2019, but wanted cleaning 24x7 carbon-free energy, which requires using solar and wind when they are available, not merely buying credits. A data center that runs flat all day is still drawing from the grid in the evening when the sun has set.

A small team built a carbon-intelligent platform. Every day it pulls two forecasts: the projected hourly carbon intensity of the local grid, and the power the data center will need for its compute tasks. It then optimises hour-by-hour guidelines to align compute with low-carbon supply.

The trick is that a lot of Google's work is flexible. Creating filter features for Google Photos, transcoding YouTube video and adding words to Google Translate can happen hours later without a user noticing. Shifting those tasks into sunny and windy periods needs no extra hardware and no slowdown for Search, Maps or YouTube.

Early pilots showed it worked, and in 2021 Google extended the platform to shift flexible compute between data centers, favouring regions whose grids are running on carbon-free power that hour. The whole system was designed to get every job done, just at a cleaner time and place.

why it works

  • Grid carbon intensity varies by hour, so timing is a real lever on emissions.
  • Non-urgent compute is flexible, so it can wait for clean supply.
  • No new hardware is needed, only a scheduler that respects the grid signal.
  • Location shifting adds a second dimension the moment grids differ by region.
the payoffShift work to where clean power is, not stand up more.inspired

what transfers

When a cost is set by a varying external signal, optimise timing against that signal rather than buying your way past it.

what came after

Google's carbon-aware load shifting became a template for data-center sustainability and influenced how hyperscalers talk about matching compute to clean energy. It was a 2026 Franz Edelman Award finalist (Google, 'Carbon and Power Aware Computing at Google'), and the method spread as operators realised that when and where compute runs matters as much as how much.

references

spotted an error? The archive wants to know.

same kind of clever