Carbon-aware computing
Edge and cloud computing appear on 151 recent papers, with 43 in the last quarter against about 34 per quarter before, and 46 of them carry an energy, carbon or green keyword, mostly about device energy. Carbon-aware scheduling, which moves computing jobs to times and places where the grid is clean, appears on one paper. How much carbon could it save, and who would pay for the delay?
What prompted this
Since October 2025, 151 papers carry a cloud-computing, edge-computing, data-centre, job-scheduling, task-offloading or serverless keyword (111 in AI journals, 14 in economics, 11 in finance, 9 in operations research): "edge computing" on 63, "cloud computing" on 22, "task offloading" on 14, "green data centers" on 4, "energy-efficient computing" on 11. The quarter to September 2026 had 43 such papers against about 34 per quarter earlier. Forty-six of the 151 carry an energy, carbon, renewable or green keyword, but "carbon-aware scheduling" appears on a single paper.
Questions
- For training and batch inference jobs that can wait hours or move between regions, how much carbon can be cut at current grid carbon intensities, and at what cost in delay and transfer?
- Do cloud providers' carbon accounting methods (hourly matching versus annual) change which schedules look clean?
- Would a carbon price on computing move jobs enough to matter for the grid, and where?
Data that could help
Hourly grid carbon intensity by region (ENTSO-E, electricityMap), cloud job traces released by providers, and electricity prices.
Next step
Simulate shifting a public cloud job trace across regions and hours with 2025 grid carbon data and compute the carbon saved per hour of added delay.