Move the work. Watch the costs move.
One fictional computing center has 100 kWh of work. Forty kWh can wait for daylight. Move that work and see which costs change—and which do not.
38.0 kgIllustrative electricity emissions
0.0 kg vs original schedule
38.0 LIllustrative cooling water
0.0 L vs original schedule
100 kWhSame work, same energy
Scheduling alone saves no energy here
Try moving half of the flexible work.
Change the assumptions
Two equal-duration periods. Assumed electricity carbon factors: daylight 0.1 kg/kWh; evening 0.5 kg/kWh. Evening cooling water: 0.2 L/kWh. No measured site, price, grid congestion, cooling electricity, embodied emissions or AI optimization overhead is modeled.
Emissions = daylight kWh × 0.1 + evening kWh × 0.5. Water = daylight kWh × your water factor + evening kWh × 0.2. Each full bar is 10 kWh.
Where the question comes from
The experience illustrates a question raised by these sources. It does not reproduce their model, data, or measured outcomes. Open the connected stories to inspect the evidence and its limits.
Making room in a data center ↗
Google says the production heuristic continuously recovers, on average, 0.7% of its worldwide compute resources.
This is a company-reported, Google-specific operational result.
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms · 2025-05-14 ↗Finding more workable grid plans ↗
The announcement reports feasible-solution rates rising from 14% to over 88%, reducing the need for costly post-processing.
The source reports a computational result, not live-utility deployment.
AlphaEvolve: How our Gemini-powered coding agent is scaling impact across fields · 2026-05-07 ↗