What is energy-aware compute?
Energy-aware compute treats electricity availability, price and carbon intensity as first-class scheduling inputs alongside deadlines and priority.
Most compute is scheduled as if electricity were uniform: always available, always the same price, always the same carbon intensity. On a site with solar generation and battery storage, none of those things are true across a day, let alone a year.
Energy-aware compute takes a different starting point. Before a workload runs, the scheduler asks what energy is available now, what is forecast for the next few hours, what the battery reserve is being held for, and how urgent the work really is.
In practice this splits workloads into three groups. Interactive work needs to run immediately and must have a reliable supply path, including grid and cloud fallback. Batch work — embeddings, fine-tuning runs, video renders, overnight index builds — can be deferred into periods of strong local generation. Background work can be paused entirely without anyone noticing.
The measurable outcome is a higher share of work completed on locally generated energy, and better visibility of the energy cost of each unit of compute. It is not a claim that solar alone can power high-demand AI continuously, and it does not remove the need for grid supply.
Start by measuring. A node that reports its own power draw alongside site generation data gives you the baseline you need before changing any policy.
- scheduling
- fundamentals
- telemetry