Problem
AI workloads can push a data center’s power demand up by tens of megawatts in minutes, often faster than its grid connection can supply. On-site gas turbines can cover the gap, but they respond over minutes while computing load changes in seconds. Keeping the two in balance without tripping any plant limit is a hard control problem.
Approach
Each side of the facility gets its own learned world model: one for the turbine fleet and one for the AI data hall. A single planner looks ahead across both at once and decides how far the turbines should ramp and how much work the data hall can safely defer. Every command is checked against the turbines’ physical limits before it reaches the plant, so the learned components never act unchecked.
Outcome
- Real-time decisions on ordinary hardware. The planner decides in 10–50 milliseconds on a CPU, fast enough for live plant control with no GPU.
- Accurate forecasts of both sides. Both world models met their accuracy targets, trained on nearly 650,000 examples that include real job traces from NREL’s Kestrel supercomputer augmented with RAPS simulator.
- Better than conventional control. The learned control policy outperforms a standard fixed-setpoint controller.
- A working end-to-end demo. The full loop runs interactively, from prediction to actuation, including an islanding mode for grid disconnects.
- Patent filed on the interacting world models approach.