Agentic AI Data Center Power & Cooling Planner

The saved model multiplies active agents by steps per task and a calibrated coefficient. At one hundred fifty thousand agents and twelve steps, IT load is approximately seventy-three point four seven eight megawatts. Multiplying by the power usage effectiveness factor one point one five gives exactly eighty-four point five megawatts before display rounding. Doubling agents to three hundred thousand doubles power to one hundred sixty-nine megawatts. Both bars share three pixels per megawatt. Doubling steps has the same linear effect; this formula has no arrival rate, task duration, GPU utilization or hardware efficiency model. The coefficient was selected to match a locked baseline, so these results are scenario assumptions rather than measurements of real agent workers or infrastructure demand. Daily modeled water multiplies total power by approximately sixteen thousand eight hundred four point seven three gallons per megawatt-day. At the baseline this gives about one million four hundred twenty thousand gallons per day. The hybrid factor is one, the air-cooling factor one point four five, and immersion point six five. Immersion therefore gives approximately nine hundred twenty-three thousand gallons, while air cooling gives two million fifty-nine thousand. Both displayed bars share three pixels per ten thousand gallons per day. The cooling selector changes only this multiplier, not calculated power. These assigned factors do not verify actual water use for any cooling installation. Substation headroom compares baseline total power with capacity; at ninety-six megawatts it is approximately twelve percent. Negative headroom is clipped to zero, hiding the magnitude of any modeled shortfall. The hourly curve is a fixed twenty-four-element profile, not observed telemetry. Its maximum multiplier is one point one nine at index thirteen, or thirteen hundred hours, despite a comment saying hour fourteen. Applying that multiplier to the rounded eighty-four point five baseline yields a displayed peak of one hundred point six megawatts. Both bars share five pixels per megawatt. That exceeds the default ninety-six megawatt capacity even though the main headroom metric uses only baseline power. Grid stress is assigned high when headroom is below fifteen percent or baseline power exceeds seventy-five megawatts, with a lower medium threshold; it is not a measured grid condition. Hourly water applies an additional one point zero five multiplier, and the profile sums to twenty-three point seven one, so the twenty-four hourly water values do not sum to the stated daily baseline. Exported projections are assumptions to inspect, not forecasts verified against a real facility. No external model or backend is called.