My friend Allen Lorch built a PUE/cost calculator for a giant AI data-center build and asked a good question: where does your own plant actually sit on the gap between 1.2 and 1.6? His numbers were sharp — at PUE 1.5, a 50 MW facility burns a third of its grid draw on cooling rather than compute.
We got to talking about payback — the number that actually gets capex signed off by people who hold the checkbook. That's a conversation I've had a hundred times across thirty years of docks. A Lean engineer calls idle, wasted effort muda. A planner calls the same thing the cost nobody billed. Both are right, and both miss the real point until you put a dollar figure on the fix.
The worked model
Allen's rough math on the Ohio build, which he's putting into the calculator, goes like this. Close the cooling gap from PUE 1.4 to 1.15 on a 50 MW IT load at industrial rates around $0.06/kWh:
| Item | PUE 1.4 | PUE 1.15 |
|---|---|---|
| Total draw (MW) | 70 | 57.5 |
| Cooling overhead (MW) | 20 | 7.5 |
| Annual cooling energy (GWh) | 175 | 65.7 |
| Annual cooling cost | ~$10.5M | ~$3.9M |
That's the gap — roughly $13 million a year of pure waste you can standardize away. The question no plant operator should skip: what capex closes it, and does it pay back inside the asset's life?
In my world, this is exactly the "last mile" discipline: the difference between the nameplate number and the meter, between the promise and the delivered hour, is where the money hides. Medellín taught me the whole thing — the trunk line is easy; the gap nobody audits is where you build the edge.
Go play with Allen's rig, and take the payback lens with you. Then pair it with the discipline in The Precision of Logistics — buffer priced right, muda hunted where it lives.