Server Cooling vs Power-Plant Cooling
A data center faces two cooling problems: keeping its chips cool, and paying for the water spent at the power plant that feeds it. They are separate loads.
Two loads, often conflated
When people discuss data-center water, they mix two things. The first is server cooling: removing heat from the chips inside the building. The second is power-plant cooling: the water evaporated at whatever plant supplies the electricity. The second is usually larger, because generating the power rejects more heat than the chips themselves dissipate, but it is off site and easy to overlook.
Why the plant load dominates
A steam plant rejects roughly twice as much heat as the electricity it delivers. So for every unit of power a data center consumes, the upstream plant may reject about two units of heat, much of it through evaporation. On-site server cooling handles only the one unit that reached the building. The upstream water is the bigger, hidden number.
What MetroVolt changes
- The upstream plant water essentially disappears, because the burner has no steam cycle
- On-site server cooling still exists and can itself be designed for low water
- Co-locating generation and load lets the two share dry-cooling infrastructure
The reframing
MetroVolt attacks the larger, upstream water load directly by removing the steam cycle from generation. It does not by itself change how a given server rack is cooled, but by supplying firm power with almost no generation-side water, it removes the part of the footprint that most site plans cannot control.