A UPS is usually bought for the server, then quietly asked to carry switches, routers, a PBX and the access control system. Runtime collapses, the battery is blamed, and the site keeps dropping during stage 4.

The fix is unglamorous: measure the real load, separate the critical path from the nice-to-have, size inverter and battery for the outage you actually experience, and monitor the battery so a failed string is a ticket rather than a Monday morning surprise.

We build every site to a documented power budget, with graceful shutdown for anything that is not on the critical path. It costs less than one lost trading day.