Breaker trip (rack power cascade)
Each rack has a power budget, and the breaker that feeds it doesn't care how important your servers are. Pack too many machines into one rack and the moment their combined draw stays over the limit, the breaker trips. Every server in that rack goes dark at once — and now you're waiting on it to reset and your customers are waiting on you.
This is the trap of a free U slot: there's physical room for one more host, but no power for it. The buy panel warns you near the ceiling, but discipline is on you. A tripped breaker is a self-inflicted outage that takes down perfectly healthy hardware as collateral, so leave headroom on every circuit instead of filling it to the brim.
Detailed explanation
Each rack has a power_budget_w (3kW garage, 8kW standard). A buy-time check warns at the 95% threshold. Per BillingTick, P_trip = clamp((draw / budget - 1) x 2, 0, 0.5) per hour — so the further over cap you run, the faster the trip, capped at a 50%/hour roll. There is no trip while you're under budget; the danger is sustained overdraw, not a momentary spike.
A trip cascades: ALL hosts AND switches in that rack drop offline for 4-10 sim-hours, and any host that leaves the fabric strands its VMs until it returns. The power-budget probe reports live draw vs cap per rack so you can see how much headroom each circuit has before the next purchase. Treat the cap as a hard ceiling and size your racks to leave room for PSU inrush and growth.