Colocation Power and Breaker Capacity for Game Servers: Amps, Watts and Cabinet Planning

Every dedicated-server quote tells you the CPU and RAM. Almost none tell you the number that actually limits how many machines you can run in a colocation cabinet: circuit amperage, breaker de-rating, and the thermal envelope of the rack you rented. Get this wrong and you buy ten servers, install eight, and trip a breaker on the one afternoon your game cluster actually needs to peak. This article gives you the power and breaker math to plan a cabinet that hosts a real game and high-traffic workload.

Amps, Watts, and the 80% Rule

US colocation cabinets are usually provisioned at 120V or 208V single-phase, or 208V three-phase, with circuit sizes of 20A, 30A, or 60A. The critical rule: you may only load a circuit to 80% of its rating continuously. A 30A/208V circuit gives you 24A of usable current. At 208V, 24A = 4,992 W. That is your real power budget — not the 6,240 W the raw number suggests.

Circuit	Usable (80%)	Usable watts @208V	Typical dual-socket game nodes
20A / 120V	16A	1,920 W	2–3
30A / 208V	24A	4,992 W	5–7
60A / 208V	48A	9,984 W	11–15
30A / 208V 3-phase	24A/leg	~8,600 W	9–12

Those node counts assume a realistic 700–900 W per loaded game node. A dual-socket EPYC or Xeon node with 256 GB RAM, populated DIMM slots, multiple NVMe drives, and a 10 Gbps NIC commonly idles at 350–450 W and peaks at 700–900 W. Plan on peak, not idle.

Dual PSU Does Not Mean Double Power

A server with redundant 1,100 W PSUs draws roughly 700 W under load, split across both if configured for load sharing, or full draw on one with the other passive. Either way, each PDU feed must independently support the node’s full draw, because during a PSU failure the survivor carries everything. This is why you budget per-rail: if feed A must survive all nodes at peak, feed A needs the whole cabinet’s wattage available, and so does feed B. Oversubscribed A/B feeds are the single most common colo mistake — the cabinet looks fine until one feed trips and the other immediately overloads.

Breaker Trips, Inrush, and Margin

Breakers do not trip at their rating instantly; a magnetic trip can fire on inrush current when many PSUs power on at once. If your cabinet reboots after a power event and all nodes power up simultaneously, inrush can momentarily exceed steady draw by 2–3× and trip a marginal breaker even though steady-state was within the 80% budget. Mitigation: stagger power-on, keep total steady load at or below 70% of the circuit rating so you retain inrush headroom, and prefer three-phase for dense cabinets — it delivers more usable power per rack unit of cabling and balances load across legs.

A practical target is to keep sustained draw at 65–70% of the circuit rating. At 30A/208V that is roughly 4,000 W of steady load — about five loaded nodes with a safety margin for a sixth’s inrush. This is the kind of headroom that separates a resilient game cluster from one that browns out during a tournament.

Cooling Is the Second Limit

Power in equals heat out. A cabinet drawing 5 kW must dissipate 5 kW, or 17,000 BTU/hr. Typical colocation air cooling handles 3–5 kW per cabinet comfortably; beyond that you need blanking panels, strong front-to-back airflow, and often containment or rear-door heat exchangers. Install high-power game nodes in the upper rack units only if your airflow design specifically supports it — heat rises, and the top of a poorly blanked cabinet runs hottest. Validate sustained CPU clocks after install: a node that throttles from 3.5 GHz to 2.8 GHz under load has effectively lost 20% of the capacity you paid for, and the cause is usually airflow, not silicon.

GPU and Accelerator Nodes Change Everything

If you host GPU-accelerated game tooling, dedicated game streaming, or ML-based matchmaking alongside traditional game servers, power planning shifts. A single modern accelerator card can draw 300–700 W on its own, and a 4-GPU node can pull 2,000–3,000 W. At that density, one node can consume half a 30A/208V circuit. Budget GPU nodes individually, not by average, and verify the cabinet’s cooling and PDU configuration can sustain the transient spikes of GPU workloads, which are sharper than CPU-only workloads. For GPU and high-density bare-metal options, sizing tables and provider comparisons live at bestdedicatedwebhostingserver.com.

Power Planning Checklist

1. Convert your circuit to usable watts using the 80% rule, then target 65–70% sustained load for inrush headroom.
2. Measure real per-node peak draw under a synthetic full load, not the datasheet figure.
3. Size each PDU feed to carry the entire cabinet alone.
4. Confirm cooling capacity in kW (watts in = watts of heat out) and blank every unused RU.
5. Stagger power-on and validate sustained clock speed after install.
6. For GPU/high-density nodes, plan per-node and verify transient headroom.

Power and thermal budget is the invisible spec that decides whether your cabinet runs five nodes or ten. Model it before you sign the colo contract. When you would rather let a provider own that math, dedicated bare-metal plans that include power, cooling, and uplink are compared at bestdedicatedwebhostingserver.com.