The rule I give anyone wiring a freezer room is short: the switch stays warm. Everything else — cable jacket, glands, heated housings, the 100 m arithmetic — follows from that decision. Cold does not break PoE. It breaks two assumptions people carry into a cold room without noticing: that copper behaves the same at every temperature, and that a port which powered up once will power up again after a defrost cycle.
Last February I stood in a frozen-food warehouse at -22 °C, looking at an 8-port switch bolted to the wall inside the freezer because it saved forty metres of cable. Eleven months of service. Then a door was propped open overnight during a delivery, the room came up to about -8 °C, and by morning three ports were dark. The PSU was fine. The switch was fine. There was water in the punch-down block and one pair read 14 kΩ to the next. Six cameras had been running on a design that was wrong from the start; it just took a bad night to prove it.
Warm Switch, Cold Device, and Nothing In Between
A commercial PoE switch is rated somewhere around 0 to 50 °C. Industrial units go to -40 °C and they are not lying, but the rating tells you what the silicon survives, not what the enclosure survives. The first thing to fail in a freezer is usually the fan — sleeve bearings stiffen, and a fan that stalls at -20 °C takes the thermal design with it. The second is every electrolytic capacitor: ESR climbs as the electrolyte thickens, ripple rating drops with it, and a PSU with 30 % margin at room temperature runs out of margin exactly when six camera heaters switch on together.
Neither of those is what actually kills the install. What kills it is condensation on the power cycle. A board sitting at -22 °C that gets switched off, opened or serviced draws moisture onto every surface, and in a room that never warms up, that moisture never leaves. It sits in the flux residue and the connector bodies until something corrodes. In a warm cabinet you get a dry-out window every afternoon. In a freezer you get one chance.
So the switch goes outside. On the warm side of the insulated panel, in the nearest conditioned space, in a proper cabinet with the door closed. If that pushes the run past 100 m, the answer is a media converter and fibre through the wall, or a properly placed extender on the warm side — not a longer copper run, and not a switch in the cold. The 100 m channel budget does not relax because the room is cold; patch cords, the wall penetration and the tray bends all count against it.
Cold Copper Measures Better. That Is the Trap.
Here is the part that surprises people, including me the first time I metered it. Copper gets better when it gets cold. Resistivity falls at roughly 0.393 % per °C, so a run measuring 18.8 Ω of loop resistance at 20 °C measures about 15.5 Ω at -25 °C. The freezer is, electrically, the best environment in the building.
That is why freezer installs get signed off in winter and fail in summer. The cable does not only live in the freezer. It lives in the ceiling void above it, in the riser, in the tray across the roof — and a ceiling void in Shenzhen in July is a 50 °C oven. A run that ate 9.3 V inside the cold room is eating 12.6 V up in the void.
| Where the copper sits | Loop resistance, 100 m of 24 AWG | Drop at 600 mA (802.3at) | Volts at the camera, from a 53.5 V PSE |
|---|---|---|---|
| Inside the freezer, -25 °C | ≈ 15.5 Ω | 9.3 V | 44.2 V |
| Bench reference, 20 °C | ≈ 18.8 Ω | 11.3 V | 42.2 V |
| Ceiling void, 50 °C | ≈ 21.0 Ω | 12.6 V | 40.9 V |
802.3at expects the PD to keep working down to 42.5 V at its own terminals. Read the middle row again: at 20 °C, a plain 100 m run of 24 AWG is already below that line at full 600 mA. The freezer section is the only reason the numbers ever looked healthy. Put that same cable in a hot void and you are 1.6 V under the floor; the camera will not tell you why — it reboots at 3 a.m., or drops its IR array, or negotiates down and looks dim at night.
I got this wrong once. A site kept losing one camera for a few minutes every afternoon. I replaced the camera, then the port, then half the run. It was 12 m of cable in a steel tray above the freezer roof, in direct sun, crossing 50 °C only when the building load peaked. The camera was never the problem; my assumption that a short run has a small drop was.
The Heater Is on the Same Port as the Camera
Nobody budgets for the heater. It is not a line on the camera spec sheet — it is buried in an accessory housing datasheet — and it turns on exactly when the room is coldest, which is to say always. Blowers and heated windows run 6 to 10 W in the sizes I have measured. A PTZ with a heater and a fan can pull 25 to 30 W on its own before the IR array kicks in.
802.3at gives you 30 W at the PSE and 25.5 W at the PD. That is the whole budget. A 12 W camera plus an 8 W heater is comfortable. A 20 W PTZ plus a 9 W heater is 29 W at the device, so the port either refuses to classify it or classifies it and folds under load. People then conclude that "PoE is unreliable in cold rooms," when the actual statement is "you asked a 25.5 W class to feed a 29 W load."
| What is actually on the port | Draw I would plan for | Port class needed | Headroom |
|---|---|---|---|
| Fixed camera, no heater | 6–12 W | Class 3 (12.95 W at PD) | Workable, thin on 12 W |
| Fixed camera + heated housing | 14–22 W | Class 4 (25.5 W at PD) | OK |
| PTZ + heater + blower | 28–34 W | Type 3 (51 W at PD) | Required, not optional |
| Door intercom, heated faceplate | 18–26 W | Class 4, Type 3 if it drives a strike | Check the strike |
A camera that has sat at -25 °C for six hours pulls more current for the first few seconds than it does at steady state, so a port sized at 96 % of its class is a port that will fail in the freezer and pass on the bench. Stagger the enable times if you stack several of these on one switch.
Water in the Plug: Condensation Kills the Handshake
PoE does not just push volts down a pair. Before it energises anything, the PSE applies a low voltage across the pair, measures the current, and looks for a signature resistance near 25 kΩ. Outside the 19–26.5 kΩ window the port stays off — no fault, no log entry worth reading, just a port that never comes up.
Now put a freezer door in the picture. Room at -25 °C, outside air at 25 °C and 60 % RH. The dew point of that air is about 16.7 °C, and in that room everything is below it. Every door opening drags a slug of it across the connector and the gland, and it does not evaporate afterward, because the room never gets warm enough to dry it.
A wet RJ45 does two things to the handshake. Water bridges the pairs, so your clean 25 kΩ signature now sits in parallel with a leakage path — 25 kΩ parallel with 60 kΩ reads 17.6 kΩ, outside the window, and the port never powers. And the IDC contacts corrode, which pushes resistance the other way over months. I have pulled plugs with green on the blades that still passed a continuity test and still refused to power a camera, because continuity at 1 mA and 100 kΩ of leakage at detection voltage are not the same measurement.
This is the one failure I fix by geometry rather than by specification: terminate on the warm side when the length allows, point every gland downward with a drip loop, seal both faces of the penetration, and never leave a plug hanging where melt water collects. For the detection window in detail, I wrote the 25 kΩ handshake piece; the short version is that a port which stays dark is usually measuring resistance, not voltage.
Jackets, Glands and the Freeze-Thaw Year
Standard PVC jacket is not a freezer cable. Most commercial Cat5e and Cat6 datasheets put the cold bend test around -15 °C, and below that the compound goes stiff, then brittle. A cable dressed at 20 °C and never touched again usually survives. The same cable, moved in January to add one camera, cracks at the bend — and you find out a year later when a pair fails.
For the cold side I specify a cold-flexible jacket — TPU or a low-temperature LSZH compound, rated to -40 °C on paper — and I double the bend radius the datasheet asks for while pulling. Most Cat6 sheets want four times the outer diameter; in a freezer I work to eight. Water expands about 9 % when it freezes, enough to split a jacket at a gland tightened in summer and filled in winter. Stainless or hot-dip galvanised tray, not painted steel.
Seal the penetration from both sides. Warm, moist air travels along the outside of a cable inside a conduit, arrives at the cold end and freezes there. I have opened a wall sleeve iced solid for 300 mm with the cable inside it. That is not a PoE fault until the ice crushes the jacket — and then it is.
What I Measure Before I Sign Off a Freezer Room
None of this is exotic. It is forty minutes with a meter and a thermometer, done on the day the room is cold and the cameras are on.
- Record the ambient at both ends at the same time — PSE cabinet intake and the coldest point on the run. Every later argument depends on those two numbers.
- Measure port voltage at the PD under full load, with every heater running. Not at the switch. At the device, or as close as the housing allows.
- Convert that to loop resistance: (VPSE − VPD) ÷ I. At 0.6 A, every ohm is 0.6 V. Compare against the 15–19 Ω you expect for the length; higher means cable or a connector, not the switch.
- Read the per-port wattage from the switch if it reports it, and compare against the housing datasheet plus 10 %. If the number is above 25.5 W, the port class is wrong, not the camera.
- Cycle each port five times cold. Count how many times it fails to come back. One failure in five is a water problem, not a firmware problem.
- Insulation test the run with the power off and the far end disconnected. A 250 V megger should read well above 20 MΩ. Anything in single digits means moisture is already in the run and you are on a clock.
- Come back in sixty days and re-torque, re-seat and re-photograph the terminations. The first winter is when the freeze-thaw damage happens; the second is when you find out.
The number I care about most is the fourth, because it is the only one that catches the heater. Voltage drop is arithmetic you can predict. A housing drawing 40 % over its label because the blower cycles on a thermostat is something only a watt reading tells you.
Questions I Keep Getting Asked
Does PoE work at -40 °C?
The cable and the powered device can, if both are rated for it. The switch should not be there: detect, classify and hold 30 W is a job for silicon that expects to be warm, and the condensation on every service visit is worse than the cold.
Can I use an industrial switch inside the freezer anyway?
You can, and I have seen an industrial unit work for years in a room that stays sealed and never cycles. I have also seen one power-cycled twice during a maintenance window, never fully dried out, and a fan that seized at -20 °C and took the thermal margin with it. If the door opens forty times a shift, keep the switch out of the room.
The camera is fine until the door opens. Why?
Almost never power. A PoE port that has classified and energised does not renegotiate because the lens fogged. If the stream drops when the door opens, look at the connector at the camera and the cable at the door frame — thermal shock plus a stiff jacket is a mechanical fault, and the fix is a drip loop and a re-termination, not more watts.
Do I need 802.3bt in a cold room?
Only when the heater and the PTZ together cross 25.5 W at the device. That is more common than people expect — a PTZ with a blower is easily 30 W — but it is a port-sizing question, not a cold-weather one. Buying Type 4 ports to "be safe" on a 12 W camera just moves the heat into the PSU.
Is an extender or a switch in the freezer ever acceptable?
An extender, sometimes, if it is rated for the temperature and feeds one cold-side run from the warm side. A switch, almost never. Both are extra power conversions in a room nobody wants to open in January.
If you are building one of these rooms now, spend your time on two decisions: where the switch lives, and what class each port needs with the heater running. The rest is cable and sealant — cheap, compared with a camera that reboots every afternoon in July because someone signed off a run in January. I test this hardware on a bench in Shenzhen all week; the bench is always 24 °C and always lies about cold rooms. Go stand in the freezer with a meter.