I keep seeing the same Zigbee question phrased as a hard number: what is the maximum number of devices Home Assistant can handle? The answer is not 32, 50, 100, or any other number you can paste into a shopping list.
A good Zigbee network is a mesh, not a USB hub. The coordinator is important, but the routers, placement, radio interference, and traffic pattern decide whether device 80 feels boring or whether device 20 feels broken. Community reports range from small cheap dongles struggling with a handful of devices to properly built networks running 80, 200, or more. That variation is the clue: capacity is a system property.
This guide explains the limits that actually matter, how to size a network before you buy hardware, and what I would change when a growing Home Assistant installation starts dropping devices. It complements my best Zigbee coordinators for Home Assistant guide, which covers specific radios.
The numbers people confuse
There are several different limits hiding behind the word capacity.
Direct children are devices that have selected the coordinator or a router as their parent. A radio stack may have a limit on how many children one parent can maintain. That is where the old 32-device advice usually comes from. It does not mean the entire network stops at 32. A router can carry its own children, and the mesh can contain many routers.
The total device limit is the number of nodes the coordinator and software can address. The exact ceiling varies by chipset, firmware, integration, and how much memory is available. It is a useful engineering limit, but it is not the first limit most homes hit.
The practical traffic limit is often the real one. A network full of sleepy temperature sensors sends little data. A network with dozens of power-monitoring plugs, frequent motion automations, bulbs reporting state, and repeated availability polling creates much more airtime. Two networks with the same node count can behave completely differently.
I would therefore plan in tiers. Under 50 devices should be easy with a decent coordinator and a few mains-powered routers. Between 50 and 100, placement and routing start to matter. Between 100 and 200, I would deliberately design the mesh instead of adding devices opportunistically. Beyond 200, check the documented limits of your coordinator and integration, and consider whether traffic, not raw count, is the problem.
The coordinator is only the beginning
My default starting point is the SONOFF Zigbee 3.0 Dongle Plus-E (also on Amazon). The Home Assistant Connect ZBT-2 is the better hands-off choice if you want official hardware and firmware management. Either can be the right answer for a normal house.
The mistake is buying the biggest-looking coordinator and then plugging it directly into a noisy mini PC. Zigbee uses the 2.4 GHz band. USB 3 interference, Wi-Fi channel overlap, a metal server cabinet, and a coordinator hidden behind the host can do more damage than a modest device count.
Use a USB 2.0 extension cable and put the radio away from the host, USB 3 cables, and Wi-Fi access points. If the Home Assistant machine lives in a rack or basement, a network-attached coordinator can be a sensible layout choice, but it adds another service and another network dependency. I would first fix local placement before introducing that complexity.
A coordinator also needs a stable firmware and a supported integration. Zigbee2MQTT gives me more visibility into links, routes, and device support. ZHA is easier to keep inside Home Assistant. Neither choice magically increases capacity. A healthy mesh running the platform you understand is better than a theoretically more tunable platform you do not maintain.
Routers make the network scale
Every mains-powered Zigbee device is not automatically a good router, but good routers are what turn a small star into a usable mesh. Smart plugs and dedicated repeaters are usually safer choices than assuming every bulb will route well. My practical baseline is one reliable router near the coordinator, one on each floor, and additional routers near walls, outbuildings, or dense groups of battery sensors.
The ThirdReality Zigbee smart plug is useful because it is both a real outlet and a router. A dedicated repeater can make sense where you do not need another switched load. Do not scatter ten routers randomly and assume more is always better. Poor routers, redundant paths, and devices that behave badly after power loss can make troubleshooting harder.
A good mesh has overlap. A sensor at the edge of the house should have more than one plausible path back to the coordinator. It should not need to cross one router in the laundry room and then fail whenever that router is unplugged. Put routers where the radio can reach the surrounding rooms, not merely where a spare outlet happens to exist.
Battery sensors are end devices. They sleep to save power, so they do not extend the mesh for anything else. This is why adding twenty battery sensors to a weak network does not improve coverage. Add mains-powered routing nodes first, then pair the battery devices where you want them.
How I would size a new installation
For a small apartment, I would start with one coordinator, two or three good routers, and leave room for growth. For a typical house, I would buy the coordinator, put it on an extension cable, and place a router on every floor before pairing a large batch of sensors. For a detached garage or workshop, I would treat the building as a radio design problem rather than hoping the coordinator reaches through exterior walls.
Pair a few devices in their final locations and watch the network before buying the rest. If those devices have poor links, fix the mesh now. A staged rollout also makes it obvious which new router or powered device changed the topology. Pairing 80 devices on day one makes every later problem look like a random failure.
Keep a written inventory of device type, room, parent or route when available, and whether it is mains powered. When a device drops, that information is more useful than the total count. If all the unreliable devices are in one wing, you need coverage. If only chatty plugs cause delays, you need to investigate traffic and reporting. If everything fails after a firmware update, the count is probably innocent.
What to do when devices start dropping
First, stop adding devices. Check coordinator placement, Wi-Fi channel overlap, and whether the host or USB cable recently changed. Then inspect the mesh map, but do not treat a snapshot as an absolute truth: routing information can be incomplete or stale depending on the integration and device.
Next, add or reposition one known-good router. Wait for the mesh to settle, then test the troublesome room. If that improves things, repeat methodically. Do not replace the coordinator before testing the cheap physical fixes. I have seen a seven-dollar extension cable solve a problem that looked like a coordinator capacity problem.
If the network is large and busy, reduce unnecessary reporting. A power sensor that reports every few seconds can consume airtime without making your automations better. Use sensible thresholds and intervals. Review automations that trigger cascades of commands to many bulbs or plugs at once. A large Zigbee network should be quiet when nothing is happening.
Finally, check the coordinator documentation and integration logs before migrating. A second coordinator creates a second Zigbee network. That can be useful for a detached building, but it does not combine into one larger mesh, and it increases your inventory and backup burden. Migration is a last resort after placement, routers, interference, and traffic have been tested.
A note about sleepy devices and recovery
Battery devices make capacity planning look worse than it is because they are intentionally quiet most of the time. A door sensor that reports two state changes per day is not equivalent to a smart plug that reports voltage, current, power, and energy continuously. Count devices, but also classify their behavior. I would rather run 120 quiet sensors on a stable mesh than 40 noisy devices with aggressive reporting and automations that constantly poll them.
Recovery behavior matters too. After a coordinator restart or power outage, many devices may check in close together. A robust mesh recovers gradually; a marginal mesh can look dead for several minutes while routes rebuild. Give the network time before repairing devices. Re-pairing a healthy device during recovery can create a second problem by changing its parent and adding more churn.
If you maintain backups of your Zigbee network data, keep them with your Home Assistant backups and test that you know how to restore them. Capacity is not just the number of devices online today. It is also how safely you can recover the network after a host, coordinator, or integration failure.
The Bottom Line
Home Assistant can handle far more Zigbee devices than the old 32-device warning suggests. For most homes, the practical target is 100 or more devices on one well-placed coordinator with a deliberately built router mesh. The number on the box matters, but the network around the box matters more.
Buy a supported coordinator, move it away from USB 3 noise, install real routers before you fill the house with battery sensors, and add devices in stages. If you follow that order, you will usually run out of rooms before you run out of Zigbee capacity.