field note

How long do security cameras keep footage?

Security cameras keep footage until storage fills or a retention limit hits, then overwrite the oldest. How to work out what your system really keeps.

2026-10-18

Security cameras keep footage until their storage fills or a set retention limit is reached, then delete the oldest recordings to make room. In practice that ranges from a few days on a small memory card inside the camera to 30, 90, or more days on a recorder sized for the job. The number that matters is not the one in the retention setting. It is the age of the oldest recording you can actually play back today, camera by camera, and on a busy commercial CCTV system that is often shorter than anyone thinks.

This post covers what decides retention, the arithmetic behind it, why retention shrinks after install, and how to check what a fleet of 100 to 500 Axis or Hanwha cameras really keeps.

Where the footage lives decides the ceiling

Where it recordsWhat limits retentionWhat happens when full
Memory card in the cameraCard size and the camera’s bitrateOldest recordings are overwritten if overwrite is on; recording stops if it is off
Recorder or VMS serverDisk capacity, the retention setting, and every camera’s bitrate togetherOldest recordings are deleted to make room
Cloud serviceThe plan’s retention tier and upload bandwidthRecordings older than the tier are removed

Most commercial fleets record to a recorder or VMS server and use the camera’s card, if fitted, as a backup for when the network or recorder is down. The server’s disk is the real ceiling. The retention setting is a maximum, and the disk enforces the minimum.

The arithmetic

Storage is bitrate multiplied by time. A camera streaming at 4 megabits per second writes 43.2 gigabytes a day: 4 Mbps times 86,400 seconds, divided by 8 bits per byte.

Average bitratePer camera per dayPer camera for 30 days200 cameras for 30 days
2 Mbps21.6 GB0.65 TB130 TB
4 Mbps43.2 GB1.3 TB259 TB
8 Mbps86.4 GB2.6 TB518 TB

These figures assume continuous recording. Motion-only recording writes less, in proportion to how much of the day the scene is moving, which is the number most estimates get wrong. A loading dock or a lobby is moving most of the working day. For a planning estimate across a mixed fleet, the camera storage calculator does this arithmetic per camera group, and it recommends planning 15 to 20 percent above the result.

Why the retention you set is not the retention you have

Retention is set once, at install, against the bitrates the cameras had that week. Then the system changes and the disk does not.

Bitrate drifts up. Most cameras use variable bitrate, so the stream grows with scene complexity. Night-time sensor noise, rain, trees in wind, a camera re-aimed at a busy road, a resolution raised during troubleshooting and never put back, a firmware update that changes a stream profile: each one lifts the average, and retention falls in proportion.

Cameras get added. Ten more cameras on the same recorder share the same disks. Nobody re-runs the storage math for a ten-camera add.

Disks fail. A failed or degraded drive reduces capacity or stops recording on the volume it held. On a recorder that nobody logs into, a warning light on the front panel can sit there for weeks.

Motion settings change. Motion recording sized for a quiet scene becomes near-continuous recording when the scene gets busy, or records nothing when someone adds a mask over the area that matters.

Three bars on the same scale from zero to 30 days. The first is the design: 200 cameras at an average of 4 Mbps on disks sized for 30 days, and it holds 30 days. The second is the same system after night noise and busier scenes lift the average to 5.7 Mbps: it holds 21 days, and the last 9 days of the setting are lost. The third adds 40 cameras on the same disks: it holds 17.5 days, and 12.5 days are lost. same disks, same 30-day setting, example fleet 30-day setting as designed: 200 cameras at 4 Mbps 30 d night noise lifts the average to 5.7 Mbps 21 d 40 cameras added, no new disks 17.5 d
The hatched span is footage the policy says exists and the disks no longer hold. Nothing alarms when retention shrinks; the oldest recordings are deleted on schedule.

None of these raises an alarm. The recorder does what it was built to do: when the disk is full, the oldest footage goes. The first sign is usually a request for an incident from three weeks ago that the policy says is kept and the disks no longer hold.

How long you are required to keep it

The retention you need is set by whoever licenses or audits you: a banking regulator, a gaming commission, a cannabis board, a payment card standard, an insurer, or your own policy. Requirements and their sources by industry are collected in video retention requirements. Size to the longest requirement that applies to each area, not to an average across the site, and remember that a legal hold on an incident clip means exporting it before the rolling window deletes it.

How to check what you actually keep

A check that takes an afternoon per site and catches most retention shortfalls:

  1. Pick the oldest required age. If the policy is 30 days, the test is whether footage from 30 days ago plays back today.
  2. Check the oldest recording per camera, not per recorder. Open playback for each camera and find the earliest footage available. On a recorder with mixed cameras, the busy ones run short first.
  3. Compare each camera’s real bitrate with the design. The design assumed a number. The camera is producing a different one, especially at night. Cameras running well above the design figure are where retention is going.
  4. Check disk and volume health on every recorder. A degraded array or a failed drive is a retention problem before it is a hardware ticket.
  5. Check the cards in the cameras. A card that is missing, failed, or worn out removes the backup copy for the times the network or recorder is down.
  6. Write the result down with a date. An auditor or an insurer asks what was kept, and when it was last verified.

Repeat it after any camera is added, any stream setting changes, and at least quarterly. Retention only shrinks between checks.

Keeping retention honest across 500 cameras

By hand, this check is one person clicking through playback for every camera. At 500 cameras across several sites it does not happen often, which is how a 30-day policy becomes a 17-day reality without anyone deciding it.

MentatNOC covers the parts of this that are about the cameras. The storage tile on the camera coverage map sizes retention against the cameras actually on the plan, with each camera’s bitrate and recording mode, so the storage math moves when the plan does. Cards in the cameras are watched continuously: a card that disappears or reports a failure raises an alert, and on cameras that report card wear, so does a card nearing the end of its life. Cameras that go offline, freeze, go black, or drift off time are caught the same hour rather than at the next playback check. It monitors device health, not video, so the age of the oldest recording on the recorder stays the VMS’s number to report, and the checklist above is how to read it.

For an integrator, a missing or failed card alert names the camera and site, so the tech carries a replacement card on the first trip and the visit is billable. See the fleet view and coverage plans on the interactive demo.