The short answer
Recorded video can live in three places: a network video recorder on site, a vendor's cloud, or on the camera itself under cloud management. An on-premise NVR has the lowest ongoing cost, keeps full resolution, and keeps recording through an internet outage, but the owner is responsible for drive replacement, firmware and the physical security of the appliance. Cloud recording gets rid of the appliance but adds an ongoing per camera subscription and a sustained upload requirement of roughly 1 to 8 Mbps per camera, depending on resolution. Edge storage with cloud management keeps the video on the camera and sends only the clips someone requests. That suits sites with limited upload bandwidth or no secure telecom room.
| Options | On-premise NVR, cloud, edge storage, hybrid |
|---|---|
| Upload per camera | About 1 to 8 Mbps sustained, by resolution |
| Storage rule | 1 Mbps continuous is about 10.8 GB per day |
| Outage behavior | NVR and edge keep recording; remote access stops |
| Federal funds | NDAA Section 889 and 2 CFR 200.216 apply |
| License | CSLB #472017, C-7 Low Voltage Systems |
Three places recorded video can live
Every recorded frame has to be written somewhere. On a commercial camera system there are three places to put it, plus one common hybrid.
On-premise network video recorder. This is an appliance or server with local drives, sitting in a telecom room. Cameras stream to it over the building LAN. Recording continues when the internet connection drops, video is kept at full resolution and frame rate, and there’s no per camera subscription. In exchange, the owner takes on drive replacement, firmware updates, array rebuilds and the physical security of the box. That last one matters, because an intruder who gets to an unsecured recorder can remove the only copy of the footage of the intrusion.
Cloud recording. Cameras upload continuously to a hosted platform. There’s no appliance to maintain, remote access works from a browser, and the vendor handles updates and redundancy. Two things change, though. Upload bandwidth becomes something you design for from the start, and the subscription is an ongoing operating cost that continues after the warranty ends.
Edge storage with cloud management. The camera keeps recorded video on internal storage, and the cloud provides the interface, search, alerting and clip retrieval. Only the clips someone asks for leave the building, so upload demand is a small fraction of full cloud recording. Cisco Meraki MV works this way, and we install and configure that platform directly. There are two tradeoffs. Retention is capped by what fits on the camera, and a camera that’s destroyed takes its recording with it unless the platform is set up to push motion or alarm clips to the cloud.
Hybrid: local gateway plus cloud archive. A local appliance records everything, and a cloud tier holds a lower bitrate copy, bookmarked events, or a subset of cameras. This is the usual answer when local recording meets the retention policy but the owner also needs footage to survive a fire or a theft at one site.
The cameras, mounts, cabling, PoE switching and VLAN design are the same in all four cases. What changes is the recurring cost, the bandwidth profile, the failure modes, and who has to log in and fix something at two in the morning.
Bandwidth and storage math in plain terms
Two numbers drive every recording decision, and both come from the same input: bitrate.
Bitrate is how many bits per second a camera produces after compression. It goes up with resolution, frame rate, scene motion and image noise. It comes down with H.265 instead of H.264, with variable bitrate encoding, and with lower frame rates. Manufacturers publish bitrate calculators for their own cameras, and those are more reliable than any general rule. The planning figures below are for budgeting a design before you’ve picked a model.
To get from bitrate to storage, you only need one constant: 1 Mbps recorded continuously is about 10.8 GB per day. Multiply by camera count and retention days for raw capacity, then add headroom, because parity and free space take up part of any array.
| Camera | Planning bitrate, H.265, 15 fps | Storage per camera per day, 24/7 | 24 cameras, sustained upload if cloud recorded |
|---|---|---|---|
| 2 MP (1080p) | 1 to 2 Mbps | 11 to 22 GB | 24 to 48 Mbps |
| 4 MP | 2 to 4 Mbps | 22 to 43 GB | 48 to 96 Mbps |
| 8 MP (4K) | 4 to 8 Mbps | 43 to 86 GB | 96 to 192 Mbps |
Here’s a worked example. Take twenty four 4 MP cameras averaging 3 Mbps, recording continuously for 30 days. Each camera writes 3 Mbps times 10.8, or 32.4 GB per day. Times 24 cameras, that’s 778 GB per day, and times 30 days it’s about 23 TB of raw video. Once you allow for parity and free space, the recorder needs roughly 30 TB of installed drive capacity.
The same system on a cloud platform needs about 72 Mbps of sustained upload, every hour of every day, on top of everything else the building does. That’s the number that often rules out cloud recording on sites with cable or DSL service, where a circuit advertised at 300 Mbps may deliver 20 Mbps up. Check the upload figure, and check it under load. Symmetric fiber changes the answer, so it makes sense to decide on the recording approach and the circuit together.
How site conditions point to NVR, cloud or edge
| Design input | Points to on-premise NVR | Points to cloud recording | Points to edge with cloud management |
|---|---|---|---|
| Upload bandwidth | Any; recording stays on the LAN | Symmetric fiber with headroom above the camera load | Limited, shared, cellular or wireless bridge backhaul |
| Camera count and resolution | High count, 4K, higher frame rates | Modest count at a tier the vendor prices well | Modest count per site across many sites |
| Retention requirement | Long retention at full resolution | Whatever the subscription tier buys | Short to moderate; capped by camera storage |
| Secure telecom room | A locked, cooled room exists | Not required | Not required |
| Who administers it | On-site IT, or a service agreement | Property manager or remote administrator | Property manager or remote administrator |
| Portfolio shape | Single site or campus | Many sites under one console | Many sites with thin backhaul |
| Internet outage | Recording and local playback continue | Depends on camera buffer; remote playback stops | Recording continues; remote access stops |
| Cost shape | Highest first cost, lowest recurring | Lowest first cost, highest recurring | Camera premium plus a moderate license |
On most projects, two of these inputs settle it. If there’s no secure, cooled, lockable room for an appliance, the appliance becomes a liability, and edge or cloud is usually the better fit. If the upload circuit can’t carry the sustained camera load with headroom, cloud recording is off the table until the circuit changes. Our network room buildout page covers what a room needs to host a recorder properly, including grounding, cooling and access control on the door.
Five year cost, with the assumptions stated
The figures below are California commercial planning ranges as of 2026 for a single building with 24 cameras and a 30 day retention policy. We’ve left out cameras, mounts, cabling, PoE switching, programming and labor, since those are the same across all three architectures. Use these for budgeting, not as a quote.
| Line item, 24 cameras, 30 day retention | On-premise NVR | Cloud recording | Edge with cloud management |
|---|---|---|---|
| Recording hardware, year one | About $4,000 to $12,000 for appliance and drives | None | None, or about $1,000 to $2,000 for a gateway |
| Camera premium for onboard storage | None | None | Roughly 5 to 15 percent per camera |
| Recording license, per camera per year | $0 to about $100 | About $100 to $500 | About $100 to $300 |
| Drive replacement across five years | About $1,000 to $4,000 | None | None |
| Five year recurring subtotal | About $1,000 to $16,000 | About $12,000 to $60,000 | About $12,000 to $36,000 |
| Bandwidth impact | LAN only | May force a circuit upgrade | Minimal |
A few things stand out in that table. The on-premise recorder is the cheapest over five years and the most expensive in year one. Cloud is the reverse, and the gap keeps widening every year after the fifth. Edge storage sits in between. It’s often sold as a multi year license bundled with the camera, which moves the recurring cost into the capital budget, and some owners prefer it there.
What the table doesn’t capture is staff time. An owner with no IT department often ends up paying for the NVR twice, once for the hardware and again in service calls. A failed drive in an array nobody monitors can go unnoticed until someone needs the footage. If nobody on site will check recorder health, we’d lean toward the architecture where the vendor does it, or a service and maintenance agreement that puts the check on somebody’s calendar.
Retention, resolution and the tradeoffs owners make
Retention is where the three architectures really pull apart. An on-premise recorder can be sized for whatever the policy requires by adding drives, so a 90 day policy means ordering more drives rather than redesigning the system. Cloud retention is a subscription tier, so 90 days costs roughly three times what 30 days costs, for as long as you subscribe. Edge retention is capped by the camera, so a 90 day policy usually rules edge storage out, unless the scene is quiet enough that motion based recording stretches the card that far.
So the retention policy needs to be settled before the architecture is chosen. Our guide on how long to keep security camera footage works through what sets that number, including insurer requirements, public records obligations and preservation duty.
Two adjustments stretch retention on any architecture. Dropping the frame rate from 30 to 15 fps roughly halves the storage on most scenes and rarely costs anything an investigator needs. Motion based recording on interior cameras that see no activity overnight can cut storage substantially. A lobby in a busy building, though, records nearly continuously and should be budgeted that way. Camera count and resolution matter more than either adjustment, so it helps to run the count method in our guide on how many security cameras a building needs before sizing storage.
Federal funding narrows the platform list
If any federal money touches the project, the recording decision is also a procurement decision. Section 889 of the FY2019 NDAA bars federal agencies from procuring covered video surveillance and telecommunications equipment from Huawei, ZTE, Hytera, Hangzhou Hikvision, Dahua and their subsidiaries and affiliates, and 2 CFR 200.216 extends that prohibition to federal grant and loan funds. That’s how the rule reaches HUD assisted housing, agency buildings with federal participation, and many campus and clinic projects.
The easy mistake to make is white label hardware. A large share of low cost cameras and recorders sold under other brand names are made by Hikvision or Dahua, and the carton doesn’t say so. That applies to cloud and edge platforms as much as to recorders, because a cloud service is only as compliant as the cameras feeding it. Specify the actual manufacturer and model, and require a written model level compliance statement from the manufacturer with the submittal. Our guide to NDAA Section 889 compliant cameras in California covers how to audit an existing system and what specification language holds up in an audit.
Public agencies should also settle data residency and export before picking a cloud platform. When a public records request for footage comes in, the agency is responsible for meeting the response deadline, not the vendor.
Next step
Send us the floor plan, the camera count if you have one, the retention you need and the upload speed at the site. We’ll come back with a recording recommendation that shows the storage sizing and bandwidth. You can start at request a bid. Our full scope is on the security camera systems page, and general contractors building a Division 28 bid list can start at for general contractors.