NVR vs Cloud vs Edge: Where Recorded Video Should Live

Three places recorded video can live#

Recorded frames have to be written somewhere, and on a commercial camera system there are three places to put them plus one common hybrid.

On-premise network video recorder. An appliance or server in a telecom room with local drives. Cameras stream to it over the building LAN. Recording continues when the internet connection drops, video is retained at full resolution and frame rate, and there is no per camera subscription. The owner takes on drive replacement, firmware updates, array rebuilds and the physical security of the box, because an intruder who reaches an unsecured recorder can remove the only copy of the footage of the intrusion.

Cloud recording. Cameras upload continuously to a hosted platform. There is no appliance to maintain, remote access works from a browser, and the vendor handles updates and redundancy. Two things change. Upload bandwidth becomes a design input rather than an afterthought, and the subscription is a permanent operating cost that does not end when the warranty does.

Edge storage with cloud management. The camera holds recorded video on internal storage and the cloud provides the interface, search, alerting and clip retrieval. Only the clips someone actually asks for leave the building, so upload demand is a small fraction of full cloud recording. Cisco Meraki MV works this way, and Telelink installs and configures that platform directly. The tradeoffs are that retention is capped by what fits on the camera, and a camera that is destroyed takes its own recording with it unless the platform is configured 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 satisfies 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 to 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 rises with resolution, frame rate, scene motion and image noise, and it falls 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 figures are more reliable than any general rule. The planning figures below are for budgeting a design before the model is picked.

Storage converts from bitrate with 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 consume 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

Work an example. Twenty four 4 MP cameras at an average 3 Mbps, recording continuously for 30 days: 3 Mbps times 10.8 is 32.4 GB per camera per day, times 24 cameras is 778 GB per day, times 30 days is about 23 TB of raw video. Allow for parity and free space and 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 is the number that kills cloud designs 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 completely, which is why the recording decision and the circuit decision belong in the same conversation.

Choosing by site conditions, not by preference#

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

Two inputs decide most projects. If there is no secure, cooled, lockable room for an appliance, the appliance is a liability and edge or cloud wins. If the upload circuit cannot 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. Cameras, mounts, cabling, PoE switching, programming and labor are the same across all three architectures, so they are excluded. These are for budgeting, not quoting.

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

Three things fall out of that table. The on-premise recorder is cheapest over five years and most expensive in year one. Cloud is the reverse, and the gap widens every year after the fifth. Edge storage sits between them and is often sold as a multi year license bundled with the camera, which moves the recurring cost into the capital budget where some owners would rather have it.

What the table does not capture is staff time. An owner with no IT department pays for the NVR twice: once for the hardware and again in service calls, because a failed drive in an array nobody monitors goes unnoticed until someone needs the footage. If nobody on site will check recorder health, buy the architecture where the vendor does it, or buy a service and maintenance agreement that puts the check on somebody’s calendar.

Retention, resolution and the tradeoffs owners actually make#

Retention is where the three architectures diverge in practice. An on-premise recorder can be sized for whatever the policy requires by adding drives, so a 90 day policy is a purchase order, not a redesign. Cloud retention is a subscription tier, so 90 days costs roughly three times what 30 days costs, forever. 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.

That is why the retention policy has to be settled before the architecture is chosen, not after. Our guide on how long to keep security camera footage works through what actually sets the number, including insurer requirements, public records obligations and preservation duty.

Two adjustments buy retention on any architecture. Lowering 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, though a lobby in a busy building records nearly continuously and should be budgeted as if it does. Camera count and resolution matter more than either, which is why the count method in our guide on how many security cameras a building needs is worth running before the storage is sized.

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 is how the rule reaches HUD assisted housing, agency buildings with federal participation, and many campus and clinic projects.

The trap is white label hardware. A large share of low cost cameras and recorders sold under other brand names are manufactured by Hikvision or Dahua, and the carton does not say so. This 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 selecting a cloud platform, because a public records request for footage is answered on the agency’s clock, not the vendor’s.

Next step#

Send the floor plan, the camera count if you have one, the retention you need and the upload speed at the site, and we will return a recording recommendation with the storage sizing and bandwidth shown. Start at request a bid. The full scope is on our security camera systems page, and general contractors building a Division 28 bid list can start at for general contractors.