Footage has to live somewhere, and the two options behave differently in ways that only become obvious at the moment you need the recording.
The on-site recorder
A box in the building holding the footage. Bandwidth is not a constraint because the video never leaves the site. Retention is a matter of how much disk you fit. There is no ongoing fee for the storage itself, and the system keeps recording if the internet drops.
The weaknesses are physical. A recorder can be stolen, which has obviously happened to somebody’s evidence before, and it can fail. Both are addressable, by putting it somewhere secure and by not treating a single disk as a backup, but they have to actually be addressed.
Cloud storage
Footage is uploaded off site. It survives the building, which is the real argument for it, and remote access tends to be simpler because it was designed in rather than bolted on.
The constraint is upload bandwidth, and it is a harder constraint than most people expect. Continuous upload from a meaningful number of cameras is a large sustained load, and many commercial connections have far less upload than download. Cloud platforms deal with this by uploading only events, or by uploading a lower quality stream, which is a reasonable trade as long as you know that is the trade.
The other consideration is that the storage is a subscription. Stop paying and the retention goes with it.
The arrangement most sites land on
Record locally at full quality, push events or a reduced stream off site. You get retention that does not depend on the connection, plus a copy that survives the building. It costs more than either alone and it removes the failure mode each one has on its own.
The question that actually decides it
What is the realistic upload speed at the site, measured rather than quoted? Everything else follows from that. A site with poor upload can still have excellent video, but it will be recorded locally, and any proposal that ignores this will disappoint in a specific and predictable way.
Getting footage out when it matters
Whichever you choose, the moment that counts is the one where somebody asks for a copy. That process should be tested before you need it, because it is more awkward than people expect.
Check that you can export a specific time range rather than only short clips. Check the export plays on an ordinary computer without a proprietary viewer, or that the viewer travels with the file. Check the exported quality matches the recorded quality, because some systems quietly downscale on export.
And check who is able to do it. If one person knows how, and the incident happens while they are away, the retention you paid for does not help.
The backup nobody has
A single recorder with a single disk is not a backup, and disks fail. Whether that matters depends on the site, but it should be a decision. For most commercial properties, a second copy of the footage that would actually be needed in a serious incident is worth more than a few extra days of retention on everything.
Where the footage lives gets decided against the actual connection at the site, which is something Elite Security checks rather than assumes. More like this in the blog, and commercial CCTV is where to start.