Cloud vs On-Premises Video: How to Choose

Cloud video as a service removes the recording server, the storage array, the operating system and database licensing, the uninterruptible power supply, the rack and the IT labor to maintain all of it. It also removes port forwarding, which is the single largest attack surface in traditional recorder deployments and the mechanism behind most publicly reported camera compromises.

On-premises recording keeps recurring cost lower once camera counts get large, keeps the video physically in the building, and does not depend on upstream bandwidth. For a single facility with high camera density and a competent IT group, that is often still the right call.

Cloud video commonly runs about 3 to 10 dollars per camera per month at an entry tier with 7 to 14 day retention, 10 to 30 dollars at a standard tier with 30 day retention, and 30 to 60 dollars at a premium tier with 60 to 90 day retention and full analytics.

On-premises video management licensing commonly runs roughly 150 to 400 dollars per channel depending on tier, with an annual software maintenance agreement commonly in the range of 15 to 22 percent of license value. Add the server, storage, power protection and the IT time to keep it patched.

State the obvious caution again. No major manufacturer publishes list pricing in this category. These are reseller and comparison estimates for budgeting only, and your quoted numbers will differ.

Almost nobody publishes this honestly, so here it is in order of impact. Retention period is the single largest cost lever in cloud video, ahead of everything else on the list.

Some cloud platforms only license their own cameras. That is a clean, easy deployment on day one and an expensive problem at renewal, because there is no path to reuse those cameras with another platform later. If pricing changes or the product direction changes, your only options are to accept it or to replace the hardware.

Ask one question of any cloud platform before you buy. If we leave you in year four, what happens to these cameras. If the answer is that they stop working, price that outcome into the decision now.

The most common correct design puts cloud at the remote and small sites where there is no rack, no server room and no local IT, and keeps on-premises recording at the high-camera-count main facility. Both are then federated into one interface so an operator does not care which is which.

That design also matches how organizations actually grow. New sites land in the cloud without a truck full of hardware, and the main facility keeps the storage economics it already paid for.

A well-engineered cloud camera can idle at as little as 30 kilobits per second, sending only metadata and thumbnails, and switch to full resolution only when a person or vehicle is classified in scene. In practice that can cut upstream consumption by 30 to 80 percent against always-on systems.

For live viewing, budget roughly 0.6 megabits per second per 2MP camera and 1.2 megabits per second per 5MP camera per concurrent viewer. Archive upload is additional and should be sized separately, because it is the number people forget until the first month's report.