The Tuesday Morning Problem
You finish a VSaaS deployment at a warehouse. Verkada or Milestone in the cloud, looks clean on the demo. Facilities manager is happy. IT signed off. Then Tuesday morning at 9:47 AM, the carrier has a brief hiccup. Twenty minutes of downtime, nothing catastrophic by ISP standards. The cloud system goes dark. No live view. No new footage uploading. By the time the connection comes back, you've got a gap in the recording.
The customer calls. Security asks why there's no video of the loading dock during that window. Facilities wants to know if this is normal. IT is now asking harder questions about SLA and failover. You're explaining that cloud surveillance depends on the uplink being solid, which it isn't, and that's not really a system problem.
This happens on maybe half the jobs where someone spec'd pure VSaaS without looking at the actual network first. The other half work fine because the uplink is genuinely bulletproof. The difference isn't the camera or the platform. It's whether you've got a plan for when the internet isn't there.
Why Pure Cloud Feels Right Until It Doesn't
Cloud video surveillance solves real problems. No NVR to maintain. Footage lives in a data center with redundancy. Multi-site monitoring from one dashboard. No PoE budget constraints. The pitch is clean.
What doesn't get mentioned in the demo:
- Upload bandwidth is almost always the bottleneck, not download. A 100 Mbps fiber connection might have 10 Mbps upload. Eight IP cameras at 2-4 Mbps each can consume most of that.
- ISPs throttle video traffic at peak hours. Not always, but often enough. Retail, office, warehouse traffic patterns hit hardest between 10 AM and 2 PM, exactly when security matters most.
- Carrier outages, maintenance windows, and flaky connections are not rare edge cases. They happen monthly at most multi-site deployments.
- Failover to cellular backup is expensive and slow. It's not a real solution for continuous recording.
- If your retention policy is 30 days and the cloud service goes down for 48 hours, you don't get those two days back. They're gone.
None of this means cloud is wrong. It means cloud alone is incomplete on most commercial jobs.
The Hybrid Architecture That Actually Works
Hybrid recording means local NVR storage plus cloud sync. Here's what that looks like in practice.
Every site gets an on-premise NVR. Doesn't have to be fancy. A 4-bay or 8-bay box running Milestone, Exacq, or even open-source Frigate. Cameras record to that box continuously. It holds 7 to 30 days of footage depending on camera count, resolution, and disk space. That's your insurance policy.
Simultaneously, a subset of that footage uploads to the cloud. Not every frame. Not every camera. You pick the critical feeds (entry points, high-value areas, parking) and push 24 hours of cloud retention. Or just the last 72 hours. Depends on your retention policy and upload bandwidth.
When the internet is solid, you get cloud convenience. Live view from anywhere, multi-site dashboard, whatever platform you chose. When the uplink hiccups, the local NVR keeps recording. No gap. When connectivity comes back, the cloud system syncs what it missed, or you pull it manually from the NVR.
For long-term retention (30+ days), you live on the local box. Cloud is your redundancy and remote access layer, not your primary archive.
The Math That Decides Your Architecture
Before you pick cloud-only or hybrid, answer these questions with actual numbers.
Uplink bandwidth available. Run a speed test. Not the download speed. The upload. If it's under 15 Mbps and you have more than four cameras, you're already constrained. If it's under 5 Mbps, cloud-only is a gamble.
Retention requirement. How many days of footage do you actually need? Not what the customer says they want. What do they actually review? Most jobs need 7 to 14 days. A few need 30. Anything beyond that is expensive in cloud and cheap on disk.
Uptime tolerance. If the internet goes down for an hour, is that a problem? For most retail and warehouses, yes. For a small office, maybe not. That answer changes the architecture.
Multi-site consolidation. If you're monitoring five locations and need a single dashboard, cloud makes sense. If each location is independent, local NVR is simpler and cheaper.
Cybersecurity posture. Cloud systems are convenient targets. Local NVR on a segmented VLAN with no internet access is harder to breach. Hybrid lets you choose which feeds are cloud-exposed.
Once you've answered those, the architecture becomes obvious. Most jobs land on hybrid because they're neither pure-cloud-friendly nor pure-local-simple.
Common Hybrid Configurations
Small Warehouse or Retail Site (8-16 cameras)
Local 8-bay NVR. 2-4 TB drives. Holds 14 days at 1080p, 30 days at 720p. Four entry cameras stream to Verkada or similar. Local retention is your truth. Cloud is your remote access.
Uplink needed: 5-8 Mbps.
Multi-Location Chain (3-5 sites, 8 cameras each)
Each site has a local 4-bay NVR running Milestone. Central cloud account (same Milestone or Verkada) pulls key feeds from each location. Facilities monitors local box on-site. IT monitors cloud dashboard from head office. Retention: 30 days local, 7 days cloud.
Uplink per site: 3-5 Mbps.
Campus or Large Facility (40+ cameras, single building)
One or two large NVRs (16+ bays) in the server room. All cameras record locally. A proxy or edge appliance selects critical feeds and pushes to cloud. Local retention: 60+ days depending on storage budget. Cloud retention: 14 days for live view and investigation.
Uplink needed: 10-20 Mbps.
The Pitfalls That Still Catch People
Assuming cloud sync is automatic and complete. Most cloud platforms don't back up everything from the local NVR. You pick which cameras and what timeframe. If you don't configure it right, you'll have gaps.
Ignoring PoE budget for the NVR. A local NVR still needs to power cameras. Make sure the switch or PoE injector can handle your camera count. Eight 5-watt cameras at the edge of a 300-foot run is different from eight cameras 50 feet away.
Putting the NVR on the main office VLAN. Segment it. Put the NVR and cameras on a separate VLAN with restricted access. Cloud upload traffic is real traffic. You don't want it competing with business apps during peak hours.
Underestimating disk failure. A 4-bay NVR with four 4 TB drives holds 16 TB raw. One drive fails, you lose 25% of your retention. Use RAID 5 or 6. Accept the overhead. Your customer will thank you when a drive fails at 2 AM and you still have footage.
Forgetting to test failover. Before go-live, kill the internet at the site and watch the NVR. Does it keep recording? Does the cloud system gracefully pause? Does failover back work? Do it once during commissioning, not during an incident.
When Pure Cloud Still Makes Sense
If your uplink is genuinely bulletproof (redundant fiber, carrier SLA), retention is short (3-7 days), and you don't need multi-site aggregation, pure VSaaS is simpler. Fewer moving parts. No NVR to maintain. Verkada, Milestone Cloud, or Genetec Cloud all work fine in that scenario.
But most commercial jobs don't land there. They land in the middle, where hybrid is the practical answer.
If you're designing a system and haven't mapped your uplink, carrier history, and retention needs yet, do that before you touch a camera. The architecture follows from the constraints, not from vendor preference or last year's playbook.
Need help sizing the NVR or calculating bandwidth for your next job? We can work through the numbers with you. Reach out.