Bandwidth and storage sizing is the infrastructure calculation that every AI video analytics deployment requires before procurement — and the one most often done informally or not at all. Getting it wrong in either direction is costly: an undersized network drops frames and creates detection gaps; undersized storage overwrites footage before incidents are investigated. This guide explains the calculation framework and includes an interactive calculator for quick project estimates.
Two failure modes dominate poorly-scoped deployments, and both are avoidable with a pre-deployment calculation:
Undersized network: When total camera bandwidth exceeds the available network capacity, streams drop frames or disconnect intermittently. The AI analytics platform receives incomplete video, creating detection gaps — events that are not detected because the frames were never delivered. Operators may not notice these gaps because the system does not alert on what it cannot see.
Undersized storage: When NVR or NAS storage fills before the intended retention period, the oldest footage is overwritten. Incidents that are reported days or weeks after they occurred — common for theft, harassment, and compliance violations — may have no footage remaining. The 30-day retention policy on paper becomes a 12-day reality in practice.
Both failures are surprisingly common because the calculation is perceived as complex. It is not — the formula is straightforward, and the variables are knowable before deployment.
The core formula: Bitrate per camera (Mbps) × Number of cameras = Total bandwidth required.
The key variable is bitrate, which is determined by three camera settings: resolution, frame rate, and compression codec. Typical bitrate ranges for common configurations:
Add 20% headroom to the total for network overhead, burst traffic, and future growth. These are average figures — motion-heavy outdoor scenes compress less efficiently and will trend toward the higher end of each range. Static indoor scenes trend toward the lower end.
For the analytics platform specifically, verify whether it requires the primary stream (full resolution) or can operate on a secondary sub-stream (lower resolution). Some platforms can run inference on a sub-stream while the NVR records the primary stream, effectively halving the bandwidth requirement for the analytics path.
The core formula: Daily storage (GB) = Bitrate (Mbps) × 0.0450 × Recording hours per day × Number of cameras. Total storage = Daily storage × Retention period in days.
The constant 0.0450 converts megabits per second to gigabytes per hour (1 Mbps × 3600 seconds ÷ 8 bits per byte ÷ 1024 MB per GB ≈ 0.439 GB/hour, simplified for estimation as 0.45 GB/hour — the calculator below uses the precise conversion).
This is the single biggest storage reduction that AI analytics enables, and it is often overlooked in sizing calculations. A camera running continuous 24/7 recording generates 100% of its calculated daily storage. A camera running AI-triggered clip recording — only saving footage when a detection event occurs — typically generates 5–20% of its continuous equivalent, depending on site activity level.
A quiet perimeter camera that detects activity for 2 hours out of 24 saves 90% of its storage allocation. Across a 100-camera deployment, AI-triggered recording can reduce total storage requirements from 50TB to under 10TB for a 30-day retention period. Factor this into the calculation by estimating the percentage of cameras that will use AI-triggered recording and the expected activity percentage for those cameras.
Use the calculator below to estimate bandwidth and storage requirements for your deployment. Results are estimates based on typical bitrate ranges — verify against actual camera specifications before finalising procurement.
After calculating, relate the bandwidth figure to your switch and uplink capacity (is your network backbone sufficient?), and the storage figure to NVR or NAS specifications (do your recording devices have enough capacity for the retention period?).
Busy outdoor scenes compress less efficiently than static indoor scenes because more pixels change between frames. Use the higher end of bitrate ranges for outdoor cameras covering roads, car parks, or public areas. Use the lower end for indoor cameras covering corridors, server rooms, or storage areas with minimal movement.
AI event metadata adds less than 1% to total storage requirements — it is negligible compared to video but worth noting for completeness. A day of metadata for 100 cameras is typically measured in megabytes, not gigabytes.
Specify storage at 1.5× the calculated requirement to accommodate additional cameras, longer retention needs as compliance requirements evolve, and the natural tendency for camera counts to grow after the initial deployment proves successful.
For on-premises NVR deployments, only LAN bandwidth matters — cameras and NVR are on the same local network. For cloud-connected systems, WAN upload capacity is the constraint — and it is typically 10–100× smaller than LAN capacity. A deployment that is comfortable on a gigabit LAN may be entirely impractical over a 50 Mbps WAN uplink.
SafetyScope's AI-triggered recording model stores full-quality clips only when a detection event occurs, reducing storage requirements by 70–90% compared to continuous recording in typical deployments. The platform supports both primary and sub-stream processing — running inference on a lower-bandwidth sub-stream while the NVR records the full-resolution primary stream. Storage management tools provide real-time visibility into capacity usage, projected fill dates, and automated retention policy enforcement.
Published: 2026-03-23 · Updated: 2026-04-02