NVR vs Cloud VMS: How Factory Video Actually Gets Stored (India)

For a mid-size Indian factory — ~20 mixed 2–4MP cameras, H.265, 30-day retention — a local NVR with 16–24 TB of surveillance HDD is the default. Cloud VMS adds off-site safety but needs sustained upload most Indian plants lack, so most run hybrid. The decision that actually moves the numbers is not cost — it is your retention window.

Key numbers

• ~20 mixed 2–4MP cameras, H.265, 30-day continuous = 16–24 TB of raw disk
• 1 Mbps of continuous video = 10.8 GB/day (fixed arithmetic)
• 20 cameras = 40–60 Mbps of sustained upload, 24/7 — not a burst
• Retention is the multiplier: 90 days ≈ 48–72 TB, 180 days ≈ 96–144 TB
• Local all-in (20 cams): ~₹80,000–₹1,30,000 one-time; cloud ~₹150–₹600/camera/month, forever

The storage decision is where a camera project turns into a cost project. Buy too little disk and footage rolls over before an incident is reviewed. Go all-cloud without checking your upload link and the recorder drops frames it never told you about. Below is the real TB and INR math, plus the two things the calculators leave out: the retention cliff, and who tells you when recording breaks.

NVR vs cloud VMS: what's the difference?

Footage lives in one of two places — a box in your plant, or a data centre over the internet.

• Local NVR / DVR. A recorder sits in your control room with hard drives inside. Cameras stream over the LAN; footage stays on-premises. This is what the overwhelming majority of Indian factories already run.
• Cloud VMS. Cameras, or a bridge appliance, upload footage over your internet link to a cloud Video Management System. You watch from a browser; retention is a subscription, not a disk.

They fail in opposite ways and bill in opposite ways. Local is capital that depreciates. Cloud is an operating cost that recurs forever, and it lives or dies on your upload pipe.

Storage sizing: how many TB for 20 cameras, 30 days?

Storage is just bitrate × time. The arithmetic is fixed: 1 Mbps of continuous video = ~10.8 GB per day (1 Mbit/s × 86,400 s ÷ 8 = 10.8 GB), the same constant CCTV calculators use (bandwidth/storage calculator). H.265 (HEVC) gives roughly the same picture as H.264 at about half the bitrate.

Typical bitrates for continuous 24/7 recording, and what they mean for 20 cameras over 30 days:

Bitrates assume ~15 fps and a moderately busy scene. Full frame rate (25–30 fps) can run 1.5–2× these numbers, so a full-rate 4K fleet lands at the top of the range. Add ~10–20% headroom.

A realistic mid-size floor — ~20 mixed 2–4MP cameras, H.265, 30-day continuous — needs roughly 16–24 TB of raw disk. That is the number every calculator gives you. It is also only true for one assumption: 30 days.

The retention cliff

Change nothing but the retention window and the disk requirement moves linearly:

For a regulated or export plant, the retention requirement — not the camera count — is what actually decides NVR-vs-cloud. There is no single legal number for general factory CCTV, so 30 days is the common default; but sector rules override it. Banking and ATM security norms commonly cite 180 days, and pharma or food customers auditing your line often want 90. A 180-day all-local build triples-to-sextuples your capex and drive-bay count — which is exactly where cloud's elastic tiered retention earns its keep: keep 30 days hot on the local NVR, push the long tail to cheaper cold cloud storage.

The box gotcha the TB math hides

Two limits a pure-TB calculation ignores:

1. Incoming bandwidth ceiling. An NVR caps how much stream it can ingest. A 16-channel unit rated ~160 Mbps handles 16 × ~3 Mbps fine; a 32-channel 4K fleet at 8–10 Mbps each can exceed a 256–320 Mbps ceiling and drop streams. Sum your camera bitrates against the recorder's rated incoming Mbps.
2. Drive bays. Most small/mid NVRs ship 2, 4 or 8 bays. Divide your target TB by (bays × max drive size). A 180-day/4K target can blow past 8 bays, forcing a NAS or second recorder — a cost the estimate never showed you.

Smart codec: don't trust the brochure on a factory floor

H.265+/scene-adaptive encoding is quoted everywhere as a 30–40% saving. It saves bytes only on static pixels. A factory floor is the worst case for it: conveyors, forklifts, robot arms, flashing stack-lights and rotating machinery keep most of the frame changing, and welding arcs spike the bitrate outright. On a real plant floor the saving largely collapses — budget close to the raw H.265 bitrate, not the smart-codec number a generic calculator hands you.

The India cost math

All INR figures below are mid-2026 planning bands. Get live vendor quotes before you sign anything.

Local NVR — one-time capital

• NVR body (16–32 channel). Basic 32-channel units start near ₹10,000–₹12,000; a 4K / H.265+ box with adequate incoming bandwidth runs ~₹15,000–₹35,000 (CP Plus NVR range).
• Surveillance HDD. An 8TB Seagate SkyHawk or WD Purple runs ~₹18,000–₹27,000 (Seagate SkyHawk). Use surveillance-rated drives built for 24/7 write — never desktop HDDs.
• Usable-capacity reality. Most small/mid NVRs run single-disk or JBOD (no parity), so three 8TB drives ≈ ~20 TB usable (~₹55,000–₹80,000 in disk). Want RAID5 redundancy? Parity costs one whole drive, so 3×8TB RAID5 = only ~16 TB usable — budget a fourth. And OS-reported capacity runs ~7% below the label (TB vs TiB), so an "8TB" drive shows ~7.3 TB.
• All-in for a 20-camera plant. NVR + ~20 TB of drives + a PoE switch on a UPS lands in the ~₹80,000–₹1,30,000 one-time band (excluding cameras and cabling), then near-zero recurring cost until a drive fails (plan a 3–5 year replacement cycle). Note: many NVRs ship with built-in PoE ports — check the port count before you buy a separate switch, or you'll double-pay.

Cloud VMS — recurring operating cost

Indian cloud-CCTV pricing is mostly quote-based. Continuous 30-day retention for 20 factory cameras sits well above entry-level home plans (commonly ~₹99–₹249/camera/month). Plan on ~₹150–₹600 per camera per month for factory-grade continuous retention. For 20 cameras that's ₹3,000–₹12,000 a month — ₹36,000 to ₹1,44,000 a year, every year. Over 3–5 years, cloud usually costs more than a local NVR: you're renting the disk, the bandwidth, and the promise someone else keeps it running. Ask too about the bill nobody quotes — retrieval/egress: pulling 30 days of footage for an insurance claim can be throttled or metered, so "off-site safety" carries a cost and a latency of its own.

Can Indian broadband handle cloud CCTV upload?

All-cloud plans die on one number, and it's not the one people check. Not download — upload. Cameras push data up; your broadband is built to pull data down.

Twenty cameras at ~2–3 Mbps each is 40–60 Mbps of sustained upload, 24 hours a day — a constant firehose, not a burst. India's fixed-broadband upload averages sit in the tens of Mbps (Ookla's Speedtest Global Index publishes the national figure; test your own line with TRAI's crowdsourced MySpeed portal). But "average speed" is not "committed sustained upload." Most business plans are asymmetric and contended, so a link that shows 50 Mbps on a test rarely holds 50 Mbps uphill against everything else the plant is doing. The vendor demos cloud on the showroom's fibre, not on your contended lease line during a shift change. Overload the pipe and the VMS just throttles — frame rate, then resolution — to fit. Nobody gets an alert. You find out the day you go looking for footage that isn't there.

The mechanism that actually beats it: buffer-and-drain

The fix that consensus articles hand-wave as "buffering helps" has a concrete shape. Record everything locally at full rate on the NVR (or camera-edge storage), then drain the copy to cloud during off-peak — nights and weekends, when the plant LAN and uplink are idle. You never need 40–60 Mbps sustained; you need enough burst headroom to clear one day's footage over a 24-hour window.

In our own multi-camera pilot this is exactly how it runs. The 2MP H.265 cameras hold ~1.7 Mbps each (close to the 2MP row above), producing ~23 GB/day across three cameras. The site's uplink measured ~61 Mbps, and the whole day's footage drains to cloud storage in ~50 minutes — but only overnight, on a schedule, to spare the daytime line. That is hybrid working as designed: on-premises reliability, plus an off-site copy, without a sustained uplink the plant doesn't have.

The reframe: who tells you it broke?

Every ranking article frames local-vs-cloud as capex-vs-opex and declares local cheaper over five years. True, and beside the point. Local's real liability isn't a stolen box — it's the recorder that quietly stopped recording and nobody noticed. A dead surveillance drive, a full disk that isn't overwriting, a degraded RAID, a recording schedule set wrong. You discover the gap only when you go looking for an incident and it isn't there.

We hit this in our own pilot. A power cut killed the PoE switch — cameras have no battery, so they die the instant mains drops even when the recorder survives — and the local recording task stood dead until the next morning, because it had no watchdog to restart it. Nobody would have known until they went looking. We added a self-healing restart; a dumb NVR has none.

That is the honest case for hybrid, and it's about reliability, not cost: a cloud VMS raises a camera-offline / no-upload alert. A local-only NVR has no one watching it. When you compare options, don't ask which is cheaper — ask which one tells you the day it fails.

Local vs cloud vs hybrid: the trade-offs

What a plant head should actually do

1. Fix retention first, then size disk. Your retention window sets the disk, not your camera count. Decide it — 30/90/180 days — before you spec the NVR, and check for any compliance, customer-audit or insurance clause that forces longer.
2. Test upload at the plant, sustained. Not a one-off speed test. If you can't hold your camera bitrate uphill over a 24-hour window, all-cloud is a trap — go local or hybrid with off-peak drain.
3. If you go cloud, demand India data residency in writing. The whole point of the 2026 BIS/STQC mandate is getting untrusted surveillance data out of Indian infrastructure; storing footage on servers abroad partially re-opens that hole and adds a DPDP cross-border-transfer problem. Confirm the provider's storage region before you sign.
4. Treat footage as governed data. Worker video is personal data under the Digital Personal Data Protection Act, 2023, which sets expectations for purpose, retention limits and notice (DPDP Act, MeitY). A retention window and access control aren't optional.
5. Buy compliant cameras regardless of where you store. New CCTV sold or installed in India from 1 April 2026 must meet BIS/STQC Essential Requirements (BIS CRS guidelines). Storage choice doesn't change the hardware rule.
6. Keep timestamped footage for safety duties. The Factories Act, 1948 imposes continuous duties around dangerous machinery (India Code). It doesn't require CCTV — but where you do record, timestamped footage is useful evidence your controls were followed.

Sizing disk assumes you already know how many cameras go where — and that placement question is where projects most often go wrong. This is the gap Mama closes: record a short phone walkthrough of the floor, and it reads the space — zones, sightlines, the hazards that need coverage — then returns a floor plan plus a camera-placement plan. From that you get the exact channel count and resolution mix, so your storage math starts from real numbers instead of a guess.

FAQ

How much storage do I need for 20 cameras for 30 days? Rule of thumb: budget ~1 TB per camera-month at 4MP H.265, then round up. For ~20 mixed 2–4MP cameras recording continuously with 30-day retention that's roughly 16–24 TB of raw surveillance HDD. The window is the multiplier — 90 days roughly triples it to 48–72 TB.

Is cloud VMS cheaper than a local NVR? Usually not for a single site. Cloud overtakes a local NVR somewhere around year two or three: the NVR is a one-time ~₹80,000–₹1,30,000, while cloud recurs at ~₹150–₹600/camera/month — ₹36,000–₹1,44,000 a year for 20 cameras, forever. Cloud earns its keep on multi-site and off-site resilience, not price.

Can Indian broadband handle all-cloud CCTV? Often not. Twenty cameras is 40–60 Mbps of sustained upload 24/7, and most Indian business links are asymmetric and contended — a 50 Mbps speed test rarely means 50 Mbps of continuous upload. Test sustained upload at the site first; if it can't hold, run hybrid and drain the copy to cloud overnight.

Should I use H.265 or H.264? H.265 (HEVC) for storage — same picture at roughly half the bitrate of H.264. Confirm your NVR and cameras both support H.265/H.265+; mixing can force a fallback to H.264 and blow up your estimate. Don't over-count the H.265+ "smart" saving on a busy floor — moving machinery erodes it.

What retention period do Indian factories need? No single legal number exists for general factory CCTV, so 30 days is the common default. But sector rules override it: pharma/food audits often want 90 days, and banking/ATM security norms commonly cite 180. Balance it against the DPDP Act's expectation not to keep personal data longer than needed. Retention, not camera count, is what usually decides NVR-vs-cloud.