Introduction: Why Nobody Publishes Honest Pricing

Ask five vendors what AI video analytics costs and you will get five numbers that are not comparable. One quotes per camera. One quotes per stream. One quotes per use case. One bundles hardware into the licence. One gives you a number for the pilot and goes quiet about year two.

This guide breaks the cost into its actual parts, with real published figures, so you can read any quote you receive and immediately see what is missing from it.

One framing rule before the numbers: cost does not scale with the cameras you own. It scales with the streams you process and the use cases you enable. A plant with 60 cameras that wants PPE detection on 8 of them is a smaller job than a plant with 12 cameras that wants PPE, forklift zones and SOP verification on all of them. Any vendor pricing purely per installed camera is either overcharging you or planning to.

The Five Things You Are Actually Buying

Keep these five line items separate on every quote you receive. Vendors blur them because blurring them hides the expensive parts.

Analytics software licence — recurring, typically 20 to 40 percent of year-one cost.

Compute, meaning the edge box or GPU server — capital expense, 10 to 30 percent.

Cameras that must be replaced, repositioned or added — capital expense, anywhere from 0 to 35 percent.

Network, cabling and installation — capital expense, 10 to 25 percent.

Integration, tuning, model retraining and AMC — both one-time and recurring, 10 to 25 percent.

The standard budgeting mistake is treating capital and operating expense as a single number. Hardware plus installation is a one-time hit, while storage growth, analytics licences and annual maintenance recur forever. Ignoring the running cost is the most common error in Indian CCTV budgeting.

Line Item 1: The Analytics Licence

This is the only line most vendors quote, and it is the smallest one.

The global benchmark. Analytics software runs roughly 3 to 15 US dollars per camera per month for the licence alone, layered on top of cameras you already own. At that rate, a 50-camera site lands somewhere around 1,800 to 9,000 dollars a year.

The India benchmark. Published Indian per-camera SaaS pricing sits at 800 to 1,200 rupees per camera per month. For a 10-camera site that is 8,000 to 12,000 rupees monthly. One Indian vendor publishes a flat rate of 1,000 rupees per camera per month with no hardware or licence fee attached. Across 25 cameras that is 2.4 to 3.6 lakh a year at the lower end.

But watch the unit carefully, because the pricing models in this market are not interchangeable.

Per camera per month. Typically 800 to 1,200 rupees. Predictable and easy to forecast, but it penalises you for owning cameras you never intend to analyse.

Per analytics unit. One Indian cloud provider prices basic analytics such as intrusion, zone monitoring, line crossing, loitering and camera sabotage at 250 rupees per unit per month, kept separate from its cloud VMS plan at 1,800 rupees per camera per month. This looks cheap until you count how many units a real deployment needs.

Per instance with an alert cap. Another publishes 2,399 rupees per instance per month at the entry tier and 7,999 rupees at the production tier, with monthly alert capacity ranging from 750 to 7,500 before you are moved onto a custom enterprise contract. Read that cap closely. A badly tuned PPE model on a busy line will burn through 7,500 alerts in a fortnight.

The full India range, end to end. Roughly 2 lakh rupees for a genuine pilot, rising to 5 crore a year for large multi-site enterprise deployments.

Line Item 2: Compute, And The Question Nobody Asks

Where does inference actually run? This single decision drives both your capital expense and whether footage ever leaves your plant.

Cloud. No capital expense, but you pay for egress and storage forever, and every frame leaves the site. For a plant with DPDP exposure and a patchy leased line, this is often the wrong answer regardless of what it costs.

Edge. A box in the panel, processing streams locally. The reference device is the NVIDIA Jetson class, and this is exactly where Indian buyers get burned, because the same part carries wildly different Indian prices.

The Jetson Orin Nano Super Developer Kit carries a 249 dollar global list price after NVIDIA's price cut, delivering 67 TOPS of performance. In India, one authorised distributor lists it at 21,000 to 23,000 rupees, the same distributor's own product page shows 49,999 rupees, and a third reseller lists 39,499 rupees. The bare production module, which is what actually goes inside deployed hardware, runs 28,900 rupees for the 4GB variant and 32,250 rupees for the 8GB. At the upper end, the AGX Orin 64GB is 1,999 dollars globally and has been listed in India at around 2.7 lakh rupees excluding tax.

What this means for your quote. A vendor quoting an edge appliance at 1,20,000 rupees per unit may be marking up a 30,000 rupee module four times over, or may be including a ruggedised enclosure, industrial power supply, warranty and annual maintenance that you genuinely need in a dusty plant. Ask which one it is. Both answers are legitimate. Only one is worth the money.

A rule of thumb for sizing. An Orin Nano class box handles a handful of 1080p streams at real-time frame rates for a single detection task. Running multiple simultaneous use cases on the same stream cuts that number sharply. If a vendor tells you one small box will run 16 cameras across four use cases, ask for a live demonstration at full load rather than a spec sheet.

Line Item 3: Cameras, And The "Works With Your Existing CCTV" Asterisk

Every vendor says analytics runs on the cameras you already own. It is broadly true. Most cameras support ONVIF and RTSP, which makes a software subscription on your current cameras the cheapest route to modern analytics with no new boxes to buy. Some platforms work exactly this way. Others depend on proprietary devices, local appliances or a wider video management upgrade, and a hardware-plus-licence system adds 600 to 3,500 dollars per camera before the software even starts.

The honest version is this: most of your cameras will work, and the ones that matter most often will not.

A camera mounted for theft deterrence points at a door from twelve feet up. A camera for helmet detection needs to see heads at usable pixel density, without backlight from an open shutter door, at an angle where a hard hat is not occluded by the worker's own shoulders. In a typical plant, expect 20 to 40 percent of the cameras covering your target zones to need repositioning, relensing or outright replacement.

Real Indian replacement costs. Dealer pricing puts a 2MP IR bullet at roughly 2,880 rupees, a 2MP hybrid with microphone at 3,516 rupees, and a 4MP AcuSense-class bullet at 7,198 rupees, all inclusive of tax. Generic 4MP IP bullets trade from 1,800 rupees upward on the open market. Budget 3,000 to 8,000 rupees per camera replaced, plus the mounting work on top.

Line Item 4: Cabling And Installation, The Line That Kills Budgets

This is where Indian quotes balloon, and it has nothing to do with AI at all.

A published Indian CCTV quote shows the shape clearly. Two hundred metres of Cat6 at 5,400 rupees. PVC pipe and fittings at 3,500 rupees. A single combined line covering installation, testing, recorder configuration, network configuration and mobile view setup at 6,500 rupees. With 18 percent GST applied, the system quote reaches roughly 86,000 rupees.

Scale that to a plant and the pattern holds. Moving one camera fifteen feet inside a live production area costs more than the camera itself. Conduit, scaffolding access, shutdown windows, and a PoE switch port that may not even exist yet.

If a vendor's quote contains no installation line at all, they are assuming you will handle it, and you will discover that in week three.

Line Item 5: The Recurring Costs Nobody Puts In The Pilot Quote

Storage. Incident frames and full footage run on different retention clocks. Cloud VMS plans commonly bill storage per gigabyte per month on top of the per-camera fee.

Model retraining. Your plant's lighting changes with the seasons, your PPE supplier changes, you add a line. The model degrades. The most common trap in Indian deployments is optimising for the year-one licence price while ignoring retraining, camera upgrades and integration. Always compare the 36-month total cost of ownership.

Integration. Pushing alerts into WhatsApp, your EHS system or a PLC is engineering work, not a checkbox on a feature list.

Annual maintenance. Edge hardware sitting in a factory panel fails. Budget for it.

The cost of false alarms. Basic motion detection carries false alarm rates above 95 percent, and every false alert consumes supervisor attention. A cheap system that cries wolf has a real, recurring cost on your payroll. It simply never appears on the invoice.

Cost By Use Case: The Four Complexity Tiers

This is the part that per-camera pricing hides completely. These four categories are not equally difficult, so they should not be equally priced.

Tier 1: Geometric events. Intrusion, line crossing, zone entry, loitering, camera sabotage. The system detects a person and checks whether the bounding box crosses a line you drew. No semantic understanding is required. This is precisely why one Indian provider can price the entire bundle at 250 rupees per unit per month. Cheapest tier, deployable in days.

Tier 2: PPE detection. Helmet, gloves, vest, shoes. This needs a trained detector and, critically, per-camera threshold tuning. Absence detection, meaning "no helmet", is considerably harder than presence detection. Expect two to six weeks of tuning against your own footage before the alerts become trustworthy. Mid-tier pricing, and the tier where accuracy claims should be tested against your floor rather than a public benchmark.

Tier 3: Interaction and behaviour. Forklift and pedestrian proximity, vehicle-pedestrian interaction, line-of-fire exposure, falls from height, slips. These require tracking objects across frames and reasoning about spatial relationships over time, not single-frame classification. Materially more compute per stream and materially more tuning. Expect a clear premium over PPE.

Tier 4: Process and SOP verification. Changeover timing, cleaning and hygiene verification, uniform compliance, defect detection on the conveyor. These are plant-specific by definition, because your SOP is not your neighbour's. They require labelled data from your own site and custom model work. Priced as a project rather than a licence, and legitimately so.

A vendor charging the same per-camera rate for Tier 1 and Tier 4 is either subsidising the hard work with the easy work, or has not actually done the hard work.

Three Worked Scenarios

These are illustrative ranges built from the published figures above, assuming existing cameras are used wherever possible. They are not quotes. They are the shape a realistic budget takes.

Scenario A: Pilot, four cameras, one zone, PPE only, three months. Analytics licence at 800 to 1,200 rupees per camera per month comes to 10,000 to 15,000 rupees for the pilot period. One edge box at 25,000 to 60,000 rupees, one time. Minimal camera work if you choose a zone with decent existing coverage. Tuning effort is the real cost here. All-in: 75,000 to 2,00,000 rupees, which is consistent with the 2 lakh figure published as the Indian pilot benchmark.

Scenario B: Single plant, 25 streams, three use cases, year one. Licence at Indian per-camera rates gives 2.4 to 3.6 lakh a year for the base tier, higher once Tier 3 use cases are added. Compute of four to six edge boxes or one GPU server, 1.5 to 4 lakh. Camera work assuming eight of 25 need replacement or repositioning at 3,000 to 8,000 rupees each, roughly 1 to 2 lakh once cabling is included. Integration and tuning, 1.5 to 3 lakh. Year-one all-in: roughly 7 to 14 lakh, settling to 3 to 5 lakh recurring thereafter.

Scenario C: Multi-site enterprise. Volume discounts apply to the licence, but integration and change management dominate the budget. The published upper bound for large Indian multi-site deployments is 5 crore a year.

Eight Questions To Put On Every Quote

1. Is your price per camera, per stream, per use case or per analytics unit?

2. Is there an alert cap, and what happens when we exceed it?

3. Where does inference run, and does any footage leave our site?

4. What is the edge hardware bill of materials, and what is your markup on it?

5. How many streams per box, running how many simultaneous use cases, at what frame rate?

6. How many of our existing cameras will you certify as usable, before we sign anything?

7. Is installation and cabling in your scope, or ours?

8. What does retraining cost in year two, and who pays when accuracy drifts?

Then ask for the 36-month number rather than the year-one number.

The Honest Summary

For a single Indian plant running PPE and zone detection across 20 to 30 streams with mostly usable existing cameras, a realistic first-year budget is 7 to 14 lakh rupees, settling to 3 to 5 lakh annually.

Below 5 lakh, something is being left out of the quote, and it is usually installation or tuning. Above 25 lakh for that same scope, you are paying for hardware markup or for use cases you never asked for.

The cheapest deployment is always the one scoped tightly to the zone where incidents actually happen. Four cameras that work will beat forty that nobody trusts.

Add GST, and confirm it appears on the invoice.