Connected Vehicles: Where the Data Costs Actually Come From

Connecting vehicles to the internet sounds like a solved problem. Put a SIM card in a box, bolt the box to the vehicle, done.

Then the bill arrives and it is several times what anyone expected. I want to explain where that money actually goes, because the biggest factor is a decision most people do not realise they are making.

The four things you pay for

The SIM subscription. A monthly fee per SIM, usually somewhere between 30 cents and 2 dollars. This is fixed — you pay it whether the vehicle sends a lot of data or almost nothing. When your data use is low, this is often your biggest cost.

The data itself. The important question here is whether your allowance is pooled across the whole fleet or allocated per vehicle. Pooled is what makes this affordable. Vehicles have very uneven usage — some run long routes daily, others sit at a depot for a week. With pooling, the heavy users borrow from the light ones. Without it, you pay penalties on the busy vehicles while leaving allowance unused on the quiet ones.

The management fee. A monthly fee per device for the software that manages it — updates, configuration, health checks. This one surprises people, because it is frequently bigger than the SIM and the data put together, and it is almost always missing from the first budget anyone writes.

Roaming. The one that destroys international budgets. And there is a legal wrinkle here that is not obvious: a device staying permanently connected to a foreign network, rather than just passing through, is restricted or banned in several countries including Brazil, Turkey, India and China. So a single worldwide SIM plan is often not legally possible. If you cross borders, work the cost out country by country.

Realistically: basic tracking runs about 1 to 5 dollars per vehicle per month. Detailed engine data runs 5 to 15. Anything with cameras runs 15 to 60 or more. International operation can roughly double any of those.

The decision that matters more than all of it

Here is the part almost nobody puts in the budget.

How much data your vehicles send is decided by where the AI runs.

Say you want driver monitoring — a camera that notices when someone is distracted and warns them. Two ways to build it.

The first way sends the video somewhere else to be examined. The vehicle uploads footage over the mobile network, software elsewhere analyses it, a verdict comes back. Every vehicle is now shipping gigabytes a month.

That is expensive, but worse than that: by the time the footage has travelled and been analysed and a warning has come back, the moment has passed. It cannot warn anyone in real time. And it stops working wherever the signal is poor — which is exactly the rural roads where tired driving is most dangerous.

The second way examines the video inside the vehicle. A small computer in the cab does the analysis. Nothing leaves except the conclusion: "distraction detected, 92% confident, 14:32, here is one still frame."

A few kilobytes. Not gigabytes.

The warning is instant because nothing had to travel. And it keeps working in a tunnel, because the detection never depended on the network.

Same result for your business. Roughly a thousandth of the data.

That is not a small saving you negotiate for. That is an architecture decision that makes the difference between a manageable bill and an unmanageable one — and it should be made before you choose a supplier, because the two approaches genuinely need different suppliers.

The catch with the better option

Putting the AI inside the vehicle is usually right, but it has a consequence worth understanding.

You now own software living inside every vehicle, and you will be updating it for years. Models get retrained, edge cases get found, rules change.

So sending software updates out to vehicles stops being an occasional task and becomes a continuous one. To moving objects. That are sometimes out of signal. And that cannot be easily undone if a vehicle is halfway through a route.

Ask any supplier for their measured success rate on remote updates — on real vehicles, on routes like yours. You will often hear something like 94%, which sounds fine.

On ten thousand vehicles it means six hundred vehicles need somebody to physically go and visit them. That is not a software statistic. That is a staffing plan.

One more thing that eats budgets

Vehicle data is not standardised. Different manufacturers report the same thing under different names, the same manufacturer changes between model years, heavy vehicles mix several standards, and electric vehicles add battery data with almost no consistency between makes.

Some suppliers clean this up for you, with wildly varying quality. Do not ask whether they support your vehicles — everyone says yes. Ask which specific pieces of data they translate, for each make and model year you run, and what happens when you add a make they have not seen before.

Ask for the list in writing. A supplier who cannot produce one does not have one.

What I would do

Decide where the analysis happens first, for what you plan to do in the next eighteen months — not just today.

Work out the monthly data per vehicle, including the busiest month rather than the average, because penalties get charged on peaks.

Compare suppliers on total cost per vehicle per month at three volume levels. You will find the rankings reverse completely between low and high usage.

Test it on twenty to fifty real vehicles for six to eight weeks, deliberately on your worst routes — tunnels, underground depots, rural stretches, borders. Good conditions tell you nothing.

Then before signing, make sure you can leave: can the SIMs move to another provider and at what cost, can you export all your data any time in a usable format without a volume-based fee, and is the contract shorter than the life of the hardware?

Full version here: How to Choose the Top IoT Connectivity Platform for Vehicles. What we build: IoT app development.

Frequently Asked Questions

What does it cost to connect one vehicle per month?

Basic tracking and diagnostics runs roughly 1 to 5 dollars per vehicle-month. Detailed engine and sensor data runs 5 to 15 dollars. Camera-based systems commonly run 15 to 60 dollars or more. Operating across borders can roughly double any of those figures.

Why is my actual bill higher than the quote?

Usually the per-device management fee, charged monthly for the software that manages each device. It is frequently larger than the SIM subscription and data combined, and it is the item most often left out of initial budgets.

Does running AI inside the vehicle really save that much?

Yes — typically around a thousandfold for camera-based features. The vehicle sends only the conclusion rather than the footage: a result, a confidence level, a timestamp and perhaps one still image. Kilobytes instead of gigabytes, for the same outcome.

Why can't I use one global SIM plan everywhere?

Because several countries, including Brazil, Turkey, India and China, restrict or prohibit a device staying permanently connected to a foreign network. You need locally provisioned profiles in those markets, so roaming must be modelled country by country.

Is a 94% remote update success rate acceptable?

Not on a large fleet. It means roughly six hundred vehicles out of ten thousand require somebody to physically visit them — a staffing and scheduling cost rather than a software one.

How do I check that a supplier handles a mixed fleet?

Ask for the written list of data parameters they translate for every make and model year you operate, and what happens when an unfamiliar make is added. Every supplier claims support; only some can produce the list.