All writing

Engineering· · 9 min read

Why per-device IoT pricing punishes fleets

Per-device and per-message IoT pricing scales with your success, not with what your fleet costs to host. A worked example at 1,000 and 50,000 devices, and the alternative.

By Alessio Grespi, Founder & CEO

Most IoT platforms charge by the device, by the message, or both. That feels fair at ten devices. At fifty thousand it means your bill tracks how many units you have sold, not what it costs to keep them connected.

This post looks at how the common pricing models work, runs the numbers for one fleet at two sizes, and sets out the alternative: paying for the infrastructure a deployment needs.

A note on scope. This is about platforms for teams building and running their own devices. A finished monitoring service, where you buy a working outcome and pay per monitored point, is a different purchase and is rightly priced differently.

How IoT platforms bill today

Three models cover most of the market.

ModelWhat you pay forExample
Per deviceA fee for every device that connects in a monthGolioth charges $0.25 per connected device per month, plus data by the megabyte
Per messageEvery message, connection minute, state update and ruleAWS IoT Core charges $1 per million messages and $0.08 per million connection minutes in its Ireland region, with shadow updates and rules billed on top
Capacity unitsBlocks of daily message quotaAzure IoT Hub sells units that each allow a set number of messages per day, from 400,000 upwards

None of these is unreasonable. Each one picks a number that is easy to count. The trouble is that the number being counted is not the thing that costs money to run.

Of the three, usage metering of the kind AWS does is the closest to how costs behave, because the bill follows activity. Its weakness is the number of separate meters, and how fast they add up once a design gets chatty.

IoT usage is highly variable

No two fleets load a platform the same way, and one fleet rarely behaves the same from month to month.

  • Reporting rate. A tank level sensor may report once a day. A vibration monitor may report many times a second, or once a minute with a far larger payload each time.
  • Payload size. A temperature reading is a few bytes. A waveform or an image can be thousands of times larger, so message count alone says little about load.
  • Bursts. Event-driven devices stay silent for weeks, then all report at once when something happens.
  • Firmware updates. A rollout sends the same file to every device, and creates a traffic spike unrelated to normal use.
  • Sleep and coverage. Battery devices spend most of their life offline. Many are connected for seconds a day.
  • Change over time. One firmware release can double a fleet's traffic overnight, or halve it.

A single fixed fee per device cannot fit that range. It will be too high for the quiet fleet or too low for the busy one.

Where it goes wrong

  • Per-device fees ignore what a device does. A sensor that sends one reading an hour pays the same as one that streams all day. Quiet fleets subsidise the pricing model.
  • Per-message fees punish chatty designs. Engineers start batching, trimming and delaying telemetry to manage the bill, not because the product is better for it.
  • Costs scale with sales, not with load. Doubling the fleet doubles the bill, even when the extra traffic would fit on the server you already have.
  • Forecasting gets hard. A firmware change that adds one message a minute can move the bill more than a year of growth.
  • The extras add up. Device state, rules, firmware updates and storage are often metered separately, so the headline rate is only part of the cost.

A worked example

Take a fleet of sensors that stay connected and each send one small message every five minutes. That is 8,640 messages per device in a 30-day month.

PlatformHow it charges1,000 devices, per month50,000 devices, per month
AWS IoT CorePer message and connection minute$14.69$734.40
Golioth$0.25 per connected device$250$12,500
DatacakeFlat plan by device band€659Custom quote
ThingsBoard CloudFlat plan by device band$749Custom quote
UbidotsPlan plus $2 per extra device$1,999Custom quote
balenaPlan plus $2 per extra device$3,219$101,219
Particle$299 per block of 100 devices$3,588Custom quote

These are published list prices, worked through for this fleet. The AWS row counts connection time, messages, and one rule and one action per message. The Golioth row is the connection fee alone, before data charges. Particle needs twelve blocks at 1,000 devices, because the message volume exceeds what ten blocks include. Large fleets usually negotiate, so read the right-hand column as the direction of travel, not a quote.

These platforms do not all do the same job. Particle bundles cellular connectivity. balena manages Linux devices. ThingsBoard, Datacake and Ubidots include dashboards. The point of the table is how each price scales, not which product is better value.

Two things stand out.

First, every model that counts devices costs far more for a quiet fleet than the one that counts activity. Across the table, the same 1,000 sensors cost anywhere from under $15 to over $3,500 a month. Even a fee as low as $0.25 a device reaches $150,000 a year at 50,000 devices, before a single byte of data is billed.

Second, per-message pricing looks cheap until the design changes. If the same devices send one message a second, the per-message bill becomes about $3,373 a month at 1,000 devices and about $168,650 a month at 50,000.

Now look at the load. Fifty thousand devices sending every five minutes is roughly 167 messages a second. That is well within what a single modest message broker handles. The infrastructure needed barely changes between the two fleet sizes, while both bills grow fiftyfold.

Serverless or monolith: why the bills differ

The pricing follows the architecture underneath. Hyperscale IoT services are serverless: nothing is reserved for you, so every message and operation is metered. The other approach is a monolith: one stack of broker, database and application on a machine sized for your fleet.

Serverless, meteredMonolith, fixed instance
The bill followsEvery message and operationThe size of the machine
Quiet fleetVery cheapYou pay for idle headroom, but the smallest instance costs a few dollars a month
Busy fleetCost climbs with every messageFlat until the instance is full
ForecastingHard, it moves with firmware behaviourKnown in advance
Scaling upAutomaticStep up to a larger instance, or add machines that start on demand and stop when idle
Who runs itThe providerYou, or a platform that does it for you

Neither is wrong. Serverless suits workloads that are small, spiky or unknown. A monolith suits a fleet whose baseline load is steady and known, which is where most sensor products end up once they are in the field.

The alternative: pay for the infrastructure

Yes, you can do it yourself. Rent a server, run an open-source message broker and a database, and your fleet costs you the price of the machine.

Then the rest of the job arrives. Someone has to design the device security, the certificates and the data pipeline. Someone has to keep it all patched, backed up and online at three in the morning. Someone has to build firmware updates that will not brick a fleet. You set out to build a product, and suddenly you are four disciplines at once: cloud architect, software and reliability engineer, and firmware engineer.

A platform should give you the economics of running it yourself without the three extra jobs. That is the gap Paqett Console is built to fill - and we use it internally too.

What a deployment costs to run comes down to three things: compute, storage and data leaving the platform. Price those, and the bill follows the load.

  • Small or early fleets share infrastructure and pay for the data they move and store.
  • Larger fleets get a dedicated instance, sized to their throughput and storage, at a flat price.
  • Device count is not on the invoice. Adding ten thousand quiet sensors to an instance with headroom costs nothing extra.

The effect is that your unit economics improve as you grow. The cost per device falls with every device you add, which is how hardware margins are supposed to work.

This is how Paqett Console is priced: by the infrastructure a deployment needs, not by device or message count. That applies to Console only. Paqett's finished monitoring products are sold per monitored point, because there you are buying a running service and its results, not infrastructure.

When per-device pricing is fine

It is not always the wrong choice.

  • Finished products and managed services. When you buy an outcome, such as a monitored building, a price per monitored point ties the cost to the value you get. The supplier carries the platform cost, not you.
  • Prototypes and small fleets. At a few hundred devices the bill is small whichever way it is counted, and simplicity wins.
  • High-value devices. If each unit earns hundreds of pounds a month, a small per-device fee is noise.
  • Fleets you never expect to grow. The scaling problem only bites if you scale.

The model starts to hurt when you are the one running the platform layer and your devices are cheap, numerous and quiet. That describes most teams building their own sensor products.

Frequently asked questions

How much does an IoT platform cost per device? It depends on the model. Per-device platforms charge a fixed monthly fee for each connected device. Per-message platforms can cost a cent or two per device for quiet sensors and several dollars for chatty ones.

Is AWS IoT Core expensive? Not for quiet fleets. Its metered rates are low, and billing by usage is the right idea. Costs rise quickly with message rate, and with the shadow, rules and storage services most products end up using.

What is infrastructure-based IoT pricing? You pay for the compute, storage and data transfer your deployment uses, in the same way you would pay for a server. The number of devices is not billed.

When should I move off per-device pricing? When the platform fee is a visible share of what each device earns, or when engineers are changing firmware to manage the bill.

Building a device fleet?

Paqett Console gives device makers and integrators device management, updates, data and an API, priced by the infrastructure a deployment needs. Talk to us about your fleet.

Sources

Prices are list prices in US dollars, read on 15 November 2026.

Keep reading

Compliance· · 4 min read

Chain of custody for sensor data

Compliance data only counts if you can prove nobody changed it. What chain of custody means for sensor readings, where logbooks fail, and what to ask a vendor.

Water hygiene· · 5 min read

Manual vs automated Legionella monitoring

Monthly thermometer rounds or sensors? How manual and automated Legionella temperature monitoring compare on cost, coverage, records and risk, and when each makes sense.

Water hygiene· · 2 min read

Your rounds, plus the 29 days in between

A monthly temperature round is one reading per outlet. This is what the rest of the month looks like when a probe stays behind.

Engineering· · 5 min read

What chip for what kind of device

ESP32, Nordic nRF or RAK3172: which chip suits which kind of device, and what each one costs you.

Get new posts by email

One email when we publish. Nothing else.

Or talk to us about a deployment: water hygiene, white-label or built to spec.