100 vs 1,000 vs 10,000 devices: how IoT cloud costs scale
A pilot on one production line is cheap to run. Teams then multiply the pilot’s bill by the number of lines to budget the rollout, and the result is wrong in both directions: some costs barely move as you grow, others jump in steps. This guide prices the same design at three sizes.
The three sizes
| Small factory | Mid-size plant | Multi-site | |
|---|---|---|---|
| Devices | 100 | 1,000 | 10,000 |
| Reporting interval | 10 seconds | 10 seconds | 10 seconds |
| Edge gateways | 1 | 10 | 50 |
| Fast-query history | 7 days | 30 days | 90 days |
| Archive | 1 year | 1 year | 2 years |
| App servers | 1 small | 2 small | 4 medium |
Each device sends a reading of about 1 KB, and every size has real-time processing switched on.
Monthly totals
| Cloud | 100 devices | 1,000 devices | 10,000 devices |
|---|---|---|---|
| Amazon Web Services | ₹52,046lowest₹520.46 per device | ₹97,428lowest₹97.43 per device | ₹5,16,588lowest₹51.66 per device |
| Microsoft Azure | ₹88,570₹885.70 per device | ₹1,53,938₹153.94 per device | ₹8,73,983₹87.40 per device |
| Oracle Cloud | ₹69,517incomplete₹695.17 per device | ₹76,273incomplete₹76.27 per device | ₹1,45,651incomplete₹14.57 per device |
| Google Cloud | ₹73,342incomplete₹733.42 per device | ₹87,346incomplete₹87.35 per device | ₹2,25,897incomplete₹22.59 per device |
Oracle Cloud's total leaves out Edge runtime (not offered), Time-series database (not offered) and IoT device gateway (can't be sized from published information). An incomplete total is never marked lowest.
Google Cloud's total leaves out IoT device gateway (not offered) and Edge runtime (not offered). An incomplete total is never marked lowest.
Live estimate: India (Mumbai / Pune), list prices refreshed 27 Oct 2026, recalculated on every refresh. Open in the estimator to change any input: 100 devices, 1,000 devices, 10,000 devices.
Read across a row: the cost per device falls sharply as the fleet grows. That’s the effect of fixed costs being shared across more devices.
Costs that barely move
Several building blocks cost about the same for 100 devices as for 1,000:
- Stream processing. Each service has a minimum job size, and one small job handles the data from thousands of devices.
- The time-series database, up to a point. Each has a minimum: two engine nodes on Azure Data Explorer, one Bigtable node, one Timestream for InfluxDB instance. (Oracle has no managed time-series database to compare.) A small plant pays for capacity it doesn’t use.
- App servers. Dashboards for 100 devices need roughly the same servers as dashboards for 1,000.
This is why a pilot’s bill looks expensive per device, and why it doesn’t predict the rollout.
Costs that grow with every message
- The IoT gateway. AWS IoT Core charges per message and per connected minute, so it grows almost exactly with device count. Azure IoT Hub grows in steps: the estimator picks the cheapest number of S1, S2 or S3 units that covers the daily message load.
- The event stream. Data volume on AWS, Google and Oracle, throughput units on Azure.
- The archive. It grows with data volume and with how long you keep it.
At 10,000 devices, the gateway is one of the two largest lines on both AWS and Azure. That is where batching and a longer reporting interval pay off most.
Costs that jump
Some services have step changes at scale:
- Time-series database nodes. Past a certain data volume, a database needs more nodes, and each node is a large step. Azure Data Explorer’s node count in particular rises with the amount of data held for fast queries, so 90 days of history at 10,000 devices costs far more than 30 days at 1,000.
- IoT Hub units. Moving from S2 to S3 units, or adding units, changes the line in large steps rather than smoothly.
Gaps on Google Cloud and Oracle
Google Cloud has no managed IoT gateway or edge runtime, so its totals leave them out and are marked incomplete at every size. To compare fairly, add the cost of what you’d use instead: a partner IoT platform, or your own MQTT brokers on VMs. See after Google Cloud IoT Core.
Oracle has no managed time-series database or edge runtime. It does offer an IoT Platform, but bills it by the hour of compute without publishing how many messages that covers, so it can’t be sized from a device count. Oracle’s totals leave these out and are marked incomplete at every size. A load test on the IoT Platform is the way to find its real cost for your fleet.
Budgeting a rollout
- Price the target fleet size, not the pilot multiplied up.
- Check which lines are fixed minimums at your pilot size: they won’t grow, so they are a poor guide to cost per device.
- Look for step changes between your current and target size, especially in the time-series database.
- Decide fast-query history per use case. It drives the database size more than device count does.
For one size in full detail, see what 1,000 sensors cost per month.