Velosyti

Edge AI box or cloud: where should factory data live?

Both work. The right answer depends on your network, your latency needs and who else must never see the footage.

· 2 min read · For factory owners and plant managers adding cameras or sensors

You want cameras watching a production line for defects, or sensors reading machine health, and someone asks: should this run on a box sitting in the factory, or go up to the cloud? Both are real options today, and the choice is not about which is more advanced — it is about your network, your speed needs, and what you are and are not comfortable sending outside the building.

When an on-site box makes sense

  • A decision has to happen in under a second — stopping a line, rejecting a part — where a round trip to the cloud is too slow.
  • Your factory's internet connection is not reliable enough to stream video continuously.
  • Footage or readings are sensitive enough that you do not want them leaving the building at all.

When the cloud makes sense

  • You need to see data across several factory locations in one dashboard.
  • The analysis is not time-critical — a weekly trend, not a per-second alert.
  • You would rather not manage hardware on-site at all, and your connectivity is solid.

What it costs you

An on-site box is an upfront hardware cost and someone locally who can handle a restart if it goes down, but no ongoing data transfer cost and no dependence on your internet line for the time-critical part. A cloud approach avoids hardware but depends entirely on connectivity, and data transfer costs grow with camera count. Many factories end up doing both — fast decisions on-site, trends and dashboards in the cloud — rather than picking one for everything. A useful test is to ask what happens to the on-site function during a two-hour internet outage; if the answer is "it stops", that function probably belongs on-site, not in the cloud.

Questions to ask any vendor

  • What exactly runs on-site versus in the cloud, and why that split?
  • What happens to the time-critical function if the internet goes down?
  • Does footage or sensor data leave the building at all, and if so, where does it go?

How we can help

Kaappu is our models in a box that sits inside your factory, so vision and sensor data can be processed on-site with nothing leaving the building, alongside cloud dashboards where you want them. See it at /products/kaappu, and tell us about your line at talk to us.

References

  1. What is edge computing? — AWS
  2. OPC Unified Architecture — Wikipedia