AI & Automation

Edge Computing vs Cloud for IoT: Where Should the Processing Happen

November 11, 2026 · 5 min read · Etzal Group Team

The first version of many IoT systems sends every reading straight to a cloud platform. It works well in a pilot. Then a site loses its connection for an afternoon and the alarms stay silent, or a vibration sensor producing thousands of samples per second multiplies the data bill. The question of edge computing vs cloud is the question of which work should stay near the machines.

Two places to run logic

At the edge, on a gateway or industrial computer at the site. It is close to the equipment, responds in milliseconds and keeps working without internet. Its computing power and storage are limited, and it is one of many units that must be maintained in the field.

In the cloud, in a central data centre. It has practically unlimited storage and computing, is easy to update and can see every site at once. It depends on the network, and each message takes time to travel.

Neither is better in general. Each function of the system belongs in one or the other.

Five questions for each function

1. How fast must the response be?

If an action has to happen within a second, such as stopping a pump when a tank is full, the decision cannot wait for a round trip to the cloud. Control and safety logic belong at the edge, and for safety functions, in dedicated controllers designed for that purpose.

If the response is a notification to a manager or a daily report, the cloud is fine.

2. What happens when the network is down?

Assume it will be, at some point, at every site. List what must continue:

  • Local alarms and interlocks.
  • Data recording.
  • Operator screens on site.

Anything on that list has to run locally. The edge device should buffer readings and forward them, with their original timestamps, when the link returns.

3. How much data is produced?

A temperature reading every minute is trivial. Vibration waveforms, high-frequency electrical measurements and video are not. Sending them raw is costly and often pointless.

Process at the edge and send the result:

  • Averages, minimums and maximums per interval.
  • Features computed from a waveform, such as overall vibration level and dominant frequencies.
  • Events only: "threshold exceeded at 14:02".
  • A short raw sample around an event, for diagnosis.

This can cut transmitted data by orders of magnitude while keeping what matters.

4. Is the data sensitive?

Camera images, production recipes and some operational details may be confidential, or regulated. Processing locally and sending only the outcome, such as a count or a pass or fail result, reduces exposure and may be required by policy.

5. Who will maintain it?

Every edge device is a small computer that needs updates, monitoring and eventual replacement. Logic placed there is harder to change than logic in the cloud. Put at the edge only what has to be there, and make sure it can be managed remotely.

A typical division

At the edge:

  • Reading sensors and controllers over industrial protocols.
  • Validation and filtering of readings.
  • Time-critical rules and local alarms.
  • Buffering during outages.
  • Reduction of high-frequency data.
  • Local operator display where needed.

In the cloud:

  • Long-term storage and history.
  • Dashboards and reports for many users.
  • Comparison across sites and assets.
  • Analytics and model training.
  • Notifications and integration with business systems.
  • Fleet management of the edge devices.

Models trained centrally can be deployed to run at the edge, which combines the cloud's capacity for learning with the edge's speed.

Designing the edge well

  • Store and forward. Size local storage for the longest outage you expect, and define what is dropped first if it fills.
  • Synchronise time. Readings with wrong timestamps are worse than missing ones.
  • Remote management. Configuration, software updates and logs must be reachable centrally, with a safe way to roll back a bad update.
  • Security. Unique credentials per device, encrypted communication, disabled unused ports and physical protection of the cabinet. Keep the control network separated from the internet-facing side.
  • Health reporting. Each device reports that it is alive, its storage level and its connection quality, so silence is noticed.
  • Industrial hardware. Rated for the temperature, dust and power conditions of the site.

Cost in both directions

Edge hardware costs money up front and to maintain. Cloud costs grow with data volume and never stop. A rough comparison for each site: the gateway and its upkeep over five years, against five years of transmitting and storing the raw data it would otherwise send. For low-volume sites, simple gateways that forward everything are often cheapest. For data-heavy sites, local processing pays for itself quickly.

Common mistakes

  • Putting control logic in the cloud and discovering the latency during commissioning.
  • No buffering, so every outage leaves a gap in history.
  • Edge devices that can only be updated on site.
  • Sending raw high-frequency data nobody analyses.
  • Treating the gateway as a consumer router and leaving it unpatched for years.

Starting point

For a first project, begin simple: a gateway that reads, buffers and forwards, with local alarms for the few conditions that cannot wait. Add more local processing only where a measured need appears, such as data cost or response time.

Gateway selection, local logic, buffering and remote fleet management are part of our IoT and Control projects. If your sites lose data during outages or your data costs are climbing, describe the setup to us.

Summary

Keep at the edge what must be fast, must survive an outage, produces too much data to send or is too sensitive to leave the site. Keep in the cloud what benefits from scale and a view across sites. Plan from the start how the devices in the field will be updated, secured and watched.

Frequently asked questions

What is an edge device in an IoT system?

It is a computer located near the sensors and machines, such as a gateway in an electrical panel or an industrial PC on the factory floor. It collects readings, can process and store them locally, and forwards what is needed to central systems.

Does edge computing remove the need for the cloud?

Rarely. The edge handles what must be fast or must work offline. The cloud remains the practical place for long-term storage, comparison across sites, dashboards for remote users and heavy analysis. Most systems use both.

Is edge hardware hard to maintain?

It needs a plan. Devices spread across sites must be monitored, updated and secured remotely, because sending a technician for each update does not scale. Choose hardware and software that support remote management from the beginning.

Author

Etzal Group Team

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

WhatsApp AI Chatbot

We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.