Cloud computing sends data to large, shared data centres to be stored and processed. Edge computing does at least part of that work close to where the data is created — on a factory floor, in a shop, inside a vehicle or at a mobile mast. The difference is location, and location decides how fast a system can react, how much data has to travel and what keeps working when the connection drops.
How the cloud model works
In the cloud model, devices collect information and send it over the internet to a provider's data centre. There, servers run the applications, store the results and send answers back. The appeal is obvious: almost unlimited storage and computing power, paid for as it is used, with the hardware maintained by someone else.
The trade-off is distance. Every request makes a round trip, and every byte of raw data has to cross a network that costs money and occasionally fails.
How the edge model works
Edge computing places smaller computers — sometimes a rugged box on a wall, sometimes a module inside a machine — next to the data source. They filter, analyse and act on information locally, and send only what is useful to a central system. "The edge" is not one fixed place; it can mean the device itself, a server in the same building or a micro data centre at the edge of a telecom network.
Side-by-side comparison
| Factor | Cloud computing | Edge computing |
|---|---|---|
| Where processing happens | Central data centres | On or near the device |
| Response time | Depends on network round trip | Very short, local decisions |
| Bandwidth use | High if raw data is uploaded | Lower; only summaries or events are sent |
| Works offline | Usually not | Often yes, for local tasks |
| Scaling up | Very easy | Means installing and managing more hardware |
| Maintenance | Handled by the provider | Spread across many sites |
| Best at | Heavy analysis, storage, model training, collaboration | Real-time control, filtering, privacy-sensitive local data |
Why the edge exists at all
Latency
A camera checking parts on a conveyor has a fraction of a second to reject a faulty one. Waiting for a distant server is not an option, so the inspection runs locally.
Bandwidth
High-resolution video, vibration readings and sensor streams add up quickly. Analysing them on site and uploading only alerts or summaries keeps network costs under control.
Resilience
A ship, mine or wind farm cannot stop working because a link went down. Local processing keeps critical functions running, then syncs when the connection returns.
Data handling
Some organisations prefer that sensitive raw data — such as video of people — never leaves the building. Processing it locally and sending only anonymous results can simplify that.
Edge computing examples
- Manufacturing: machine vision for quality checks and vibration monitoring that flags a bearing before it fails, often feeding data to the line's PLC.
- Retail: shelf cameras and self-checkout systems that need instant results even on a slow store connection.
- Vehicles: driver-assistance systems must process sensor data on board; there is no time to ask a server whether to brake.
- Energy: substations and turbines that balance loads and protect equipment locally.
- Content delivery: servers near viewers cache video and web pages so they load quickly.
- Smart buildings: heating, lighting and access systems that keep working during an internet outage.
Where 5G fits in
Mobile networks are adding computing resources inside the operator's own infrastructure, a setup often called multi-access edge computing. Combined with the lower latency of standalone 5G, it lets mobile devices hand off demanding tasks without a long trip to a distant data centre. Our comparison of 5G vs 4G explains which network changes make that possible.
The hybrid reality
Very few systems choose one model exclusively. A typical pattern looks like this:
- Sensors and cameras produce raw data.
- Edge devices filter it, run quick checks and act immediately when needed.
- Summaries, exceptions and samples go to the cloud.
- The cloud stores history, runs heavier analysis and trains or updates models.
- Updated rules or models are pushed back out to the edge.
The same split appears in software that answers questions from company documents: indexing and search can run centrally while results are served close to users, a pattern touched on in our piece on retrieval-augmented generation.
Common mistakes when choosing
- Moving everything to the edge because it sounds modern, then struggling to patch and secure hundreds of small devices.
- Uploading every raw reading to the cloud and discovering that the network bill, not the computing, is the main cost.
- Ignoring physical conditions such as dust, heat and vibration, which ordinary servers are not built for.
- Forgetting updates: edge hardware needs a plan for remote management from day one.
Quick answers
Is edge computing replacing the cloud?
No. It handles the jobs that need speed or local autonomy, while the cloud remains the place for storage, large-scale analysis and coordination.
Is a smartphone an edge device?
Yes, when it processes data itself — for example, recognising a face to unlock the screen without sending the image anywhere.
Tell us what you think.
Corrections are always welcome.