What edge computing actually means
Edge computing moves computing power to the edge of the network – to the place where data is created. Instead of first sending sensor data, camera images or machine signals over a wide area connection to a central data center, they are processed directly on site. Only what is genuinely needed centrally is then transmitted: aggregated metrics, anomalies, results.
The practical difference is substantial. Vision-based quality control generates large volumes of data continuously, the vast majority of which is unremarkable. Transmitting all of it costs bandwidth, time and money – whereas locally only one answer matters: pass or fail, right now.
The four drivers
1. Latency
In manufacturing control loops, autonomous vehicle control in plant traffic or safety shutdowns, milliseconds count. The round trip to a distant data center is limited by physics – there is no negotiating with the speed of light in a fibre. If a system has to react within a few milliseconds, processing has to happen on site.
2. Data volume
High-resolution cameras, vibration sensors and measurement equipment generate data volumes that are commercially unattractive to transmit in full. Local pre-processing frequently reduces the volume by orders of magnitude before anything leaves the plant at all.
3. Data sovereignty
Production data, formulations, design data and personal information are often subject to requirements that make processing outside your own premises difficult or impossible. An edge data center on your own site settles that question physically: the data stays where it is created.
4. Resilience
A site whose production depends on a WAN connection comes to a standstill when that connection fails. If the critical processing runs locally, by contrast, the plant keeps working – central synchronisation catches up as soon as connectivity returns. For sites with weak or unreliable connectivity this is often the strongest argument.
The most useful test questionWhat happens at this site if the internet connection is down for two hours? If the answer is “production stops”, that is a strong argument for local computing power – regardless of any latency or volume considerations.
Practical examples from industry and infrastructure
| Application | Main reason for edge | Typical scale |
|---|---|---|
| Optical quality control in manufacturing | latency and data volume | 10–50 kW |
| Predictive maintenance across machine fleets | data volume | 5–20 kW |
| Logistics centre with sorting and scanning systems | latency and resilience | 20–100 kW |
| Substation or energy plant | sovereignty and resilience | 3–15 kW |
| Mobile network or network node site | latency | 10–80 kW |
| Construction site or temporary location | lack of infrastructure | 3–30 kW |
The scale is striking: edge applications mostly sit in the single- to double-digit kilowatt range. That is exactly what compact modules are designed for, and they can be operated sensibly from around 3 kW of IT load – a dedicated data center building is neither necessary nor economical for this.
What an edge site needs technically
The appeal of edge sites lies in the fact that they are rarely located where IT infrastructure already exists. A substation, a hall on the edge of the plant or a logistics yard has no server room – and usually no IT staff either. An edge data center therefore has to bring four properties with it:
- Self-contained cooling. Cooling has to work independently of the building services – including at summer outdoor temperatures and in dusty environments.
- Its own power supply with UPS. Industrial power networks are not always stable. A UPS rides through dips, and a backup generator is added where needed.
- Physical security. Access control and fire protection are more important at unstaffed sites than in a supervised building, not less.
- Remote monitoring. With no staff on site, temperature, power supply, access and faults have to be reliably reported to a control centre.
These four points are precisely what a prefabricated modular system already solves at the factory – a key reason why edge roll-outs are almost always implemented with modules rather than in-house builds in practice. How that compares commercially with a conventional room fit-out is set out in the article Container data center or your own server room.
Planning an edge site?
Tell us which application is to run there and what infrastructure exists – we will propose an appropriately sized configuration, for rent or for purchase.
Edge and cloud: not opposites
Edge computing is occasionally presented as a countermovement to the cloud. In practice the two complement each other. The sensible division of labour usually looks like this:
- At the edge: capture, pre-processing, real-time decisions, short-term buffering, local control.
- Centrally in the cloud or data center: long-term archiving, cross-site analysis, model training, reporting.
A fault detection model running at the edge is therefore trained centrally and executed locally. The heavy computing work happens once, centrally; the time-critical application runs thousands of times on site.
Is edge worthwhile at your site?
These five questions give a reliable first assessment. Two or more clear yes answers usually argue for local computing power:
- Do systems at the site have to react in less than about 20 milliseconds?
- Are large volumes of data generated there continuously, of which only a fraction is needed centrally?
- Are there regulatory or contractual requirements that restrict processing outside the premises?
- Does operation come to a standstill if the WAN connection fails for a few hours?
- Is the bandwidth available at the site permanently scarce or expensive?
Starting small is expressly allowedAn edge concept does not have to cover every site from day one. A rented module at a pilot site delivers solid figures on benefit and operating effort within a few months – and the decision on a wider roll-out is then made on measured values rather than assumptions.
Frequently asked questions
What is edge computing, put simply?
Edge computing means processing data where it is created – on the production floor, at the logistics site, in the substation – instead of transmitting it to a central data center or the cloud first. That cuts latency, reduces the volume of data transferred and keeps operations running even when connectivity fails.
At what size does a dedicated edge data center become worthwhile?
There is no fixed lower limit – what matters is the reason. Compact modules start at around 3 kW of IT load, which is enough for most edge applications running a few racks. The deciding factor is rather whether latency, data volume, sovereignty or resilience make local processing necessary.
Does an edge site need a fibre connection?
Not necessarily. Fibre is ideal, but many edge sites are connected via 5G or LTE. Since the data-intensive processing happens locally anyway, usually only the result is transmitted over the link – so the bandwidth required is correspondingly small.
How does edge computing differ from the cloud?
The two are not opposites but complements. Time-critical and data-intensive processing happens at the edge, while analysis, long-term archiving and model training run centrally in the cloud or the data center. This division is often referred to as a hybrid architecture.
How is an edge data center maintained when there is no IT staff on site?
Through remote monitoring. Cooling, UPS, access and environmental readings are monitored continuously and reported to a control centre. Maintenance is carried out on site at regular intervals, faults are handled through 24/7 customer service – staff at the site are not required.
Last updated on