Three claims travel with edge computing wherever it is sold into industry: it is faster, it is more secure, and it is inevitable. Two of those hold up under scrutiny. The middle one is the opposite of what the current breach data says, and getting it wrong is expensive in a way the other two are not.
Key takeaways
- IEEE 802.1 TSN targets 1 ms cycle time, 1 microsecond jitter and 99.999 percent reliability.
- 3GPP Release 16 added URLLC and the interworking that lets 5G carry TSN traffic.
- Edge devices and VPNs were 22 percent of exploitation targets in the 2025 Verizon DBIR.
- That share was 3 percent a year earlier, an increase of roughly eight times.
Claim one: it is faster. Correct, and the number that matters is jitter
This one is straightforwardly true, but the usual framing misses why. The problem with a distant data centre is not average speed, it is variance. A control loop can be designed around a fixed delay. It cannot be designed around a delay that is usually two milliseconds and occasionally forty.
The industrial networking standards make the target explicit. Time-Sensitive Networking, the set of IEEE 802.1 extensions to standard Ethernet, is specified around figures of the order of 1 millisecond cycle time, 1 microsecond of jitter and 99.999 percent reliability for demanding automation traffic. Those are not cloud numbers and no amount of bandwidth turns them into cloud numbers, because the constraint is distance and queuing rather than throughput.
On the wireless side, 3GPP Release 16 introduced ultra-reliable low-latency communication together with Ethernet PDU sessions, over-the-air time synchronisation and quality of service mapping, which is what allows a 5G system to carry time-sensitive traffic and interwork with an IEEE 802.1 TSN domain. Release 17 continued that work. This is the practical mechanism behind the private 5G deployments now appearing in plants, and it is worth knowing that the specification exists rather than treating the capability as a vendor promise.
| Workload | Where it belongs | Why |
|---|---|---|
| Motion control, safety interlocks | Controller, on the machine | Deterministic timing, no network dependency tolerated |
| Vision inspection, anomaly detection | Edge node on the line | High data volume, low value once the verdict is known |
| Model training, fleet comparison | Central cloud | Needs data from many sites, latency irrelevant |
| Scheduling, ERP integration | Central systems | Business-layer decisions, seconds are acceptable |
Claim two: it is more secure. This one is wrong as usually stated
The argument runs that processing locally keeps sensitive data off the network, so exposure falls. The half of it that is true is narrow: data not transmitted cannot be intercepted in transit. The half that is missing is that edge deployment multiplies the number of internet-facing, physically accessible, infrequently patched devices in the estate.
The 2025 Verizon Data Breach Investigations Report is unambiguous on the direction. Edge devices and VPNs accounted for 22 percent of the targets in breaches involving exploitation of a vulnerability, against 3 percent in the previous edition, an increase of roughly eight times. The remediation picture is worse than the exposure one: of the edge device vulnerabilities organisations worked on, around 54 percent were fully remediated over the year, with a median of 32 days to get there. In an industrial context, where a patch window competes with a production run, both of those numbers get harder rather than easier.
This is also why the American cybersecurity agency CISA has published dedicated guidance on edge device security and on reducing the attack surface of end-of-support edge devices, alongside international practitioner guidance issued in February 2025. The infrastructure that makes the plant responsive is the same infrastructure attackers now prioritise.
The correct conclusion is not to avoid the edge. It is to budget for IEC 62443, the international standard series covering security for industrial automation and control systems, at the same time as the hardware, and to treat network segmentation and a patch process as part of the project rather than as a later phase that never arrives.
Claim three: it is inevitable. Probably, and the famous number is a forecast
The statistic quoted in almost every article on this subject is Gartner’s projection that by 2025, 75 percent of enterprise-generated data would be created and processed outside a traditional centralised data centre or cloud, up from around 10 percent when the forecast was made. It is a reasonable directional statement and it is worth being precise about what it is: a forecast, not a measurement. We have not seen a credible published audit confirming the figure was reached, and the honest position is that the trend is real while the number is a projection that has been repeated until it sounds like an observation.
What is observable is the standards activity, which tends to be a better indicator than market forecasts because it is expensive and slow. The OPC Foundation’s Field Level Communications initiative, and the OPC UA FX specifications that came out of it, exist to make controller-to-controller and controller-to-device communication vendor-independent at exactly the layer edge computing operates in. Standards bodies do not do that work speculatively.
Where the edge is the wrong answer
Three cases come up repeatedly, and in each the edge deployment is a cost rather than a capability. The first is any workload whose value comes from comparing sites, since a model that only sees one plant learns one plant. The second is anything the organisation cannot maintain: an edge node running an unsupported operating system in a cabinet nobody opens is a liability with a payback calculation attached to it. The third is data that was never time-critical, where local processing is being used to solve a bandwidth cost problem that a schedule change would solve more cheaply.
The reliable test we apply is to ask what breaks if the link to the central system is down for an hour. If the answer is “production stops”, the workload belongs at the edge and the case is made. If the answer is “a report is late”, it does not, whatever the architecture diagram says. The same logic applied to a whole site is what separates a genuinely smart factory from an instrumented one.
Three questions we get from plant teams
Does edge computing replace the PLC?
No. Programmable logic controllers still handle deterministic control and safety functions. Edge nodes sit alongside them, taking the analytical work that a controller was never designed for. Projects that treat the two as competing tend to end up with neither being trusted.
Can older machines be brought in without replacement?
Usually yes, and it is the majority of the work in practice. Retrofitting means adding sensing and a gateway rather than changing the machine, which is why the exceptions list on any site is dominated by equipment whose control system cannot be touched during production rather than by equipment that is too old.
How much of this needs 5G?
Less than the marketing suggests. Wired Ethernet with TSN covers most fixed installations more cheaply and more predictably. The wireless case is genuinely strong for mobile assets, reconfigurable lines and anything on a moving vehicle, which is a real set of applications but not the default one.
Want the application where this pays back first?
Predictive maintenance is the workload that most often justifies putting compute on the line, and its claims are the easiest to test.
Sources: Verizon, 2025 Data Breach Investigations Report, for edge devices and VPNs representing 22 percent of targets in vulnerability-exploitation breaches against 3 percent in the previous edition, for approximately 54 percent full remediation and a median of 32 days; IEEE 802.1 Time-Sensitive Networking working group material, for cycle time, jitter and reliability targets in industrial profiles; 3GPP Release 16 and Release 17, for ultra-reliable low-latency communication, Ethernet PDU sessions and 5G to TSN interworking; OPC Foundation, Field Level Communications initiative and OPC UA FX; IEC 62443 series, security for industrial automation and control systems; United States Cybersecurity and Infrastructure Security Agency, edge device security guidance and the February 2025 practitioner guidance on mitigation strategies for edge devices; Gartner, projection that 75 percent of enterprise-generated data would be created and processed outside a traditional centralised data centre or cloud by 2025, cited here as a forecast rather than a verified outcome. Updated August 2026.
