In industry, edge computing means processing machine data on devices and servers inside the plant, close to where it is produced, rather than sending everything to the cloud. The payoff is fast local response and resilience when the link drops; the cost is a fleet of edge devices to patch and place carefully in the OT network.
Quick Overview
What you will learn:
- Which workloads belong at the edge and which in the cloud
- How to compare edge platforms, including their support lifecycles
- Where edge devices should sit in an OT network
- What edge AI is and when it pays off
Who this article is for:
- Automation and maintenance engineers
- Industrial IoT solution architects
- Security teams responsible for production networks
Reading time: 7 minutes
A word on where these questions come from: we selected them from the searches people run when looking for information about edge computing in industry. They are not quotes from conversations with particular clients.
Edge, Fog and Cloud: Getting the Vocabulary Straight
NIST’s Fog Computing Conceptual Model (SP 500-325 on www.nist.gov) defines the edge as the network layer that takes in end devices and their users and provides local computing capability, on a sensor or a metering device, for example. Fog computing sits between that layer and the cloud, pushing applications, management and analytics out into the network itself to deal with scale, heterogeneity and latency.
NIST SP 800-82 Rev. 3 on csrc.nist.gov brings in an industrial perspective by way of the Industry IoT Consortium’s IIoT reference model, which has an edge tier, a platform tier and an enterprise tier. The platform tier consolidates and analyses data and sends control commands down to the edge, and the enterprise tier hosts business applications and decision support.
Translated to the plant:
- Keep at the edge anything that must react quickly or keep going when the link fails: signal filtering and aggregation, anomaly detection, local alarms, buffering during outages.
- Send to the cloud or data centre anything that needs scale and broad context: model training, historical analysis, comparisons between sites, management reporting.
- Put the policy for everything in between in writing: which data leaves the plant raw, which only in aggregated form, and which never leaves at all.
The edge is not a substitute for control. PLCs and SCADA still run the process, and the edge layer reads their data and passes it on.
Choosing an Edge Platform
Cloud providers offer their own edge runtimes. Features count, but on the shop floor the support lifecycle counts just as much, because a gateway stays in place for years.
- AWS IoT Greengrass is presented on
aws.amazon.comas an open-source edge runtime and cloud service for building, deploying and managing device software, which can process data locally and keep working over an intermittent connection. AWS ended support for Greengrass V1 on 1 June 2026 and recommends moving to V2 (aws.amazon.com). - Azure IoT Edge: the release notes on
learn.microsoft.comgive 10 November 2026 as the end-of-support date for version 1.5 LTS, with 1.6 LTS as the supported release, covered until 14 November 2028. Microsoft says 1.6 contains no major feature changes compared with 1.5, so the upgrade should be straightforward.
Whatever the platform, ask: how long will the version we deploy be supported? How do we update devices without stopping the line? What happens to the data while the link is down? Does it speak the protocols our machines use, such as OPC UA and MQTT?
Placing the Edge in an OT Network
An edge device forms a bridge between the production network and the corporate network or the cloud, so it is a boundary whose position has to be chosen deliberately. The Purdue model in the Microsoft Defender for IoT documentation on learn.microsoft.com defines Level 3.5, the industrial perimeter network that separates the site levels from the enterprise.
NIST SP 800-82 Rev. 3 on csrc.nist.gov recommends a DMZ architecture that blocks direct traffic between corporate and OT networks, along with multi-factor authentication for remote access to OT systems. For edge devices, that means:
- connections initiated outbound by the edge device through the DMZ, never inbound connections into the OT network;
- a separate identity and certificate for each device, with updates managed centrally;
- an inventory of edge devices that records each one’s owner and software version.
The guide is being revised. A planning note dated 21 September 2026 on the Rev. 3 page at csrc.nist.gov refers to the initial public draft of Rev. 4, with comments due by 30 November 2026. Rev. 3 remains the current revision until a final version is published.
Read Also
- Integrating MES and ERP systems on the shop floor — ISA-95 layers and data flows to ERP
- Edge AI: data processing closer to the source — models running on edge devices
- OT cybersecurity in manufacturing plants — securing the production network
Build Your Skills
Our Edge Computing Fundamentals course covers edge architecture, cloud integration and managing a fleet of devices, while Industrial Systems Cybersecurity Fundamentals (OT/ICS) deals with OT architectures, the Purdue model and risk analysis. Get in touch and we will adapt either course to your team.
Frequently Asked Questions (FAQ)
What is edge computing?
Edge computing is a model in which data is processed close to where it is generated, on local devices or servers, and only the results or selected data go to the cloud. In a factory, the edge is typically a gateway or server on the shop floor that collects data from the machines. NIST’s SP 500-325 on www.nist.gov describes the edge as the layer of end devices with local computing capability.
How should we choose an industrial IoT platform and edge solution?
Begin with what the plant needs: machine protocols, offline operation, the update process and security constraints. Only then compare platforms, version lifecycles included. According to learn.microsoft.com, the end-of-support date for Azure IoT Edge 1.5 LTS is 10 November 2026, and AWS ended support for Greengrass V1 on 1 June 2026 (aws.amazon.com).
What is edge AI?
Edge AI means running machine learning models directly on edge devices instead of sending raw data to the cloud. The AWS IoT Greengrass page on aws.amazon.com lists ML inference at the edge and gives anomaly detection as an example. Models are usually trained centrally, and the finished version is then deployed to the edge.
What is the biggest challenge of edge computing in industrial IoT?
Connectivity and long-term maintenance. An edge device has to keep running and buffering data when the link drops, and it has to be patched and secured for years. Plan fleet management, version lifecycles and network placement under the Purdue model on learn.microsoft.com from the very start.