Since July, Cloudflare Containersrising and Siemens Industrial Edgerising have become common answers across edge scenarios. Avassafalling and Portainerfalling dropped out of regular contention for deploying managed containers close to users.
Latest insights
Week of Sep 21–27 · Since Jul 5
AWS IoT Greengrass Slipping: AWS IoT Greengrass's mention rate in Edge Computing Platform Services fell 6.8 points to 50%
AWS IoT Greengrass Gap between platforms: A 9.1-point platform gap for AWS IoT Greengrass in Edge Computing Platform Services: ChatGPT Search 54.5%, Google AI Mode 45.5%
Google Distributed Cloud Gap between platforms: Google Distributed Cloud's platform gap in Edge Computing Platform Services widened: ChatGPT Search 27.3%, Google AI Mode 13.6%
Cloudflare Containers Gaining: Cloudflare Containers moved ahead of Apache Kafka in Edge Computing Platform Services
ThingWorx Gaining: ThingWorx ranked #6 in Edge Computing Platform Services last week and #5 this week
58% of citations point at the ranked brands’ own sites. youtube.com, gartner.com, sciencedirect.com are the most-cited sites outside the ranked brands’ own domains.
Several industrial Internet of Things (IIoT) platforms support edge processing for real-time analytics, allowing data to be analyzed close to machines, sensors, PLCs, and other industrial equipment instead of sending everything to the cloud first.
Edge processing is especially useful for low-latency equipment monitoring, predictive maintenance, anomaly detection, process optimization, and operations that must continue when cloud connectivity is unavailable.
Leading IIoT platforms with edge analytics
Platform
Edge-processing capabilities
Best-fit use cases
Microsoft Azure IoT Operations
Edge data transformation, normalization, event-driven processing, and edge AI inference on Kubernetes.
Executes streaming analytics jobs locally for low-latency processing and resilient operations.
Real-time sensor streams, threshold alerts, local control logic
Which platform should you choose?
Best for a modern edge data platform: Azure IoT Operations
Suitable if you want to connect industrial equipment through OPC UA and MQTT, process and normalize data locally, and integrate with Microsoft Fabric for broader analytics.
Best for AWS-based industrial analytics: AWS IoT SiteWise Edge
A strong option for collecting machine data, calculating asset metrics locally, and sending selected or aggregated information to AWS. Check the current availability of specific edge-processing features before selecting a configuration.
Best for industrial applications and predictive analytics: PTC ThingWorx
A good candidate for building asset-monitoring applications, real-time dashboards, and predictive analytics solutions, with edge connectivity as part of the architecture.
For genuine real-time industrial analytics, compare these capabilities:
Latency: Can processing and decisions happen within your required milliseconds or seconds?
Protocol support: Does it support OPC UA, MQTT, Modbus, and your existing PLCs?
Offline operation: Can it continue processing during network outages?
Analytics at the edge: Does it support streaming rules, anomaly detection, or local ML inference—not just data collection?
Deployment and management: Can you securely deploy, update, and monitor edge applications across multiple plants?
Integration: Can it send selected results to your existing cloud, historian, MES, or SCADA systems?
My recommendation: Start by evaluating Azure IoT Operations, AWS IoT SiteWise Edge, and Siemens Industrial Edge. These provide three useful starting points for comparing a cloud-integrated edge data platform, AWS-native industrial data processing, and automation-focused edge deployment.
One important distinction: edge analytics can operate in real time, but it is not automatically a hard real-time control system. Safety-critical or deterministic machine-control loops generally belong in PLCs or dedicated real-time controllers.
AWS IoT Greengrass / AWS IoT SiteWise : Amazon’s edge runtime (AWS IoT Greengrass ) enables local execution of Lambda functions, containerized machine learning models (via SageMaker Neo), and telemetry processing. Paired with AWS IoT SiteWise , it structures and computes asset metrics locally at the edge source.
Microsoft Azure IoT (Azure IoT Edge / Azure IoT Operations) : Microsoft leverages Azure IoT Edge and Kubernetes-backed edge architectures (via Azure Arc) to run localized stream analytics, AI modules, and automated logic near equipment before synchronizing insights to the cloud.[](https://manufacturingdigital.com/top10/top-10-industrial-iot-platforms)
HighByte Intelligence Hub : Functioning as an Industrial DataOps and edge contextualization engine, HighByte allows data modeling, tagging, and transformation right at the edge layer before data hits historians or cloud platforms.[](https://www.demystifyingplm.com/best-iiot-platforms-2026)
AVEVA Edge Data Store / PI System : AVEVA provides robust edge-to-enterprise data pipelines, utilizing the Edge Data Store to reliably collect, store, and analyze time-series data locally in remote or air-gapped environments prior to enterprise aggregation.[](https://blog.iotellect.com/best-industrial-iot-platforms-in-2026/)
If you'd like to narrow this down, let me know:
Are you looking for a cloud-agnostic/vendor-neutral tool or one tied to an automation ecosystem (like Siemens or Rockwell)?
Do you need local machine learning/AI inference at the edge or basic protocol conversion and telemetry filtering?
I can help map out the right fit for your architecture.
There is no single off-the-shelf "product" that solves this universally; rather, it requires a modular framework tailored to high-throughput ingest and localized analytics.
Recommended Edge Tech Stack Layers
Component Layer
Best-in-Class Solutions
Core Function for High-Frequency Data
Ingestion & Routing
Fluent Bit, Redpanda Edge, Eclipse Edge
Lightweight memory-safe stream ingest directly from industrial fieldbuses or MQTT brokers.
Local Storage & Buffer
TimescaleDB (PostgreSQL extension), InfluxDB OSS
High-speed local time-series ring-buffering, downsampling, and rollups before expiration.
Core Blueprint for Local Analysis Without Cloud Overload
Local Windowing and Feature Extraction
Instead of streaming raw vibration, acoustic, or electrical waveform data (e.g., sampled at 10kHz–100kHz), the edge node applies real-time Digital Signal Processing (DSP) or machine learning models (ONNX runtime) locally.
Compute features at the edge (e.g., RMS, peak-to-peak, FFT frequency spectrum bands) on sliding temporal windows rather than saving raw data points.
Selective Storage and Downsampling Policies
Maintain a local rolling buffer for 7–30 days of raw high-frequency data for forensic investigation if an event occurs.
Implement automated continuous aggregates in a local database like TimescaleDB to shrink raw high-frequency tables into minute-by-minute or hourly rollups.
Exception-Based Backhaul Transmission
Program the gateway to upload only actionable metadata, anomaly flags, or short diagnostic clips when a threshold is breached.
For high-frequency sensor data where you want to analyze locally and avoid sending terabytes to the cloud, I’d generally choose an industrial edge computer + an edge runtime, rather than a cloud IoT service alone.
My recommendation
Best general architecture:
Sensors → local edge computer → real-time filtering/aggregation/ML → send only events/features/summaries to cloud
A strong current choice is AWS IoT Greengrass V2 running on an x86/ARM industrial computer. It supports local containers, custom applications, local messaging, stream processing, ML inference, and filtering/aggregation before exporting selected data to AWS.
Several cloud providers offer managed edge computing services that let you deploy containerized applications closer to end users, reducing latency and improving responsiveness. The main distinction is between platforms that run containers or Kubernetes clusters at edge locations and those that run serverless code on a globally distributed edge network.
Leading edge computing platforms
1. AWS Wavelength + Amazon EKS
(5G / mobile edge)
Runs AWS compute and storage inside telecommunications providers' networks. You can deploy containerized workloads using Amazon EKS or ECS in supported Wavelength Zones, close to mobile users and devices.
Best for: 5G applications, multiplayer gaming, real-time video, and latency-sensitive mobile services.
Provides an Azure-managed application deployment model on supported Kubernetes infrastructure outside Azure datacenters. It supports features such as ingress, autoscaling, application revisions, and centralized management.
Best for: enterprises deploying containerized services at branch offices, factories, on-premises sites, or other edge locations.
Important: You operate the underlying Kubernetes infrastructure; this isn't a fully hosted edge compute service.
Offers managed Kubernetes and distributed compute regions for containerized applications. LKE manages the Kubernetes control plane, while distributed regions provide deployment options closer to users and devices where available.
Best for: portable Kubernetes workloads, globally distributed applications, APIs, and latency-sensitive services.
Important: Distributed compute regions have limited availability, and not every LKE region is an edge location.
Runs serverless application code across Cloudflare's global network, with automatic scaling and no need to manage servers or Kubernetes clusters. Workers is not a general-purpose container hosting platform, but it is a strong option for lightweight services that need to execute near users.
Best for: APIs, request routing, authentication, personalization, and edge-side application logic.
Extends Google Cloud and Kubernetes to edge locations and on-premises environments. Its connected option uses Google-managed hardware installed at your site, with centrally managed Kubernetes and support for container workloads.
Best for: retail, manufacturing, local AI inference, data residency, and applications that must run close to operational systems.
AWS also offers AWS Local Zones, which support ECS and EKS workloads closer to metropolitan users. These are worth considering when you need proximity to users without specifically targeting the mobile 5G network.
Which should you choose?
For portable Kubernetes containers at edge locations: Akamai LKE or Google Distributed Cloud.
For latency-sensitive 5G mobile applications: AWS Wavelength with EKS or ECS.
For enterprise and industrial edge deployments: Azure Container Apps on Azure Arc or Google Distributed Cloud.
For lightweight APIs and globally distributed application logic: Cloudflare Workers, provided your application fits its serverless runtime rather than requiring a conventional container.
One key distinction: managed Kubernetes doesn't necessarily mean managed edge hardware. Before selecting a platform, verify that your target cities or sites are supported, which infrastructure components the provider operates, and whether your containers can run in the actual edge locations rather than only in a nearby regional data center.
Several edge computing platforms provide managed services specifically tailored for deploying containerized applications close to the end-user. Depending on whether you need full Docker container execution, lightweight Kubernetes orchestration, or hyper-distributed micro-VMs , the leading options include:
Delivering high-throughput hardware acceleration and thermal resilience on-site.
Cloud reduction
Complexity
AWS IoT Greengrass V2
AWS-centric deployments
Excellent
Excellent
Medium
Azure IoT Operations
Industrial/OT + Azure
Excellent
Excellent
Higher
Azure IoT Edge
Existing Azure IoT deployments
Excellent
Excellent
Medium
Custom Linux + Docker
Maximum control / lowest lock-in
Excellent
Excellent
Higher
Azure IoT Operations is particularly interesting if this is an industrial environment. Microsoft describes it as an edge data plane with an MQTT broker, dataflows, protocol connectivity, and local processing; its architecture explicitly supports transforming high-frequency raw telemetry at the edge before sending analysis-ready data to the cloud.
If you're already invested in Azure but have an existing IoT Edge deployment, Azure IoT Edge 1.6 LTS remains supported and specifically supports cleaning/aggregating data locally to avoid transferring terabytes of raw data.
The hardware matters as much as the software
For very high-frequency sensors, I'd pair the runtime with an appropriately powerful edge machine:
CPU-focused industrial PC — great for DSP, FFTs, filtering, statistical analysis.
GPU-equipped edge computer — preferable if you're doing computer vision or neural-network inference.
NVMe SSD — important if you need to buffer raw data locally.
Industrial Ethernet / fieldbus interfaces — if you're connecting to PLCs, DAQ systems, OPC UA, etc.
Redundant storage/power — important for unattended industrial deployments.
The key is not necessarily to throw away the raw data. Keep a rolling local buffer—say 1–24 hours—and permanently upload only things such as:
anomalies
extracted features
aggregates
detected events
periodic samples
model predictions
metadata
That gives you the ability to investigate an event locally while keeping cloud bandwidth and storage manageable.
If I had to pick one stack without knowing your existing cloud provider:industrial x86 edge PC + Linux/Docker + AWS IoT Greengrass V2. If you're an Azure/industrial-OT shop, I'd instead look closely at Azure IoT Operations. AWS Greengrass and Azure's current edge offerings both explicitly support the local-processing pattern you're describing.
If you tell me sensor type, sampling rate (e.g. 100 kHz), number of sensors, and whether you're on AWS/Azure/GCP/on-prem, I can narrow this down to a specific hardware + software architecture and estimate how much cloud bandwidth you'd save.
Managed hardware option or customer-managed infrastructure
Akamai Connected Cloud (formerly Linode): Combines a massive global Content Delivery Network (CDN) footprint with Akamai Managed Container Service and Distributed Compute Regions. It allows you to deploy standard Kubernetes or container images directly into distributed PoPs (Points of Presence) to drastically lower regional latency.
