There’s a classic problem in IoT projects that any telecom engineer knows well: thousands — sometimes millions — of devices generating data in the field, and no real capacity to process all of it at the edge. Industrial temperature sensors, surveillance cameras, smart energy meters, connected vehicles. Each with its own memory, power, and processing constraints.
Each one producing a constant stream of data that simply doesn’t fit on the local hardware.
The natural answer is to integrate IoT with Cloud Computing as a well-defined division of responsibilities: the edge collects and transmits, the cloud processes and analyzes. What looks simple on paper reveals a series of engineering challenges that are still being solved in 2025 — and that’s exactly what this post is about.
The Three-Layer Architecture
The canonical topology of a cloud-based IoT solution is organized into three distinct layers.
At the base are the IoT devices — sensors, actuators, cameras, wearables, connected vehicles. Each device has a unique IP address and the ability to send and receive data over the network, but with severe processing and power constraints.
They weren’t built to think; they were built to sense and send data.
In the middle sits the collection and aggregation layer — what the classic diagram calls IoT Data Collection. This is where the physical gateway or software middleware lives. It’s responsible for pre-processing, filtering, and compressing the data before sending it to the cloud. Without this layer working well, you’ll flood the network core with unnecessary noise and blow through your data transfer budget in weeks.
At the top, the cloud layer provides all the elastic infrastructure: virtualized compute, distributed storage, analytics at scale, ML models, digital twins.
The devices can cover the most diverse verticals. The same three-layer architecture supports all of these scenarios, with variations in communication protocol (MQTT, CoAP, HTTP) and in the chosen middleware.
Challenges of IoT-Cloud Architectures
IoT-Cloud integration is not a solved problem. There’s a well-documented list of issues that remain the subject of active research:
Security and privacy top the list. Devices with limited hardware are easy attack surfaces. Proper authorization — guaranteeing that only authorized users access critical data — still has no standardized solution.
Unnecessary data communication: without intelligent pre-processing at the gateway, the volume of data going up to the cloud generates unacceptable delays and unnecessary cost. The gateway needs to decide what’s worth transmitting.
IPv6 deployment: IPv4 can’t address all the devices in a meaningful IoT fleet. IPv4/IPv6 coexistence still brings specific security risks.
Resource allocation and management, interoperability between platforms, service discovery for mobile fleets, scaling with proportional security, and energy efficiency complete the list. Each of these is a separate post waiting to be written around here — or a product waiting to be built 😉.
Conclusion
Ten years ago, players like AWS IoT showed that IoT-Cloud integration went from laboratory experiment to mission-critical infrastructure at global scale. The three-layer architecture proved to be the standard to follow. The open problems became more specific and harder — which is a sign of maturity, not of market failure.
For anyone designing IoT systems, the good news is that the fundamental building blocks are available and well documented today with AWS IoT, Azure Hub, or on-prem technologies like EMQX, Kafka, and the like.
The bad news is that the devil still lives in the details — security, interoperability, and scale management still require engineering decisions that no platform makes for you or hands you ready-made today.