Fog Computing
Fog computing places useful computing power between a device and a distant cloud. Instead of sending every camera frame, sensor reading, or audio clip across the internet, a nearby gateway or local server can handle the urgent work close to where the data is created.
How it works
A fog layer is made of devices such as industrial gateways, Wi-Fi routers, roadside units, local servers, or small edge clusters. These machines sit near many constrained endpoints—battery sensors, cameras, phones, or microcontrollers—but have more memory, processing power, and network capacity than the endpoints themselves. They can run model inference, combine data from several devices, filter unimportant events, and send only selected results to the cloud for long-term storage or heavier analysis.
The practical split
A useful way to picture fog computing is as a local control room: close enough to react quickly, but powerful enough to coordinate more than one small device. For example:
- A factory gateway collects vibration readings from dozens of machines, runs an anomaly-detection model, and raises a local alarm without waiting for an internet connection.
- A smart-camera system sends video to a nearby NVIDIA Jetson-based server for object detection, while uploading only detected incidents rather than continuous footage.
- A vehicle or roadside unit processes traffic data locally, where delays measured in seconds could make a decision useless.
Why it matters for Edge AI
Fog computing helps when an endpoint cannot carry the full model or workload. A Cortex-M sensor node might run a tiny wake-word or threshold model, then pass a suspicious audio segment to a local gateway running a larger model through ONNX Runtime or another accelerator-aware runtime. This arrangement reduces bandwidth use, protects sensitive raw data from leaving a site, and keeps responses working during cloud outages. Ignoring the fog layer can force every device to choose between an oversized on-device model and slow, costly cloud inference—both poor fits for real-time, power-limited deployments.
Fog computing is a distributed computing model that places processing, storage, and network services on intermediate nodes—such as gateways, routers, or local servers—between edge devices and the cloud. For Edge AI, it lets resource-constrained sensors and devices offload heavier inference or coordinate locally while avoiding cloud round trips. This reduces latency, bandwidth use, and dependence on continuous internet connectivity.
Think of fog computing as a helpful middle stop between a device and a faraway cloud server. Instead of every security camera, factory sensor, or smart speaker sending all its information across the internet, a nearby computer can handle much of the work.
For example, cameras in a stadium might send video to a small computer in the building, which quickly spots overcrowding and alerts staff. Only important results—or selected footage—need go to the cloud. This can make AI responses faster, reduce internet traffic, and keep more sensitive data close to where it was created. Fog computing is especially useful when devices need quick decisions but are not powerful enough to do everything alone.