Edge computing moves compute to the edge of the network — to where data originates and decisions take effect: the factory, the branch, the cell tower, the provider’s point of presence. Instead of sending everything to the central cloud, data is pre-processed, filtered and acted on locally.
The drivers are physical and economic: the speed of light sets latency limits, bandwidth costs money, and some data must not leave the site. Edge is not a replacement of the cloud but its extension — orchestration stays central, execution moves outward.
The flavours of edge
On-premises edge sits at the company itself: industrial PCs and local clusters in production and branches, evaluating machine data in milliseconds — camera-based quality inspection, anomaly detection, local buffering during WAN outages. Network edge sits with the provider: compute capacity in the points of presence of global networks, placing applications and security functions a few milliseconds from users.
In between lie telco edge (MEC in mobile networks, relevant with 5G campus networks) and regional micro data centers. The common pattern: latency-critical and data-intensive work runs outside, training, long-term analytics and management inside — connected by a WAN that must carry this architecture.
Typical use cases
- Industry: predictive maintenance and vision processing at the line, decoupled from the WAN.
- Retail: branch systems that keep selling and deciding locally even during line outages.
- Security at the network edge: WAF, bot and DDoS defence in the PoP before traffic reaches the origin.
- AI inference at the edge: models answer near users without collecting raw data centrally.