An edge server is the operational unit of a CDN. It sits in a POP (Point of Presence) with a public IP, accepts client requests, serves content from its local cache, and on cache miss either pulls from a shield/regional cache or from origin. Edge servers are commodity x86 hardware running specialised caching software — Apache Traffic Server, Varnish, NGINX with extensions, or a CDN vendor's proprietary stack. A large CDN has tens of thousands of edge servers across hundreds of POPs.
For streaming, the edge's job is to absorb the per-viewer traffic. A single 2-second video segment for a 4K bitrate is roughly 5 MB; an edge serving 10,000 concurrent viewers of that stream pushes 25 GB/s out of its NICs. Modern edges have 100 Gbps interfaces, multi-terabyte SSD caches and tens to hundreds of CPU cores. Cache hit ratio at the edge — the share of requests served from local cache without touching origin — is the single most important operational metric.
Edge servers also run application logic. Signed-URL validation, geo-blocking, header manipulation, watermarking, ad insertion stitching, ABR rendition steering all happen at edge, often programmable through edge-compute interfaces (Cloudflare Workers, Fastly Compute@Edge, AWS Lambda@Edge, Akamai EdgeWorkers). This brings logic close to viewers, reducing round-trips and enabling per-session personalisation without going back to origin.
