the network core
Think of the road system between cities: you do not store anything on the highways, you just drive your car along them from one place to another. The network core is the equivalent — the mesh of packet switches (mostly routers) and the high-capacity links between them that move data across the network, from one edge to another. It is the middle of the picture, away from the end systems.
The core's one job is to forward packets toward their destination. A packet arrives, the router reads its destination address, looks it up in a forwarding table, and sends it out the right link toward the next router — over and over, hop by hop, until the packet reaches the far edge. The core does not understand your application, does not open your messages, and ideally keeps no per-conversation state; this is what packet switching buys you. The links here are fast and heavily shared, carrying many users' packets mixed together (statistical multiplexing).
There is a real choice in how a core works: today's Internet core uses packet switching (data is chopped into packets that share links on demand), whereas the old telephone network used circuit switching (a dedicated path reserved for the whole call). Packet switching wins for bursty computer traffic because it shares capacity flexibly, but it offers no guaranteed rate or delay by default — which is why quality-of-service and congestion control exist.
A packet from London to Tokyo may pass through ten or more core routers; at each one it waits in a queue, gets its address read, and is sent on — no router keeps a record of your whole session.
The core forwards packets hop by hop and, by design, stays simple and stateless.
The core is fast but not magic: it cannot beat the speed of light, and shared links can congest. Adding core capacity raises throughput but does not lower the propagation delay of a long path.