an interconnect
If a shared bus is a single hallway everyone must squeeze through one at a time, an interconnect is the building's whole system of private corridors and a switchboard that routes each message directly to its destination. An interconnect is the general name for the wiring and switching that ties components together. Modern designs favour point-to-point links — a private pair of wires between two endpoints — and switches that forward messages between links, so many conversations can happen at once instead of taking turns on one shared wire.
The key contrast is shared versus switched. On a shared bus, all devices listen to the same wires and must arbitrate for a turn, so total bandwidth is fixed and split among them. In a switched interconnect, each device has its own dedicated link to a switch; if device A talks to B at the same time C talks to D, both transfers run in parallel at full speed because they use different links. Aggregate bandwidth grows as you add links, and one slow device no longer blocks everyone. The price is more wires and the cost and latency of the switches themselves.
Interconnects appear at every scale, and the same idea repeats. Inside a chip, a network-on-chip links many cores. Between a CPU and its peripherals, PCI Express is a switched point-to-point interconnect. Between machines in a datacenter, the network is an interconnect of switches. Beginners should notice the through-line: as systems grew past what one shared bus could feed, architecture moved from buses to switched fabrics. The word 'bus' lingers in product names (the 'PCI bus'), but the topology underneath is point-to-point.
Shared bus with four devices and 800 MB/s total: if all four want to transfer at once they each get about 200 MB/s. Switched interconnect with a private 800 MB/s link per device: two non-overlapping pairs can each run at the full 800 MB/s simultaneously, for 1600 MB/s of useful aggregate traffic.
Switching turns one shared pipe into many private pipes, so independent transfers no longer compete.
Do not assume an interconnect is always faster per link than an old bus — its win is parallelism and scaling. A single transfer over one link may even have slightly higher latency than a bus because it passes through a switch.