Part of the day: Board-level connectors & sockets

Assembled by the nightly build from this catalog’s own entry for Board-level connectors & sockets (4.07, Layer 4 — Servers, racks & integration). The tables below are rebuilt every night; first published 2026-09-23, refreshed 2026-09-24.

The physical connectors — CPU and GPU sockets, board-to-board and cable connectors, and I/O cages — that link chips, trays and cables together on and between boards.

Connectors sit at nearly every junction in the rack: CPU and GPU sockets attach silicon to the baseboard (4.03), board-to-board connectors link trays together, and I/O cages (4.07a) terminate the optical and copper interconnect coming off layer 3. As pin counts, power currents and signal speeds all rise with each GPU generation, socket and connector design becomes a harder mechanical and electrical problem — more contact points, tighter tolerances on insertion force and contact resistance, and more attention to high-speed signal integrity through the connector itself. AI racks also simply need more connectors per system than a conventional server, both because there are more trays to link and because power delivery at rack scale increasingly moves through busbars and whips (4.15) rather than cables, each with its own connector interface. For investors, this tier tends to be sticky once a connector is designed into a socket or board generation, since requalifying a different supplier’s socket against a new GPU package is not a drop-in swap; the interesting question by generation is who wins the socket design-in for the next GPU package.

AI delta: Higher pin counts, higher currents and faster signaling with each GPU generation make socket and connector design a harder mechanical problem, and AI racks need more connectors per system than a conventional server simply because there are more trays to link.

Who makes it

The full entry — sub-categories, news, videos and sources — lives on the Board-level connectors & sockets page.