Telecom equipment we source

Remote radio units sit outdoors at the top of the tower, carrying the power amplifier and transceiver away from the cabinet.

An active antenna unit integrates the radio and the antenna array into one assembly, which is what makes massive MIMO practical on a tower that cannot carry separate radios and panels.

The baseband unit is where a site's capacity actually lives, and because it is normally a chassis populated with plug-in boards rather than a sealed product, most enquiries are for a specific board rather than a whole unit.

A base station cabinet is an enclosure, and what makes it a working site is everything fitted inside it — baseband, radios, power, battery backup and the interconnect between them.

A GPS antenna and the antenna line devices around it are low-cost items that stop a site commissioning when they are wrong, which is why they generate more urgent enquiries than their value suggests.

Microwave links carry backhaul where fibre is uneconomic, which describes a great deal of network in Africa and the interior of the Gulf.

This is one of the largest categories in the catalogue, and the one where sourcing is hardest.

Fibre access rollouts across Africa and the Gulf run predominantly on a handful of OLT platforms, and expansion is normally a service board added to an existing chassis rather than a new chassis.

Pluggable optics are the most commodity-like items in this catalogue and also the ones where vendor coding causes the most trouble.

Network switch and routing hardware in carrier networks is almost always modular, and almost always tied to a software release.

Most telecom equipment failures are board failures, and most board replacements are cheaper than the platform documentation implies — provided the board is identified correctly.

Site power is the least glamorous and most failure-prone part of a telecom site, and unlike radio equipment it is largely vendor-agnostic — a −48 V rectifier of the right output and form factor will serve a site regardless of whose radios are on the tower.

Fibre optic cable and cable assemblies are ordered by length and connector, and the length is usually encoded in the part number suffix — which means a single assembly appears as a dozen near-identical part numbers that differ only in metres.

Fan units fail predictably, sunshields crack in the sun, and neither is expensive — but a fan failure will take a shelf offline and the replacement part is model-specific in a way that is almost never documented publicly.

Rack servers are the single largest line on most data centre purchase orders, and the one where specification error is most expensive.

Server processors are bought two ways: as spares to match an existing socket population, and as upgrades to lift core count inside a chassis that is otherwise fit for purpose.

Accelerator demand has changed shape faster than any other line in this catalogue.

Server memory is the most frequently mis-ordered line in the catalogue, because a module that is physically identical and electrically incompatible looks the same on a purchase order.

Server drives are bought against four independent attributes — interface, capacity, rotational speed and form factor — and getting any one of them wrong makes the drive useless in the bay it was bought for.

Enterprise storage is bought as a system rather than a part: a controller enclosure, disk shelves behind it, interface modules in the front, and disk units qualified by the array maker.

Network adapters are ordered against port count, port speed, physical media and the form factor the chassis expects.

Power supplies are a high-turnover spare and a frequent source of ordering error, because rating alone does not determine fit.

Storage controllers divide into two jobs that are often confused on the purchase order: RAID controllers, which present logical volumes and generally carry cache, and host bus adapters, which present drives directly to the operating system for software-defined storage.

Data centre switching sits between the server estate and the network, and is bought either as a like-for-like replacement for an installed switch or as capacity for a new row.

Mechanical and board-level spares are what actually hold up a server rebuild.