Made to measure comes out cheaper. And you are the one left with the work.

Where a large manufacturer sells closed configurations from a catalogue, here you start from a board and a chassis and build what is needed. That allows things that do not exist in a catalogue or cost far more: a chassis with a great many drive bays, a server with room and cooling for several real graphics cards, or a short machine for a shallow cabinet.

In video that has a clear use. The two problems that press hardest are terabytes and capacity to analyse, and both are solved with density: many drives in little rack space, and card slots with the power supply and the airflow those cards really ask for. Per terabyte and per card, this route almost always comes out cheaper.

The trade-off is not the quality of the part: it is the service around it. What a large manufacturer sells along with the metal —a support network, a contract with hours written into it, a part number that will still exist in five years— here depends on who builds it and who maintains it. It can be perfectly well sorted, and it has to be checked rather than assumed.

Where a lot of terabytes per rack unit are needed: installations with hundreds of cameras, or with few but very detailed ones and many days of retention. There the difference between twelve bays and thirty-six is not a luxury, it is whether the system fits in the cabinet you have.

And where there is real analytics, with several graphics cards working at once. A chassis designed for that has enough power supplies, the power connectors needed and an airflow that goes through the cards; putting those cards in a server that was not expecting them is the most common way for them to throttle themselves on heat and deliver half.

It also fits when the client has an IT team with judgement and appetite: people who know what they want, keep spares on the shelf and prefer to control the configuration rather than buy a closed package. In that case this route is a good decision and not a cut.

You gain price per terabyte and per card, and you gain fit: the machine is exactly the one you need, without paying for what you do not use and without giving up what you do. You also gain density, which in a full cabinet is worth real money.

What you lose, above all, is replacement time with an owner. A large manufacturer's contract says how many hours pass before somebody arrives with the part; on a built machine, that answer depends on who you ring and what is on the shelf. For a recording server that matters: every day of fault is a day of video that does not exist, and it does not come back afterwards.

You also lose management integration. The remote console exists and works, but fitting it into the tools the IT department already uses for the rest of the servers usually takes more work. And the firmware is tracked piece by piece rather than as a set, which means somebody has to write down what versions that machine carries, because nobody is going to remember in three years' time.

That is why the rule we use is simple: this route works out when there is somebody —the client's or ours— who takes charge of that machine by name, and there are spares decided in advance. If neither of the two exists, the saving is paid back with interest on the first bad day.

In a remote site with nobody who can open the cabinet, no. A machine you have to travel to needs the best replacement contract there is, and that is the large manufacturers' ground.

Nor where the client has standardised on one brand and one procedure for all their servers. Coming in with a different machine for the video breaks their inventory, their firmware plan and their contract, and that cost does not show up in the quotation but it gets paid.

And we would not put it forward as an argument for squeezing a quotation. If the only reason to choose this route is that it comes out cheaper, the first fault will eat the saving. There has to be a technical reason: drive density, card slots or a particular shape that does not exist in a catalogue.

A machine of this kind is administered like any server: remote console, monitoring of drives and power supplies, firmware and alerts. There is nothing exotic. The difference is that the work of knowing what is inside it and where each part is obtained does not come packaged: somebody has to do it, once, and leave it written down.

That is why the question we ask before proposing this route is who that somebody is going to be. If the client has a department with judgement, this is a good decision and they run it without us. If they do not, there are two honest ways out: that the maintenance with spares is ours and is in the contract, or buy from a catalogue and pay the difference.

What does not work is leaving it half done: a made-to-measure machine with no owner and no paperwork is a fault waiting to happen, and the day it comes nobody will even know which power supply to order.