Both arrangements put a physical machine in a data centre with power and connectivity. The difference is ownership, and everything that follows from it.
Rented dedicated hardware
The provider owns the machine. You pay monthly for the use of it, and hardware failure is their problem: they hold spares, they replace the failed disk, and the replacement time is part of what you bought.
No capital outlay, a predictable cost, and an upgrade is a migration to a different machine instead of a purchase.
The constraint is that you take what is offered. An unusual requirement. A specific card, a particular disk arrangement, more drive bays than the standard build, may simply not be available.
Colocation
You buy the server. You rent space in a rack, power, connectivity and an address.
The recurring cost is lower, sometimes substantially. In exchange you own the hardware, which means you own its failures.
You choose the exact specification, which is the real reason organisations colocate: a requirement the rental market does not serve.
The question that actually decides it
Not the monthly figure. Who replaces a failed disk at three in the morning, and how quickly?
With rented hardware, the provider, from spares held on site, within an agreed time.
With colocation, you, either by travelling there, or by asking the facility's staff to do it, which is billed by time and requires them to have the part. If your spare is in a cupboard at your office, the recovery time includes getting it there.
That answer should be worked out before the first failure, not during it. Monitoring server health explains detecting the failure in the first place, which on a redundant array is silent by design.
Remote hands
Every facility offers staff who will perform physical tasks on request, billed by time and usually with a minimum.
It works well for simple, well-described jobs: reseat a cable, swap a labelled disk, press the power button. It works badly for anything requiring judgement, because the person doing it does not know your system.
Clear labelling inside the machine is what makes this usable. A colocated server with unlabelled drives is one where a remote-hands request is a risk in itself.
What colocation demands of you
Spare parts you actually hold. A documented build, so any competent person can act on your instructions. Firmware and hardware maintenance that nobody else is doing. Server hardening explains the management controller, which on your own hardware is entirely your responsibility.
And a replacement plan, because hardware ages. A rented machine is refreshed by moving; an owned one is refreshed by buying again.
When each makes sense
Rent for one or a few machines, when you have no hardware expertise on hand, when predictable cost matters more than total cost, and when the standard specifications fit.
Colocate when you have several machines to spread the effort across, when someone is genuinely responsible for hardware, when you need a specification nobody rents, or when a regulatory requirement means owning the equipment.
For a single server, colocation rarely pays. The saving is real and the operational burden is the same whether you have one machine or ten.
The middle option
Managed dedicated hosting sits between: rented hardware where the provider also administers the operating system.
That trades money for the administration burden in place of the hardware one, and it is a different question from this one. Comparing shared, VPS and dedicated goes into where each tier sits.
Either way, the same disciplines apply
Backups off the machine. Monitoring that alerts rather than displays. A tested recovery path.
Owning the hardware changes who fixes it, not whether you need to be able to recover without it. Backups and disaster recovery deals with that half, which is identical in both arrangements.
Cost it over the whole term, not per month
The two arrangements are compared on a monthly figure and the difference is mostly in the parts that are not monthly.
Owned hardware carries the purchase, the spares you have to hold, the replacement cycle and whatever it is worth at the end. Rented hardware carries none of those and a higher recurring figure that continues indefinitely.
Over three years the totals are frequently closer than either side suggests. Over five, ownership usually wins on hardware and loses on the time spent maintaining it, which is the cost nobody puts in the comparison.
Work it out for your own numbers rather than accepting the general claim. The break even point depends heavily on how much hardware and how much of your time is involved.
The failure you plan for is a disk
Whatever the arrangement, the component that fails is storage, and the difference between the two options is what happens next.
smartctl -H /dev/sda 2>/dev/null | tail -2 storcli /c0 show 2>/dev/null | grep -iE 'state|degraded'
With rented hardware a replacement is the provider's obligation and their timescale. With owned hardware it is your spare, on a shelf, in the building, and somebody has to be there to fit it.
Both work. What does not work is owning the hardware without holding spares, which is the arrangement that turns a two hour repair into a week of waiting for a delivery. Decide which model you are actually in and stock accordingly.
Know what you can reach without a person there
The practical difference between the two shows up at three in the morning, and it is about access rather than about hardware.
ping -c 2 ipmi.example.com 2>/dev/null | tail -2 nc -zv -w 3 ipmi.example.com 443 2>&1 | tail -1
Out of band access to the machine, a way to power cycle it, and console access that works when the operating system does not are the three that determine whether a problem is fixed remotely.
Confirm all three work before you need them, and confirm the credentials are held by more than one person. A machine you can only recover by visiting is a machine whose recovery time is a travel time. Using IPMI and out of band management covers setting it up.