Hardware Refresh Timing: Replace on Risk, Not on Age
Refresh cycles are usually inherited rather than decided, and a fixed number of years is a weak reason to replace working hardware or keep failing hardware. This covers timing refreshes on risk instead.
Why a fixed cycle is the wrong default
Three or five years is an accounting convention, not a hardware lifespan.
Two failures follow from treating it as one. Working machines get replaced because the calendar says so, spending money that had better uses. And failing machines get kept because their replacement is not due yet, running risk that nobody has priced.
The useful question is not how old a machine is. It is what happens if it fails, and how quickly could you recover.
The four signals that actually mean replace
Spares are becoming hard to source. When the parts a platform needs are getting scarce or expensive, the machine has a fixed remaining life whether it is working or not. Our availability guide covers reading this.
It cannot run what it needs to. An operating system or hypervisor version dropping support for the platform is a hard deadline. Check the compatibility list for the version you will need to move to, not the one you run. Our compatibility guide covers this.
Failures are clustering. Several components failing in a short period on the same machine means the whole thing is ageing, not that you were unlucky.
Running cost exceeds the difference. Where power and cooling for an older machine exceed what a newer one would cost to buy and run over the remaining horizon. Our TCO guide covers the arithmetic.
The signals that do not mean replace
Three things that trigger unnecessary refreshes.
Age alone. A well-maintained machine with available spares, running its workload comfortably, is not a problem because it is old.
A newer generation existing. There is always a newer generation.
Warranty expiry. Warranty ending changes your support position, and that is a question about spares and response time rather than an automatic replacement trigger. Our entitlement guide covers assessing it.
Refresh is not the only option
Four alternatives, all cheaper.
Upgrade components. Memory, drives or a controller frequently cost a fraction of a new machine and extend its life meaningfully. Our upgrade or replace guide covers the call.
Move the workload down. A machine no longer adequate for its current job is frequently fine for a lighter one. Demotion beats disposal.
Consolidate. Several under-used machines onto fewer hosts reduces the estate rather than renewing it. Our consolidation guide covers this.
Hold spares and keep running. For a machine that works, has available parts, and does a job that is not growing, a shelf spare is far cheaper than a refresh. Our spares guide covers sizing that.
Planning the sequence
Where several machines need attention, order them by consequence rather than age.
First: anything whose failure would be an outage, with no spare and no redundancy.
Second: anything blocking a software or platform change you need to make.
Third: anything where running cost genuinely exceeds replacement.
Last: everything that is simply old.
And stagger it. Replacing an entire estate at once means replacing it all again at once later, which is a budget problem every few years rather than a manageable annual one.
What to do with what comes out
Decide before it is unplugged, not after it has sat in a storeroom for a year losing value.
Keep as a spare if it is a platform you still run and the parts are useful.
Redeploy to a lighter role.
Sell while it still has value. Our selling guide covers what holds value and how to prepare it.
Whichever, sanitise the data properly first. Our decommissioning guide covers the order.
Frequently asked questions
How often should servers be replaced?
Not on a fixed schedule. Three or five years is an accounting convention rather than a hardware lifespan. Replace on risk: scarce spares, an inability to run what is needed, clustering failures, or running cost exceeding the difference.
Does warranty expiry mean I should replace a server?
No. It changes your support position, which is a question about spares and response time rather than an automatic replacement trigger. Many machines run well past warranty with a spares holding.
What are the alternatives to a refresh?
Upgrading components, moving the workload to a lighter role, consolidating several machines onto fewer, or simply holding spares and continuing to run. All are cheaper than replacement.
What order should I replace machines in?
By consequence, not age. First anything whose failure would be an outage with no spare and no redundancy, then anything blocking a needed platform change, then anything where running cost exceeds replacement, then everything that is simply old.
Should I replace the whole estate at once?
Stagger it. Replacing everything at once means replacing it all again at once later, which turns into a budget problem every few years rather than a manageable annual one.
When does spares availability force a replacement?
When the parts a platform needs are getting scarce or expensive. At that point the machine has a fixed remaining life whether it is working or not, and the decision is about timing rather than whether.
Send us the platforms you run and we will tell you honestly which still have parts available and which are on a clock.
