Five-year refresh planning is one of the more common mistakes I see procurement teams make. Not because they refresh, that part is necessary, but because they refresh based on age, vendor recommendation, or the configuration that worked last time, rather than on a measured view of what their estate actually does and what would replace it best.
The pattern is consistent. A team identifies that the existing estate is approaching end of life. They request quotes from their incumbent vendor. The configurator produces a like for like replacement, often with some headroom built in. The capex is signed off, the order is placed, and the new hardware lands roughly the same shape as what it replaced. A year later, utilisation graphs show the same patterns. Three years later, the budget cycle starts again.
There is a better way to do this. Three principles.
One. Start with the workload, not the hardware.
The first question is not "what should we buy." It is "what are we running, and how is it actually performing." A refresh plan that is not informed by current utilisation, peak load profiles and headroom usage is solving for the wrong thing. The cheapest, cleanest carbon outcome is almost never a like for like replacement. It is a denser configuration, or a wider one, or a consolidation play that retires more machines than it deploys.
Two. Address the worst-performing 10% first.
A full estate replacement is a procurement event most organisations cannot or should not undertake in a single cycle. The more useful move is to identify the 10% of servers performing worst on energy per unit of work, and replace those on a rolling basis. The carbon and cost return from that subset typically dwarfs the equivalent return from a uniform refresh, because the lowest performers are doing the most damage per Watt.
Three. Model maintenance, not just acquisition.
The full cost of an installed server includes not just its purchase but its maintenance contract over its life. Third party maintenance can be procured at materially lower rates than the vendor's own. AWS estimates that maintenance alone runs at 18 to 22% of upfront server cost. A refresh plan that does not include the maintenance line will under-state the saving from consolidation by a significant margin.
What good looks like
A defensible 5 year refresh plan is, in essence, a series of evidenced decisions. Each decision uses a current measurement of the estate as the starting point. Each decision models candidate replacement configurations against measured behaviour rather than vendor spec sheets. Each decision is sized in cost, energy and carbon. And the whole plan compounds: the savings released from the first wave fund the next, the floor space released by consolidation removes the need for new building, the carbon released against SECR or equivalent reporting holds up against scrutiny.
That is what evidenced procurement looks like at the planning horizon a board cares about.