Three-Year TCO of a Dedicated Server: Hardware Refresh, Power and the Depreciation Line

A dedicated server is rarely cheap at the point of purchase and rarely expensive at the point of exit — the real cost lives in the three years between. Most buyers compare monthly rental figures and stop there, which flatters the cheapest box and hides the fact that power, replacement parts, and refresh timing dominate the true cost of ownership. This model breaks a mid-range dedicated server into its four cost layers and spreads them across 36 months so the comparison is honest.

The numbers below use a representative 1U single-socket platform: 16-core Xeon or EPYC, 128 GB DDR4 ECC, 2×1.92 TB NVMe, dual 10 GbE, rented from a provider who includes a /29 and 10 TB of transfer. Adjust the inputs to your own quotes; the structure and ratios hold across hardware generations.

The Four Cost Layers of a Rented Dedicated Server

Every dedicated server bill decomposes into four layers: the base rental (hardware amortisation plus provider margin), power draw, bandwidth and port allocation, and the operating layer you pay for in labour or add-on fees. A $180/month quote and a $240/month quote for the same CPU are almost never the same machine — the difference is usually drive class, power allowance, or whether remote hands is metered.

Separate fixed from variable before you compare anything. Rental is fixed and predictable. Power is quasi-variable because a loaded server pulls more watts than an idle one. Remote hands and replacement drives are event-driven. Budget the three separately and you will catch the quotes that look cheap because they priced 1 TB of transfer and 200 W of headroom.

Building a 36-Month Depreciation Schedule

Straight-line depreciation is the right mental model here even though you do not own the box. The provider is recovering $4,000–$6,000 of hardware over 36–48 months and pricing your rent accordingly. When you add a second NVMe drive or bump RAM at month 14, you are extending a new depreciation curve, not resetting the old one — which is why upgrades mid-term rarely pay back.

The practical rule: any hardware change costing more than three months of rent should be evaluated against the full remaining term, not the next invoice. A $300 RAM upgrade with 24 months left costs $12.50/month of remaining life, which is fine. The same upgrade with four months left costs $75/month, which is not. If you are considering a platform refresh, the dedicated server hosting plans and hardware tiers page lays out current generations with their power envelopes.

Three-Year Total Cost Table

Line itemMonthly36-month totalShare
Base rental (hardware + margin)$180$6,48058.6%
Power (avg 320 W @ $0.12/kWh)$28$1,0089.1%
Bandwidth / port allocation$25$9008.1%
Replacement NVMe (1 drive, year 2)—$3403.1%
Remote hands (4 hrs @ $60)—$2402.2%
Control panel / OS licensing$15$5404.9%
Backup storage (2 TB offsite)$20$7206.5%
Setup / provisioning fee—$1501.4%
Contingency (10% of rental)$18$6485.9%
Total$286$11,026100%

Read the table as ratios rather than absolute dollars. Rental is still the largest line at roughly 59%, but the operating layer — power, bandwidth, licensing, backups, contingency — adds another 35% on top of the sticker price. That is the number to carry into a managed-versus-unmanaged decision: if a managed plan adds $90/month, you are paying $3,240 over three years for something the schedule above values at roughly $1,300 in labour and licensing.

Where Power Quietly Beats Hardware Spec

Power scales almost linearly with CPU utilisation until you hit the platform’s boost ceiling, then plateaus. A 16-core box idling at 90 W and loading to 380 W costs you $18/month more at full tilt than at idle. Over 36 months that is $648 — more than a drive replacement, more than a year of licensing.

This is why two providers quoting the “same” CPU can differ by $60/month. One sells you a 400 W power allowance; the other caps you at 250 W and overages bill at $0.15/kWh. Ask for the watt allowance in writing, and check whether it is per-socket or per-rack-unit. The difference shows up only under load, which is exactly when you cannot afford a surprise.

Hardware Refresh Timing and Residual Value

Most dedicated platforms reach their economic end around month 40–48, when replacement parts become scarce and the CPU falls behind two generations. Providers know this and price the tail of the term accordingly. If your workload is CPU-bound and single-thread sensitive, plan the refresh at month 36 rather than squeezing the last twelve months out of a depreciated box.

Residual value is not yours to capture on a rental, but it is yours to avoid paying for. Never accept a migration fee on a same-provider refresh inside the same chassis, and negotiate the setup fee away on any refresh — the provider is re-using the case, PSU, and network drop, so the marginal provisioning cost is close to zero.

Break-Even Maths Against Cloud and Colocation

At the $286/month all-in figure above, a dedicated server buys roughly 16 vCPU, 128 GB RAM, and 4 TB of NVMe. Equivalent cloud compute with sustained-usage discounts runs $450–$600/month for the same resources, so dedicated wins on raw spec at steady load. Colocation only wins if you already own hardware with more than 18 months of useful life left.

The cross-over point is spikiness; if you are comparing tiers, the feature breakdown of dedicated server packages shows which ones bundle the power and bandwidth that dominate the tail of the schedule. The cross-over point is spikiness. If your load profile is 70% idle and 30% saturated, cloud’s elasticity can beat dedicated on cost — you are paying for capacity you do not use. Anything above 50% average CPU utilisation for 20 consecutive days moves the maths decisively toward dedicated.

Putting the Model to Work

Build the schedule before you sign. List rental, power allowance, bandwidth, licensing, backups, and a 10% contingency. Multiply by 36 and compare that total — not the monthly figure — against managed alternatives and cloud equivalents. Then check the refresh clause: what happens at month 24 and month 36.

The single most useful output of this exercise is the contingency line. A dedicated server with no contingency budget is a server that gets fixed late. Ten percent of rental, held aside, covers a drive replacement, a remote-hands visit, and one emergency RAM stick without touching your operating budget.

Leave a Reply