Server rooms and datacenters: mastering critical technical constraints
4 technical trade-offs that structure the 10-year TCO
Targeting Tier III on a 20-rack room generates a significant capex overhead for an SLA that is rarely used: a substantial share of French commercial server rooms are over-specified. On datacenter and server room projects delivered by Kytom since 2022, the SLA actually required by the business is lower than the contracted SLA in most cases. A server room is sized around 4 structuring trade-offs: electrical redundancy (N+1 or 2N), cooling architecture, energy density (3 to 8 kW/rack) and modular scalability. These choices determine most of the 10-year TCO, weighing on both the initial capex and the cumulative opex. Kytom’s 5-phase design and build methodology, spanning 4 to 8 months, frames budgets by synchronising thermal audit, MEP design and progressive commissioning.
Designing a server room requires four structuring trade-offs, to be locked in before any detailed costing.
- Electrical redundancy: N+1 (one backup UPS for N active units) or 2N (full dual chain). 2N doubles the electrical investment but secures a maximum uptime level, suited to high-criticality environments.
- Cooling architecture: direct or indirect free cooling depending on latitude and air quality, direct expansion (DX) air conditioning or chilled water. In the temperate French climate, indirect free cooling significantly reduces the air conditioning’s annual energy consumption.
- Energy density: 3 to 5 kW per rack for extended office use, 6 to 8 kW for virtualised workloads, and beyond for intensive computing (ASHRAE TC 9.9 recommendations, Thermal Guidelines for Data Processing Environments, 5th edition). This density drives the choice between contained hot/cold aisles and liquid cooling.
- Scalability: a modular architecture with progressive commissioning avoids initial over-sizing and smooths opex over time.
Kytom’s position, contrarian on the Tier. Industry doxa recommends Tier III by default whenever production is involved. Our reading differs: the business SLA genuinely justified by RTO/RPO constraints rarely exceeds 99.9% in non-sovereign commercial environments. A well-designed Tier II (N+1, shared generator, mechanical bypass) covers the majority of non-sovereign commercial uses for a capex significantly lower than a 2N architecture.
When these trade-offs do not apply. Below 30 m² of usable space or 4 racks, 2N logic and indirect free cooling are generally no longer cost-effective: a simple N+1 architecture with redundant DX air conditioning is sufficient in most cases.
For the CFO and the Asset Manager: decode the hourly cost of downtime before costing redundancy
The technical reflex is to choose redundancy before calculating what an outage costs. The reverse is what secures the capex trade-off.
The question preceding MEP costing is a business one: how much does one hour of room unavailability cost for the processes it hosts? Three typical cases emerge depending on the criticality of the hosted uses: cold backups, test environments and non-critical EDM present a generally low cost of downtime, compatible with a Tier II.
- ERP, email, virtualised workstations: hourly cost of downtime between €2,000 and €8,000 (depending on the headcount affected). N+1 with indirect free cooling balances capex and continuity.
- Payments, trading floor, real-time production: hourly cost exceeding €20,000. 2N becomes rational again, even mandatory depending on the operational risk mapping.
For the Asset Manager. An over-specified server room (Tier III for Tier II use) ties up €800 to €1,500/m² of non-productive capex over the depreciation period. On a 200 m² room, the gap represents €160,000 to €300,000 of asset value tied up with no offsetting rent or used SLA. The tertiary decree adds an opex challenge: an over-sized room consumes more energy than necessary in the early years, which degrades the mandatory reduction trajectory (-40% by 2030 vs reference baseline).
For the CFO. The capex/opex trade-off plays out before the MEP phase. The phase 1 technical scoping note incorporates an hourly cost of downtime × probability matrix, jointly signed by IT and finance management, before any redundancy costing.
3 recurring mistakes observed on recent projects
Three technical mistakes regularly compromise server room projects delivered in France.
The safeguard relies on two systematic deliverables: a 3D thermal audit (CFD) validated before MEP design, and a modular architecture allowing commissioning in steps of 25% of the target capacity. The prior electrical audit checks compliance with NF C 15-100 and flags anomalies before works.
Limit of the method. The CFD audit is not justified below 25 kW of installed IT load: the study cost absorbs the potential gain. On these small configurations, a tabulated thermal calculation and a probe-based check after commissioning are sufficient.
design and build methodology: 5 phases locked by deliverables
The design and build methodology applied by Kytom structures the project into five phases locked by validated deliverables.
- Audit and SLA: energy audit of the existing room, definition of the target uptime (99.9% or 99.99%), business RTO and RPO. Output: technical scoping note jointly signed by IT and finance management.
- 3D thermal modelling: CFD simulation of air flows, identification of hot spots, validation of
