Migration
A data center migration moves an IT estate from one facility to another — physically, or by rebuilding it at the target and cutting over. For an enterprise estate of 100–300 racks it takes 9 to 18 months and costs roughly $8,000–$20,000 per rack before any hardware refresh. The physical move is under a tenth of that. Everything else is discovery, network lead times, dual-running and the exit.
Migrations fail in discovery and slip in networking. Those two phases carry most of the risk and almost none of the visible progress, which is exactly why they get compressed under schedule pressure. The phase gate matters more than the phase: a phase is not finished when the work looks done, it is finished when the gate is evidenced.
4–8 weeks · Infrastructure architecture
Inventory every asset and — the part teams skip — every dependency between them. Most migrations fail here, not on the loading dock: an application nobody documented turns out to talk to a database that was scheduled for a different wave.
Gate A dependency graph signed off by application owners, with every system assigned to a move group.
6–12 weeks · Infrastructure + procurement
Shortlist facilities against power, density, latency, carrier mix and compliance. Design the target layout: cabinet elevations, power draw per rack, cross-connects, network handoff.
Gate Signed colocation contract with commissioned space, power energised, and cross-connects ordered.
8–16 weeks · Network engineering
Circuits are the long pole. New WAN links, dark fibre or waves between old and new sites, carrier onboarding and IP addressing all run on carrier lead times you do not control — typically 60 to 120 days, longer in emerging markets.
Gate Both sites live on the same L2/L3 fabric, tested at full bandwidth, with the temporary inter-site link in place.
2–4 weeks · Migration lead
Move a small, non-critical, self-contained group first. The pilot exists to test the runbook, the logistics chain and the rollback — not to save time. Every subsequent wave inherits its corrections.
Gate Pilot systems running in the target site for a full business cycle with no rollback and no unplanned outage.
8–26 weeks · Migration lead + application owners
Repeated cutover windows, usually weekends, grouped so that each wave is dependency-complete. Physical moves, replatforming and lift-and-shift-to-new-hardware can run in parallel across waves.
Gate Each wave signed off by its application owner after a post-move validation pass.
4–12 weeks · Facilities + procurement
Certified data destruction, asset disposal, cabinet clearance, cross-connect cancellation and — the item that quietly costs the most — formally exiting the old contract without triggering an evergreen renewal.
Gate Written confirmation from the outgoing provider that the space is returned and billing has stopped.
The instinct is to budget for trucks and hands. In practice labour and project management is the largest line, and the second largest is a decision rather than a cost: how much of the estate you refresh instead of move. Shares below are typical for enterprise-scale programmes — use them to sanity-check a vendor estimate, not as a norm.
| Cost driver | Share | What it covers |
|---|---|---|
| Labour and project management | 30–40% | Internal time plus any migration partner. Dominates every migration budget and is the line most often left out of the first estimate. |
| New hardware and refresh | 20–35% | Equipment replaced rather than moved. Often the largest discretionary lever: a migration is the cheapest moment to retire old kit. |
| Network and circuits | 10–20% | New circuits, the temporary inter-site link, cross-connects and the overlap period where both sites are billed. |
| Dual-running overlap | 10–15% | Paying for both sites simultaneously. Directly proportional to how long the wave programme runs — the strongest argument for a tight schedule. |
| Physical logistics | 5–10% | Packing, specialist transport, insurance, receiving, racking, cabling. |
| Decommission and exit | 3–8% | Certified data destruction, disposal, cabinet clearance, and any early-termination charge. |
Run your own numbers in the migration cost calculator, and check the target market rate against the Colocation Price Index before you sign.
Six risks account for most migration failures we see discussed publicly. None of them are exotic — each has a known control, and each control has to be in place before the phase where the risk lands, not after.
| Risk | If it happens | Control |
|---|---|---|
| Undocumented dependency discovered mid-cutover | Extended outage, forced rollback, business escalation | Two-week traffic capture in discovery; dependency-complete move groups; pilot wave before anything critical |
| Carrier circuit delivered late | Whole programme slips; dual-running cost accrues with no progress | Order circuits the week the contract is signed; contractual delivery dates; a temporary link as a fallback path |
| Hardware fails to power on after transport | Wave misses its window; unplanned procurement at list price | Retire ageing kit instead of moving it; spare parts on site; insurance at replacement value |
| Old contract auto-renews before exit | A full extra term billed for empty space | Notice date tracked from day one of the project, with the reminder owned by procurement, not the migration team |
| Target power density below what the estate actually draws | Stranded racks, emergency redesign, renegotiation from a weak position | Size from measured draw, never nameplate; contract headroom and an expansion clause |
| Licences break on new hardware or a new site | Applications unlicensed and legally unusable after cutover | Licence audit in discovery; re-issue requests raised before the pilot |
The migration plan is downstream of the site decision, and a bad site decision cannot be fixed by good execution. Three things constrain the choice more than price: available power density at the time you need it, the carrier mix on site, and whether the facility holds the certifications your compliance team will ask for. All three are checkable before you talk to sales.
For an enterprise estate of 100–300 racks, 9 to 18 months end to end. Discovery takes 4–8 weeks, target selection and contracting 6–12 weeks, network build 8–16 weeks (carrier lead times dominate), then production waves over 2–6 months and decommissioning for another 1–3 months. Programmes that claim six months usually mean the wave phase only.
Budget $8,000–$20,000 per rack for labour, logistics, network, dual-running and decommissioning. Hardware refresh sits on top of that and can double or triple the programme cost, which is why it is treated as a separate decision rather than a migration line item. Within the migration itself, labour and project management is the largest share and the physical move is under 10%.
An undocumented dependency found during cutover. It is the most common cause of extended outages and forced rollbacks, and it is why move groups must be dependency-complete and why a pilot wave exists. Late carrier circuits are the most common cause of schedule slip, which is a budget problem rather than an outage.
A migration is the cheapest moment to retire ageing equipment: you are already paying for the labour, the transport risk applies only to kit you actually move, and hardware near end of warranty is the most likely to fail on power-up at the far end. As a rule, anything within 18 months of warranty expiry is a refresh candidate rather than a move candidate.
The week the target contract is signed. Carrier delivery runs 60–120 days in mature markets and longer in emerging ones, and it is the only part of the plan you cannot compress by adding people. Every other workstream can be resequenced around a late circuit; nothing can be resequenced around a missing one.
Yes, for the duration of the wave programme — there is no realistic migration where the old site goes dark the day the new one lights up. Dual-running is typically 10–15% of migration cost and is directly proportional to programme length, which is the strongest financial argument for a compressed wave schedule.
Tell us the rack count, the target market and the deadline. We will come back with facilities that match on power and compliance, plus benchmark pricing so you know what a fair quote looks like.
