Most multi-site businesses never choose where their systems live. They inherit it.
The usual inheritance is a server room at head office, with every new system added to it because that is where the last one went. The usual overcorrection is to move everything to a data centre or the cloud, and then find that a branch in Port Harcourt cannot raise an invoice while its one link is down. Neither is a decision about co-location, on-premise or edge. Both are defaults.
The decision worth making is smaller: where each system should sit. Not where the company sits. A branch is a set of workloads with different needs, and one branch can reasonably use all three models at once.
What each model actually buys you
On-premise is equipment in a room you run: a head-office server room or a comms cabinet in a branch. You own the hardware and everything around it, from the UPS and the generator changeover to the air conditioning and the lock on the door. Control is total. So is the bill for every battery and every litre of diesel.
Co-location is rented rack space, power, cooling and connectivity in someone else's data centre. The servers are still yours, and so is managing them. What you pay for is the room: power and cooling built to run all day, every day, and a choice of carriers in the building. A branch server room in Nigeria finds exactly those things hardest to provide.
Edge is a small amount of compute at the site where the work happens. It processes locally and syncs to the centre when it can. Its job is to keep a branch working through a WAN outage.
One local fact shapes the co-location option before anything else. As of 2026, most of Nigeria's commercial co-location capacity is in Lagos, close to the submarine cable landings. Abuja has a small number of facilities, and much of that capacity serves government hosting. Outside those two cities the choice is thin. For a business with sites in Abuja, Lagos and Port Harcourt, co-location usually means a rack in Lagos, reached from everywhere else over the WAN. That makes your links part of the co-location decision, not a separate one.
Five questions to ask of each system
Take each significant system in turn. Skip the weighted scoring sheet. Ask these questions and write the answers down.
What happens when the link goes down? This one decides more placements than the other four together. A finance ledger used from head office can live in a Lagos data centre and tolerate the round trip. A branch till or a clinic's patient check-in cannot tolerate the link being gone. If staff stop work when the circuit drops, that function needs a local copy at the site, whatever happens to the data afterwards. Many branches run on a single provider, so plan for the outage as a matter of when.
How good is the power where it would sit? A branch server room is as available as its generator, its fuel and its UPS runtime. If the site has one generator with a manual changeover, a system that must stay up is safer in a facility built for continuous running. An edge box still has to ride through the changeovers at its own site, so it needs a UPS sized for that gap and a clean shutdown when the batteries run out. Our piece on disaster recovery when the main risk is power covers how to set those recovery targets.
Where is the data allowed to go? The Nigeria Data Protection Act 2023 does not require personal data to be stored in Nigeria. It controls transfers out of the country. Section 41 allows a transfer where the recipient is subject to a law, binding corporate rules, contractual clauses, a code of conduct or a certification mechanism that gives adequate protection, or where one of the conditions in section 43 applies. So a Lagos rack and a cloud region abroad are not equivalent. The first keeps the data in Nigeria. The second is a cross-border transfer you have to justify and document. Some sectors add their own hosting rules on top. Public bodies and CBN-regulated institutions should check their regulator's position before moving anything.
Who will look after it? On-premise gives you the most control and asks for local skill at every site that has it. Co-location removes the building work and leaves the systems with you. Be honest about the team you have. A server in a branch that nobody there can restart is a liability.
What does it cost, and in which currency? Server hardware is imported and priced in dollars, so expanding on-premise puts a capital purchase at the mercy of the exchange rate on the day you order. Co-location turns the room into a monthly fee, but that fee also tracks diesel and the exchange rate, so do not treat it as fixed. Edge looks cheap per site and adds up across twenty sites. The server room cost breakdown shows what sits behind the on-premise number.
If an hour without a system stops the business, that fact decides where it lives. Cost gets a vote after that, not before.
What the answers usually look like
Patterns come out quickly. Tills and local sign-in go to the edge, as does anything clinical or on a production line. They run at the site and sync when the link returns. The finance system and consolidated reporting are strong co-location candidates, because they need the power and the carriers more than they need to be near anyone. Systems tied to equipment on site, or that the team has to touch often, stay on-premise, provided you accept the room that comes with them.
Now argue yourself out of the expensive answer where it is not needed. A ten-person branch that works in a web application and email may need no local server at all. Two links from different providers and a UPS on the network kit will do more for it than an edge box. Local compute earns its place only where there is work that must continue offline.
The output is a placement list: every significant system, where it sits, and one line saying why. That list is the thing to keep. It turns an argument about architecture into a decision someone else can check. The network that joins those places is covered in LAN and WAN design for multi-branch businesses.
Hybrid, done in order
For most multi-site businesses the answer is a mix. The risk in a mix is not the design. It is the sprawl: three models, each with its own patching and backups, and gaps between them that nobody owns.
So do it in order. Start with the systems whose current home is clearly wrong. Usually that is a critical system in a single-generator branch server room, or one that stops completely whenever the WAN drops. Build one standard edge configuration and give every site the same one. A different build at every branch is how a hybrid estate becomes unsupportable.
Keep one view of monitoring and backups across all three places. A mixed estate you cannot see is more fragile than the single server room it replaced. That is the multi-site support model applied to placement.
Then revisit the list once a year. Links improve and sites open and close. The list should describe the business you have now, not the one you had when it was drawn.
Frequently asked questions
Is co-location more expensive than on-premise in Nigeria?
Compared line for line, often not. On-premise hides its running costs: diesel, UPS batteries, air-conditioning servicing, security and staff time. Co-location puts the room into a monthly fee and gives you redundant power and carriers that a single site rarely matches. Price the uptime you need, not the rack rental on its own. We do not publish a naira figure, because the fee depends on the power you draw and the exchange rate, and a number written today would not hold.
Where does public cloud fit?
It is a fourth place to put a workload, and the same five questions apply. It suits central systems whose load changes. It does nothing for a branch when the link is down. Putting personal data in a region outside Nigeria is a cross-border transfer under the NDPA that you have to justify. Price the cost of moving data back out before you commit a large system to it.
Does the NDPA require us to keep data in Nigeria?
No. The Act regulates transfers out of Nigeria rather than banning them. A transfer needs a basis under sections 41 to 43, such as adequate protection at the recipient or one of the listed conditions, and you need to be able to show which one you relied on. Sector rules can be stricter, so check your regulator. Our guide to NDPR compliance for IT teams covers the rest of the record-keeping.
Can one branch use more than one model?
Yes, and most should. A typical branch runs its offline-critical work on a small local system and uses central co-location or cloud for everything else. You place systems, not sites.
Where to start
List every system you run across your sites. Next to each, write what happens when that site's link drops. That one column will show you most of the wrong placements. Then send us the list through infrastructure solutions. We will come back with where we would put each item and why, including the ones we think should stay exactly where they are.
