Designing Resilient Software Around Nigeria's Power and Connectivity Realities | Blog

# Designing Resilient Software Around Nigeria's Power and Connectivity Realities

TL;DR: In Nigeria, the grid can drop to zero megawatts nationwide with no warning and broadband still reaches barely more than half the population, so software built for Lagos, Abuja, or Kano has to treat power and connectivity failure as a normal operating condition, not an edge case — that means offline-first UX, aggressive queuing, and infrastructure architected for interruption from day one.

Why "just add a generator" isn't a software strategy

Most engineering teams outside Nigeria treat power and network outages as rare disaster-recovery scenarios. In Nigeria, they're Tuesday. The national grid collapsed twice in the space of five days in January 2026 — on January 23 and again on January 27 — with generation falling to zero megawatts and all 11 distribution companies (Ikeja, Eko, Abuja, Kano, Port Harcourt, and others) recording no allocation each time (Nairametrics, Guardian, ICIR). The Lagos Chamber of Commerce and Industry has warned that without structural reform, Nigeria could see tens of grid collapses across the year (ICIR).

The market response tells you everything about how normalized this is: 70.7% of Nigerian firms now own or share a generator, and outage losses come to about 3% of annual sales (AfDB, via Legit.ng). Nigerians spent ₦16.5 trillion on diesel, petrol and generators in 2023 — against roughly ₦1 trillion in revenue for the entire formal power sector — on the power minister's own figures (Businessday), while the Energy Commission of Nigeria puts yearly spend on fuelling off-grid generation at around $22 billion (The Sun). Energy swallows as much as 40% of factory operating costs (MAN, via Naija247news).

That's the physical-world coping mechanism. Software needs its own equivalent — and a generator under a server rack doesn't fix a mobile app that freezes when a user's phone loses signal mid-transaction on a bus in Ibadan.

Connectivity is improving, but "average" hides the real problem

Nigeria's telecom numbers look genuinely good in aggregate. Active telecom subscriptions hit 192.23 million by June 2026, broadband subscriptions reached 123.11 million, and broadband penetration climbed from 53.07% in January to 56.79% by mid-year (NCC data, via Tribune and Tekedia). 4G still dominates mobile connections at 54.31%, while 5G has only inched up to 4.61% — 2G and 3G together still carry more than 41% of connections (Techeconomy).

But "56% broadband penetration" means close to half the country is still on 2G/3G or intermittent connections, and even "connected" users experience wide swings in quality — congested cell towers during peak hours, rain fade on satellite links, and periodic submarine cable damage that can take down large chunks of West African connectivity at once. A single incident in March 2024 — cable damage near Abidjan affecting the West Africa Cable System (WACS), ACE, MainOne, and SAT3 cables — knocked out connectivity across 13 countries including Nigeria, disrupted bank services, and carried an estimated $8 million bill to fully restore, roughly $2 million per cable on WIOCC's own estimate (Internet Society, PressReader/ThisDay).

Starlink has become a meaningful alternative for businesses that can afford it — the NCC licensed it in May 2022 — Nigeria and Mozambique were the first African approvals, granted together — and Nigeria became the first African country where the service actually went live, in early 2023, and 2026 pricing runs from roughly ₦57,000/month (residential) up to ₦159,000/month for the priority business tier, on top of a ₦669,000 hardware kit (Legit.ng, TalkTalk Nigeria). That's real redundancy for an office or a branch location, but it's not something you can assume your end users have in their pockets.

What this means for how you actually build

The practical implication is that resilience has to be designed in at three layers: the client, the application/data layer, and the hosting infrastructure.

1. Client-side: assume the network will disappear mid-session

Offline-first, not offline-tolerant. Local-first data models (writes go to local storage first, sync happens opportunistically) prevent a dropped connection from destroying a user's form entry, cart, or in-progress transaction.
Idempotent, queued writes. Every mutating action — a payment, an order, a status update — should be safe to retry. Nigerian fintech apps in particular need to handle "did my transfer go through?" gracefully, since a network drop between request and confirmation is common enough to be a designed-for state, not a bug report.
Aggressive payload discipline. Compress images, lazy-load non-critical assets, and design for 3G-equivalent throughput as the baseline experience, not the degraded one.
Local caching with clear staleness signals. Show users what they're looking at is cached/offline data rather than silently failing or showing a blank screen.

2. Application layer: decouple availability from any single dependency

Message queues over synchronous calls for anything crossing a third-party boundary — an NIBSS Instant Payment transfer, a Paystack or Flutterwave charge, a bulk-SMS gateway. When the March 2024 cable cut took bank services down with it, the systems that came through were the ones that had already accepted the request and could settle it once the link returned, rather than holding a connection open against a processor they could not reach.
Circuit breakers and graceful degradation. If a non-critical service (recommendations, analytics, image processing) is unreachable, the core transaction should still complete. On a network where a tower congests every evening, the difference between a degraded feature and a failed checkout is a timeout budget someone chose deliberately.
Event sourcing or append-only logs for critical business state, so a mid-write power cut on either end doesn't leave a record in an ambiguous state — with the grid dropping to zero megawatts twice in five days, that is a routine event to design for, not an edge case.
A low-bandwidth path for every critical journey. With 2G and 3G still carrying more than 41% of connections, the checkout, the balance check and the delivery update each need a route that works without the full app payload: USSD, an SMS confirmation, or a stripped-back HTML view.

3. Infrastructure: get closer to users, and build for interruption at the hosting layer

Neither AWS nor Azure operates a full region inside Nigeria — the nearest AWS region is Africa (Cape Town), and Azure's nearest is South Africa North (Johannesburg), with cross-region latency from Lagos commonly reported in the 95–140ms range — a practitioner measurement rather than a published provider figure, so treat it as indicative (Launchverse). AWS has, however, operated a Local Zone directly in Lagos since January 2023 — the first of its kind in Africa — letting workloads that need single-digit-millisecond latency or local data processing run physically closer to Nigerian users while still using standard AWS tooling, linked back to the Cape Town parent region (AWS, DataCenterDynamics). That's a meaningful architectural option most teams haven't factored into their region-selection decisions yet.

Beyond region choice, the infrastructure checklist for Nigeria-facing products should include:

Layer	Practice	Why it matters here
Compute	Multi-AZ / multi-region failover, health-checked auto-recovery	Local outages (power, ISP) shouldn't require a human to notice and restart anything
CDN/Edge	Cache static and semi-static content at the edge	Reduces round trips over congested or degraded local last-mile links
Database	Automated backups + point-in-time recovery, tested restores	Abrupt power loss at a client site or self-hosted server is a real failure mode, not hypothetical
Monitoring	Alerting tuned to distinguish "our service is down" from "the user's network/grid is down"	Avoids false-alarm fatigue and misdirected incident response
Deployment	Zero-downtime deploys, feature flags, rollback automation	Bad deploy timing compounded by a grid collapse is a genuinely bad day

This is precisely the discipline behind proper cloud and DevOps engineering — CI/CD pipelines, infrastructure-as-code, monitoring, and failover design aren't nice-to-haves for a Nigerian-market product; they're the difference between an app that "works in the demo" and one that holds up when the grid drops mid-transaction on a Tuesday afternoon.

FAQ

Does my app really need to support offline mode if most users have smartphones?

Having a smartphone doesn't guarantee a stable connection. With broadband penetration around 57% and frequent grid collapses disrupting cell towers and home routers alike, treating connectivity as intermittent — rather than binary online/offline — produces a noticeably better experience even for users who are "usually" online.

Is hosting in Cape Town or Johannesburg good enough for a Nigerian audience?

It's workable for most business applications, but latency of 95–140ms round-trip adds up for interactive or real-time features. For latency-sensitive workloads, AWS's Lagos Local Zone (linked to the Cape Town region) or a CDN edge layer in front of whichever region you use will noticeably improve responsiveness.

How should fintech apps specifically handle power/connectivity interruptions?

Treat every payment or transfer as needing idempotency keys and asynchronous status confirmation rather than assuming a synchronous request-response will always complete cleanly. Given how common mid-transaction network drops are, "is my money safe?" needs to be answerable from queued, auditable state — not from whether the user's screen showed a success message.

What's the single highest-leverage change a team can make first?

Move from synchronous, request-blocking architecture to a queue-backed, retry-safe one for anything that writes data or calls a third-party service. It's a smaller lift than a full offline-first rewrite and immediately reduces the blast radius of both grid and network interruptions.