An update that sat quietly on Microsoft’s servers for more than three months met a Teams failover and a lost slice of server capacity on 12 November. By Friday, Equity Bank, Stanbic Bank Kenya and Standard Chartered Kenya were telling customers their email was down and pointing them to WhatsApp and phone lines. The City of Cape Town and the South African Revenue Service warned that their own services were disrupted. The mass tech outages of the past week showed how much of Africa’s economy runs on infrastructure that African institutions neither own nor can inspect.
Microsoft was one of five failures in six days. A Japanese supercomputer, a US bank, a web-testing vendor and a Norwegian energy giant followed it into the headlines. None of the companies involved has linked these incidents to each other. Microsoft traced its outage to a software defect, and Chase said its fault was not cyber-related. The common thread is not a plot. It is dependence, and Africa has less room to absorb it than anyone.
What Failed at Microsoft, and Why It Went Global
Exchange Online and the services that lean on it stopped working worldwide from 16:20 UTC on 12 November until 16:10 UTC on 13 November, almost 24 hours. Microsoft’s preliminary post-incident report, dated 15 November, names three factors. A latent defect in a Windows fix built for Microsoft 365’s networking gateway servers exhausted processor and memory under certain loads. About 15 hours earlier, Teams had failed over onto the primary gateway that also serves Exchange Online, which raised the load. And nearly 14% of gateway capacity was already out of service because of an unrelated problem the day before.
The cascade did the rest. Microsoft’s traffic systems moved users to healthy servers, which pushed the same load onto fewer machines. At the lowest point, more than 80% of gateway servers were unresponsive. The report says normal bulkheads between regions failed to contain the spread, so a regional fault became a global one.
Two details matter for African customers. Teams, OneDrive, SharePoint and Office apps could fail out of the broken gateway, which is why they recovered first. Exchange Online could not, for scale reasons. Microsoft also said its automated recovery tools were not built for a fleet-wide collapse, and that its usual rollback and feature flags could not quickly undo a change that had shipped months earlier. A final report is due on 20 November.
Four More Failures in Six Days
The other incidents had different causes, which is the point. On 16 November, JPMorgan Chase said a software bug in on-premise hardware systems slowed transactions. Customers of other banks reported delays on payments involving Chase, and Zelle said its network was not down. Early claims that several banks had failed at once came from user reports, not from the banks.
Attackers had a different route into Fugaku, Japan’s flagship supercomputer. In its 16 November notice, RIKEN said an outsider reached internal servers between 10 and 13 November through the public dashboard that displays Fugaku’s operating status. The institute suspended operations on 14 November. Up to 4,222 Fugaku users and 375 users of a second system may be affected, and RIKEN said SSH private keys and access keys stored there may have been taken. It asked users to revoke those credentials everywhere they unlock access.
Then came the vendor problem. On 16 November, attackers hijacked scripts hosted by the web-experimentation provider Optimizely and served a crypto-address swapper and card skimmer through them. StepSecurity’s analysis describes the payload, and researchers at Report URI found infected scripts at 178 organisations. On 17 November, Equinor said it had detected an ongoing cyberattack on some IT systems, that production was unaffected, and that it had found no indication of data theft.
Why Africa Takes the Hardest Hit
African firms buy most of their productivity software, cloud capacity and payment tooling from a handful of foreign providers. When one of them fails, the failure arrives in a region where fallback options are thin. The Kenyan banks’ answer, which was to route customers onto WhatsApp, swapped one foreign platform for another. It worked as a patch, not as a plan.
Local hosting does not solve this alone. Microsoft’s report describes a global gateway failure, so a customer with data in a South African region still lost access to mail. Region choice protects against a building failing. It does not protect against a provider’s global control plane failing. Nigerian lenders also reported email problems, though TechMoonshot could not confirm which institutions on the record.
Infrastructure is improving, and TechMoonshot has tracked Africa’s data centre build-out, including operators such as Rack Centre that connect directly to all eight undersea cables serving the West African coast. Better facilities and cables reduce one class of risk. They do not reduce the risk of depending on a single vendor’s software.
Start With a Map of What You Cannot Do Without
Most organisations cannot name their real dependencies. They know they use Microsoft 365, but not that password resets, payment approvals, supplier invoices and incident alerts all travel through the same tenant. Draw the map by function, not by vendor. For each critical service, write down which email, identity, DNS, payment and messaging providers it needs, and what happens in the first hour if each disappears.
The German hotel that reportedly lost a chlorine alarm because the alert went out by email is the extreme version of this mistake. Every institution has a version of it. Banks that route fraud alerts, hospitals that route lab results and utilities that route fault reports through one mailbox have built a single point of failure and called it a workflow.
Build a Second Way to Talk and a Second Way to Log In
The ordinary advice is to keep a backup contact list. The better advice is to build a communications path that shares nothing with your primary one. That means a different provider, a different identity system and a different network where possible. Give incident teams an out-of-band channel, such as a satellite phone, a mobile broadband device on a different carrier, or a messaging group on a separate platform, and rehearse using it.
Do the same for identity. Keep emergency administrator accounts that do not depend on your main directory, store their credentials offline, and test them quarterly. Route safety and security alerts through at least two independent paths, so a failed mail gateway cannot silence an alarm. Prepare paper or offline fallbacks for the five processes your customers notice first, such as account queries, payments and appointments.
Keep Backups Where Missiles, Bugs and Billing Disputes Cannot Reach Them
AWS gave the clearest lesson of the year. After drone strikes damaged facilities in the UAE and Bahrain in March, AWS said in September that it could not restart several sites. It said most UAE customers had restored operations in other regions from backups or copied data. For Bahrain, it said it had exhausted every option for restoring data and resources that had not been migrated before the region went down. Customers who had copies elsewhere recovered. Customers who did not lost data.
Beyond the ordinary rule of three copies, copy critical data to a different provider and a different jurisdiction, keep at least one immutable copy that an attacker with admin rights cannot alter, and store it in formats another platform can read. Then time a full restore. A backup you have never restored is a hope, not a control. Budget standby capacity or a pre-negotiated contract with a second provider, because spinning up a replacement during a regional crisis is when everyone else is doing the same.
Treat Every Vendor Script as Code You Own
The Optimizely incident worked because websites load a vendor’s script at runtime and trust it with the whole page. Inventory every third-party script on your payment, login and customer-account pages, and remove what you cannot justify. Use a Content Security Policy to restrict where scripts can load from and send data to. Where a vendor file never changes, pin it to a fixed hash, though experimentation and analytics tools often change constantly.
Use hosted payment fields that isolate card entry from your own page code, so a hijacked script cannot read card numbers. Monitor the scripts your pages actually serve to real visitors, not just what your developers deployed. For African fintechs and merchants, whose customers often transact on mobile browsers with little protection, this is cheap insurance with a large payoff.
Shrink What the Internet Can See
Fugaku’s breach began at a status page, the kind of dashboard teams publish for convenience and forget. Audit everything your organisation exposes to the internet, including status pages, admin panels and test environments, and put anything non-essential behind authentication. Never store long-lived private keys on shared systems where one intrusion exposes them. Prepare a playbook for mass credential revocation, because RIKEN’s instruction to users to revoke keys at every connected service is a task most organisations have never rehearsed.
Slow Your Own Releases Down
Chase blamed a software bug and Microsoft blamed a change that shipped in July. Your own releases carry the same risk at smaller scale. Roll changes out to a small slice of users first, keep a tested rollback for every release, and test on the newest and largest hardware you run, not only the cheapest. Microsoft said its load testing favoured weaker hardware, and the defect worsened on machines with more cores. Limit the blast radius so one failing component cannot take down the others, and rehearse recovery from a total collapse, not only from a partial one.
What Governments and Regulators Should Demand
Regulators cannot protect what they cannot see. South Africa’s Financial Sector Conduct Authority and Prudential Authority published Joint Communication 5 of 2026 on 1 September, setting a notification template for material IT and cyber incidents. Other African regulators should adopt comparable templates, then use the data to map which vendors sit under which critical services. Concentration risk is invisible until someone aggregates the reports.
Procurement is the other lever. Government contracts with cloud and software vendors should require exit plans, tested data portability, post-incident reports on a fixed timetable and a named fallback for every critical service. Microsoft promised a final report within five business days of resolution, and that is a reasonable benchmark to write into contracts. Nigeria’s sovereign cloud taskforce under NITDA is the obvious place to start, because sovereignty without a failover plan only moves the single point of failure inside the border. The same logic applies as Nigeria builds its identity infrastructure, where an outage becomes a national service failure.
Governments should also run joint exercises that simulate a vendor-wide failure across banking, health, tax and courts at once. The week’s incidents showed that sectors fail together, so rehearsing them separately proves little.
What to Do in the Next 90 Days
Mass tech outages will keep happening, and no contract will prevent them. What separates a bad day from a crisis is whether an organisation has rehearsed running without its most trusted tool. In the next 90 days, map your dependencies, stand up a second communications and identity path, copy critical data to a second provider and test the restore, and audit every script and public endpoint you expose.
The first test of those plans will arrive without warning, probably at the start of a working day. Organisations that treat the week of 12 November as someone else’s problem will learn how much they depend on a stranger’s software at the worst possible time.