Table of Contents
Email is one of the oldest and most stubbornly decentralized technologies we have. SMTP doesn’t care who you are or which company you pay. Anyone can run a mail server, and any mail server can talk to any other. Email was built so that no single machine and no single company would be in charge of it.
On Thursday afternoon, European time, a large part of the world found out how little that design matters in practice.
Starting around 16:00 GMT on 12 November, Outlook stopped working. The web app wouldn’t load mailboxes, the desktop client went quiet and the mobile app just spun. Within about an hour Downdetector had reports from 74 countries and territories. Teams and Copilot started failing too. By the evening around 140,000 people had reported problems, and that only counts people who bothered to file a report. As I write this on Friday evening, more than a day later, Microsoft still hasn’t declared the incident resolved.
I don’t want to just pile on Microsoft here, although they’ve earned some of it. Outages happen and every large system fails eventually. What’s worth looking at is how much stopped when this particular system failed, and why so many organizations had nothing to fall back on.
What Actually Happened
The timeline is still being pieced together. Microsoft has promised a post-incident report once things are stable, so some of this may change. Here’s what’s known so far.
The trouble started between 17:00 and 18:00 UTC on Thursday. At first Microsoft said it was looking into problems in the Americas, then admitted users anywhere could be affected. Its early explanation was “elevated load across multiple regions.” It then said this wasn’t a traffic spike: a number of servers had become unexpectedly unavailable, and the remaining ones were struggling with the load.
Later updates got more specific. A process on the affected servers that sets up secure connections had become overwhelmed, and the servers stopped handling requests. Microsoft pointed to settings applied to servers after an update as a possible factor, and began rolling back a recent network configuration change whose timing matches the start of the outage. It has not said that change caused it, and the root cause remains officially unconfirmed until its post-incident report is out. Australia’s Home Affairs Minister Tony Burke was less careful about it. He said Microsoft had told him it was dealing with “an update that has gone awry,” and compared it to the CrowdStrike disaster of July 2024.
One detail Microsoft repeated was that this wasn’t an attack. Charles Lamanna, Microsoft’s executive vice-president for Copilot, Agents and Platform, wrote on X that “there is no indication this is the result of malicious or cyber activity.” Analysts seem to accept that. Rupert Goodwins told Al Jazeera that elevated load like this usually comes from either an outside attack or a bad internal change, and that he believed Microsoft’s denial.
Recovery has been the painful part. On Friday morning, European time, servers that had been brought back were falling over again. About 16 hours in, the problem had mostly narrowed to Exchange, the backend behind Outlook and nearly all corporate Microsoft email. Microsoft described its progress at one point as “marginal success,” and later as “encouraging recovery progress.” Its service health page said it was “reviewing server performance and confirming that developed patches have been applied.” That’s the kind of sentence an engineering team writes when it’s still on a bridge call at 4 a.m.
The two people Al Jazeera spoke to explained the slow recovery well. Walter Pasquarelli, an AI researcher at Cambridge, compared it to a four-lane motorway narrowing to two lanes. Traffic piles onto the lanes that are left, and those clog up too. Goodwins described the other half of the problem. When a server comes back online, every client that’s been waiting to reconnect hits it at once, and it falls over again. Engineers call this a thundering herd. It’s a well-known failure mode, and it’s hard to manage while millions of Outlook clients keep retrying.
The outage also came two days after November’s Patch Tuesday. Nobody has officially connected the two, and I won’t either. People will wonder, though, and Microsoft’s post-incident report should answer the question directly.
Who Got Hit
This is where it stopped being a tech story.
In the UK, the NHS cancelled some operations because much of its scheduling runs on Microsoft systems, and there were reports of electronic prescription problems. A Canadian healthcare network running several hospitals declared a “code grey,” the alert for an infrastructure failure. English schools had communication problems, and so did the exam regulator Ofqual, the Department for Education and the school leaders’ union ASCL.
Governments were hit across several continents. Japan’s Government Solution Service, the network civil servants use, was disrupted. Singapore’s courts had problems. The Supreme Court of the Philippines warned that “responses to emails and other transactions may be delayed.” Australian federal agencies, including its spy agency and the federal police, reported disruption. In the US, a House of Representatives staffer told Politico the outage had ended “a day of legislative work.” Some law enforcement agencies had trouble processing bail payments, which is more serious than it sounds: people may have stayed in a cell longer because an inbox wouldn’t load.
Businesses lost a full working day of email, or more. Australian experts warned that losses could reach billions of dollars, with small businesses hit hardest. Some companies delayed quarterly financial reporting. The disruption also spread past Microsoft’s own customers: integrations used by OpenAI and Anthropic were affected, because Microsoft infrastructure runs underneath a lot of other services.
The UK’s defence readiness minister, Luke Pollard, called for “greater resilience” against technology failures. It’s a reasonable thing to say, but it doesn’t commit anyone to doing anything.
This Isn’t the First Time This Year
The “rare event” defence gets harder to make each time it’s used.
In January 2026, Microsoft 365 was down across North America for almost ten hours. That time Microsoft blamed heavy load combined with capacity it had taken out for maintenance. In late August, thousands of Outlook users around the world were locked out of their accounts. Last October, an AWS outage took a large part of the web offline. And everyone remembers July 2024, when a faulty CrowdStrike update crashed Windows machines around the world, grounded flights and caused billions in losses.
Each incident had its own technical cause. A config change, a capacity miscalculation, a bad driver update. Look at them separately and each one seems like bad luck. Together they show a pattern. A small number of companies now run critical services for a very large share of the world. When one of them makes a mistake, it affects everyone at once, and there’s often no backup in place.
Goodwins also made a less comfortable point. He suggested that experienced engineers leaving Microsoft have drained knowledge of how these systems actually work, and that the company’s huge push into AI may have left unglamorous services like email and Teams underinvested. I can’t verify either claim from the outside. Both are plausible, though, and both fit with how a company behaves when AI announcements get more attention than keeping mail flowing. He also pointed out that Microsoft’s AI couldn’t do anything to fix Microsoft’s own network, which says something about where the money is going.
The Monoculture Problem
Farmers have a word for planting one crop variety across a whole region: monoculture. It’s efficient while things go well. Everything grows on the same schedule and responds to the same treatment, and the economics scale. Then one disease arrives that the variety can’t resist, and you lose the entire harvest.
That’s where office IT is today. For a huge number of organizations, Microsoft provides the email, the calendar, the chat, the video calls, the document storage, the identity system everyone logs in with, and more and more often the AI assistant on top. It’s sold as one integrated bundle, and that’s a big part of the appeal. One contract, one admin console, one vendor to call.
It also means one vendor is a single point of failure. When Exchange went down on Thursday, people didn’t just lose email. Calendar invites stopped arriving. Password reset emails didn’t come through. Approval workflows that send notifications by email stalled. Customer support queues that come in through shared mailboxes stopped. Every process that depended on “we’ll send you an email” stopped too, and those processes run through everything.
The irony is that email itself was designed to avoid this. A federated protocol like SMTP is supposed to survive any single server going down. If my mail provider has a bad day, yours keeps working, and mail to me queues up and gets delivered when my server recovers. The protocol is open, but the market has concentrated anyway. When a big share of the world’s business mail runs on one company’s servers, the openness of the protocol makes little practical difference.
This isn’t really a technical problem. Microsoft’s engineers know how to build redundant systems, and Exchange Online is spread across many data centres in many regions. The trouble is that all those regions run the same software, get the same configuration changes and are managed by the same deployment pipeline. Spreading them around the world protects you from a fire in one building. It doesn’t protect you from one bad update that goes out to every building.
Europe Has Been Here Before
For European readers this outage lands in the middle of a debate that was already heated.
Europe’s dependence on US cloud providers has been an uncomfortable topic for years, and it got sharper in 2025 and 2026 as relations with Washington became less predictable. Tariffs, public threats over Greenland and a general sense that the transatlantic relationship can’t be taken for granted have all moved “digital sovereignty” from think-tank papers onto ministers’ agendas. France’s prime minister, Sébastien Lecornu, said French officials would stop using Zoom and other US video tools in favour of French software. As I wrote back in June, Euro-Office went live as an open-source attempt at a European alternative to Microsoft 365.
Usually the sovereignty argument is about control and legal jurisdiction: who can read your data, which courts can demand it, and whether a foreign government could cut you off. Goodwins mentioned that side too, saying that when a US company hosts your data “you can’t trust that your data is secure.”
This week added a more practical argument: availability. Leave aside surveillance, the CLOUD Act and geopolitics. If a public health system can’t schedule surgery because a configuration change in another country broke its inbox, then the system isn’t resilient, and its owner doesn’t really control it. You don’t need to be anti-American to see that. You only need to look at the list of hospitals and courts that were affected.
Pasquarelli was more cautious. He doesn’t see a credible European alternative to Microsoft’s office software yet, and thinks organizations will spend on “digital resilience” rather than build their own. He may be right about the short term. I’d just point out that you can only be resilient if you have something to switch to, and alternatives don’t appear out of nowhere when an emergency hits.
What “Resilience” Should Actually Mean
“Resilience” is about to appear in a lot of board presentations, so it’s worth saying what it should mean in practice.
It doesn’t mean buying another Microsoft add-on that promises higher availability. If the backup runs on the same platform, gets the same updates and depends on the same identity system, it’ll go down at the same time as the main system.
Real resilience means some degree of diversity. Critical communication needs at least one channel that doesn’t share a failure mode with the main one. A hospital might keep a separate messaging system for clinical coordination that runs on different infrastructure, from a different vendor, ideally on open standards. A government department might keep an emergency mail relay that isn’t Exchange. Most organizations should have a plan B that’s been tested in practice, not just written down in a document that’s stored, of course, in SharePoint.
This is where open standards and open-source software matter, and not out of ideology. When your email runs on standard protocols (SMTP, IMAP, CalDAV) and your data is in open formats, switching to a different provider is a migration you can actually do. You can keep a secondary system that speaks the same language. You can move part of your workload somewhere else without rewriting everything. Lock-in isn’t only a licensing problem. In a crisis, it decides how quickly you can route around a failure.
The open ecosystem already has serious tools for this. Matrix, an open, federated chat protocol, is used by the German armed forces, the French government and a growing number of public bodies, largely because no single server is a point of failure. Nextcloud and the European office suites built around open formats are used in production by real governments. Providers like Proton, Tuta, mailbox.org and many smaller hosting companies offer standards-based email that isn’t Exchange. None of them will replace Microsoft 365 for a 50,000-seat organization by next quarter. They don’t need to. They need to exist, be maintained and be good enough for organizations to diversify, so that one bad update can’t take out the whole office.
To be fair, open source isn’t immune to outages. An open-source project hosted in one place and maintained by an overstretched team can fail as badly as anything Microsoft runs, and plenty have. The difference is structural. With open protocols and open code, you can choose to run things in more than one place, with more than one provider. With a closed, vertically integrated platform you get the vendor’s redundancy and nothing else.
The Accountability Gap
There’s another uncomfortable part of this story, and it’ll come up again when Microsoft’s licence renewals arrive.
Goodwins predicts Microsoft won’t take a short-term revenue hit, because customers are tied into multi-year licensing contracts. He’s almost certainly right. Enterprise 365 agreements usually include service level commitments, and missing them earns customers service credits: a percentage off next month’s bill. For a hospital that cancelled operations or a court that couldn’t process cases, a discount on a subscription has nothing to do with what the outage actually cost.
That’s the structural problem in a sentence. The costs of a centralized failure fall on everyone, and the consequences for the vendor are capped by contract. Microsoft won’t pay for the NHS’s rescheduled surgeries or the lost legislative day in Washington. It’ll publish a post-incident report, apply credits where required, and carry on. Its share price may dip briefly, as Pasquarelli expects. A year from now this outage will be a footnote in an article about the next one.
That’s not a criticism of Microsoft’s engineers, who almost certainly had a miserable 36 hours. It’s a criticism of a market where one company can become critical infrastructure for public services worldwide without carrying anything like the accountability we’d expect from a power grid or a water utility. Regulators are starting to notice. The EU’s Digital Operational Resilience Act already requires financial institutions to manage concentration risk in their ICT providers, and it’s fair to ask why hospitals and courts don’t have something similar.
What I’d Like to See Happen Next
Here’s my short wish list, in rough order of how likely I think each item is.
A detailed, honest post-incident report from Microsoft. It should explain what changed, why it went out globally instead of in stages, why the rollback took so long to work, and whether it had anything to do with Patch Tuesday. Microsoft has written good post-mortems before, and this incident deserves one.
Public bodies auditing their single points of failure. Every ministry, health service and court that went dark this week should ask a simple question: if our main provider disappeared for 48 hours, how would we communicate? If the answer is “we’d wait,” that’s a policy decision, and it should be made deliberately.
Procurement that rewards diversity and open standards. Public money buys a huge amount of IT. If tenders favoured interoperable, standards-based solutions and treated vendor concentration as a real risk, the market would change faster than any amount of advocacy could manage.
Continued investment in open alternatives. That includes European suites like Euro-Office, federated tools like Matrix, and the dull but essential work of maintaining open-source software that public infrastructure depends on. Spending on it is spending on insurance.
The Lesson We Keep Not Learning
At some point, probably by Monday, Outlook will be fine again. The backlog will clear, people will skim three days of email, delete most of it and get on with things. The memes will fade. On Tuesday most of the organizations that were hit will be running exactly the same setup they were running on Wednesday.
That’s the part that bothers me most. The outage itself was a technical failure, and those happen. The bigger problem is that we keep building systems where one technical failure can stop hospitals, courts, parliaments and schools on several continents at once, and we keep calling that efficiency.
The open internet was designed so that this wouldn’t happen. Email, the web and the protocols underneath them assume that any single node can fail and the network carries on. We didn’t lose that resilience because the technology stopped working. We gave it up a contract at a time, because one vendor, one bill and one login were easier to manage.
Thursday showed what that convenience costs. Whether we act on it will depend on what organizations do once the backlog is cleared and this isn’t news anymore.
Sources: Wikipedia: 2026 Microsoft outage · Al Jazeera · ABC News (US) · ABC News (Australia) · TechRadar · SecurityWeek · CIO&Leader · The Register (January 2026 outage)
