Microsoft has said it is "deeply sorry" for the outage to its products, including Outlook and Teams, that caused disruption around the world.
Services were broadly recovering by Friday afternoon, nearly 24 hours after the problems began. The company said it was making "encouraging progress" but added that it does not know when services will fully recover.
The cause it has named
Microsoft said the outage was a technical issue introduced during a "network capacity upgrade", and that it was not related to a cyber-attack or to artificial intelligence agents.
Both denials are worth noting, and the second one is new. Ruling out AI agents is not a sentence vendors needed to write a year ago; it is there because the question is now routinely asked, and because an outage caused by automated systems acting on their own would be a different and much larger story than a bad change during planned work.
What remains is a familiar and unglamorous failure: an upgrade intended to add capacity took capacity away.
How far it reached
The outage hit more than 70 countries, disrupting business operations, government services and personal users alike. More than 430 million people use the affected products worldwide.
By Friday it was affecting institutions from the Japanese government and the Philippines Supreme Court to the Australian spy agency, along with trains, hospitals and courts in North America.
In the UK, parts of the NHS including GPs and hospitals were affected, as were some schools and police forces.
Why that list is the real story
Read the institutions again: a national government, a supreme court, an intelligence agency, hospitals, police. None of those chose Microsoft 365 casually, and none of them can be said to have been careless. They made the same decision almost every large organisation has made over the past decade, for good reasons — cost, security patching, and the difficulty of running mail servers well.
Experts described the event as a "stark reminder" of how heavily businesses and vital services rely on a small group of big tech providers. That framing is right but incomplete. The concentration is not an accident or an oversight; it is the outcome of each organisation individually making the sensible choice. A court that ran its own email would be less reliable on an ordinary week and more independent on this one.
That is the trade nobody made explicitly, and it is the one worth revisiting now — not by leaving the cloud, but by asking which functions genuinely cannot share a single provider with everything else.
Why an apology with a cause is worth more than an apology
Microsoft's statement did two things, and only one of them is routine. Saying sorry costs nothing. Naming the trigger — a change made during a network capacity upgrade — is a commitment, because it can be checked against whatever the post-incident review eventually says.
That matters for customers deciding whether to trust the next reassurance. A vendor that identifies the trigger on day one and is later shown to be right builds something; a vendor that stays vague keeps its options open and earns nothing. The useful test is not how quickly a company apologises. It is whether what it said in the first 24 hours survives contact with the full review.
