Microsoft's servers failed their own health checks, and then the whole world got sick. For about 12 hours, Outlook, Teams and Copilot went dark for government agencies in three countries, two of the biggest AI companies on Earth, and at least one Taiwanese financial firm that had to cancel a bond transaction it had already scheduled. The company that wants to sell you an AI assistant to run your entire workday could not keep your email inbox loading.
130,000 Reports and a Very Long Day
The trouble started Thursday and peaked around 1 p.m. ET, when Downdetector logged more than 130,000 reports of Outlook problems in the US alone. Users could not send mail, could not receive mail, or could not get their inboxes to load at all. Thousands more reported trouble with Teams, Copilot and the wider Microsoft 365 suite, according to the New York Post and NPR.
By 11:30 p.m. ET there were still more than 18,000 Outlook complaints in the US, and the mess had followed the sun into Asia-Pacific business hours, with outages showing up in India, Australia and Japan early Friday local time. That is a 12-hour outage that rolled across continents like a very boring apocalypse.
Microsoft's service notice said that anything depending on Exchange Online, the company's cloud email platform that also handles calendars, contacts and tasks, was affected. Features that did not rely on Exchange kept working. So the good news is that if you only needed things that were not email, you were fine. Please enjoy your career in the 1990s.
The Servers Flunked Their Own Physical
Here is what Microsoft says happened. A portion of its infrastructure was "experiencing elevated load," which the company said was "likely contributing to the connectivity errors." A number of servers became "unexpectedly unavailable" because something caused them to fail during "automated health checks."
Read that again. The automated system built to confirm the servers are healthy is what the servers failed. Microsoft later said it was testing a fix and rolling out an alternative method for those checks to bring the impacted servers back online. Cause of the original failure? Not immediately clear, per NPR.
Then came the part nobody wants to see in a status update. Microsoft said it applied adjustments to the affected servers, but the adjustments "did not result in sustained recovery." Partial recovery in some regions. Then the company said it was "reviewing what additional actions we can take to improve overall infrastructure stability." Translation: we fixed some of it, it did not stay fixed, and we are now brainstorming.
Spies, Lawmakers and a Bond Deal Walk Into an Outage
Look at who got caught in this. Australia's spy agency and federal police were among the organizations hit on Friday. The US House of Representatives said the outage hit its users and that its engineers were working with Microsoft. Japan's Deputy Chief Cabinet Secretary Masanao Ozaki told reporters the outage was affecting the government's information system infrastructure.
The private sector fared no better. OpenAI and Anthropic both said they faced problems with Microsoft connectors. Indian airlines IndiGo and Akasa Air said their email support services were impacted. In Taiwan, the outage disrupted fund management at securities firms, life insurers and investment trust companies, and one of them cancelled a bond repurchase agreement transaction it had already scheduled, a source told Reuters.
And Workhorse Group, the electric truck maker, pushed its third-quarter earnings report from Nov. 12 to Nov. 16, citing business disruptions from the global Microsoft outage. Somewhere an investor relations team had to write the sentence "our earnings are late because Outlook." Nobody cheered.
Not a Hack, Just 'An Update That Has Gone Awry'
Microsoft said it was "not aware" of any malicious activity behind the outage. Australian Home Affairs and Cyber Security Minister Tony Burke relayed the company's view at a press conference, saying Microsoft believes this is similar to CrowdStrike, "where there is an update that has gone awry."
CrowdStrike, for those who blocked out July 2024, was one of the largest internet disruptions in history, hobbling hospitals, banks and airports. The comparison is not comforting. The best-case explanation for a global outage is that nobody attacked us, we did this to ourselves.
To be fair, a self-inflicted wound beats a foreign adversary rummaging around in government email. But it also means there is no villain to blame and no enemy to rally against. Just a very large company, a very large dependency, and a very large oops.
Third Time in 13 Months Is Not a Fluke
This is the second disruption to Microsoft 365 this year. There was also an outage in January and another in October of last year. And last month, Microsoft reported an outage of its Azure cloud platform that hit its productivity software and industries worldwide.
EToro market analyst Josh Gilbert said investors will likely shrug this one off, because "very few of these customers can realistically walk away from the Microsoft ecosystem overnight." That is the whole business model in one sentence. You are not a satisfied customer. You are a hostage with a subscription.
Gilbert also pointed at the real problem. If reliability starts to look like a recurring issue, it "hands every IT department a reason to think twice before putting even more of their business in one provider's hands." And the market, he said, will want to see Microsoft "keep the lights on before it asks customers to spend more on Copilot." Exactly. Copilot was one of the things that went down.
The Dingo Take
Microsoft says it is "aggressively pursuing all pathways to reduce impact." The facts say the company spent 12 hours failing to reduce impact, watched its first round of fixes not stick, and is still reviewing what other actions it can take. Aggressive is not the word. Hopeful is the word. Hopeful with a status page.
The real story is not the outage. Outages happen. The real story is that a US legislature, an Australian spy agency, a Japanese government, two AI giants, two Indian airlines and a Taiwanese bond desk all hung their operations on the same cloud email system, and when it hiccupped they all hiccupped together. Nobody planned for this because planning for this means paying for redundancy, and redundancy does not show up on a quarterly earnings slide.
Meanwhile Microsoft wants everyone to hand even more of their work to Copilot, an AI assistant that lives inside the same infrastructure that just tripped over its own health checks. Get the lights to stay on first. Then tell us about the future of work. Until then, maybe keep a phone number written down on actual paper.
The cost of consolidation always arrives eventually. Today it arrived as a canceled bond deal and a delayed earnings call. Next time it might arrive somewhere that matters more. Every IT department that just spent a Friday refreshing a blank inbox now has a very good question to ask its vendor, and a very good reason to ask it loudly.
Comments