How fragile the cloud really is, and why a yacht should run on board
Last week Microsoft 365 was down for a day. It was not the first big outage and it will not be the last. Two years of failures at the world's largest providers point to one conclusion for a yacht: the systems that matter should run on board, under the owner's control, and keep working when the internet is gone.
On Thursday 12 November, Microsoft 365 stopped working for much of the world. Email, Teams and the security consoles administrators use to watch their devices went down together, and Exchange Online did not fully recover until the following afternoon. It was not an attack, and it was not unusual. In the last two years Amazon, Microsoft, Google, Cloudflare, CrowdStrike and Starlink have each had at least one failure that reached around the world, and in March 2026 drones took Amazon data centres in the Gulf offline for days.
The lesson for a yacht is not that the cloud is bad. It is that the cloud is somebody else's computer, run by somebody else's change process, a long satellite hop away. The systems a vessel cannot do without should run on board, on hardware the owner controls, and the yacht should be able to disconnect from the internet completely and keep working.
What happened at Microsoft last week
Microsoft's incident record for Exchange Online, EX1550543, puts the impact between 16:20 UTC on Thursday 12 November and 16:10 UTC on Friday 13 November, with a wider advisory, MO1550821, covering Outlook, Teams, SharePoint, Defender, Intune and the admin centres in more than 70 countries. A fix was deployed early on Friday morning, and mail queues took the rest of the day to clear.
Microsoft's preliminary post-incident report, published on 15 November, describes three things coming together. A latent defect had been sitting on its networking gateway servers since an update in July. About fifteen hours before the outage, Teams traffic was moved onto the same gateways that serve Exchange Online. And part of the gateway capacity was temporarily offline. Exchange Online cannot route around those gateways, so email was hit hardest. Microsoft found no malicious activity. We wrote on the day about why we keep crew software on board, and you can follow the major cloud services on our live Cloud Watch page.
Not an exception: two years of outages
Taken one at a time, each of these looks like bad luck. Taken together, they are the normal operating record of the services most yachts now depend on.
- 19 July 2024, CrowdStrike. A faulty content update for the Falcon security agent crashed about 8.5 million Windows computers. Airlines, hospitals and banks stopped, and many machines had to be repaired by hand, one at a time. The update arrived from the cloud, automatically, and nobody on site had a say.
- 12 June 2025, Google Cloud. A policy change with blank fields crashed the service that authorises Google Cloud's API calls, in every region at once. Gmail, Compute Engine and BigQuery were affected for around seven hours, and the failure spread to companies built on Google, including parts of Cloudflare.
- 24 July 2025, Starlink. A failure in the internal software that runs Starlink's core network took the service down worldwide for about two and a half hours. Terminals at sea were no exception.
- 20 October 2025, Amazon Web Services. A race condition in the automation that manages DNS for DynamoDB in AWS's US East region knocked out a database that much of AWS itself relies on. Alexa, Snapchat, Fortnite, Venmo and banks went down, and full recovery took about fifteen hours.
- 29 October 2025, Microsoft Azure. A configuration change to Azure Front Door slipped past its safety checks because of a software defect. For more than eight hours, Microsoft 365, Azure and many customer sites returned timeouts and errors.
- 18 November 2025, Cloudflare. A database permissions change doubled the size of a configuration file. Cloudflare's proxies could not load it and crashed, and for around three hours a large part of the web, including X, ChatGPT and Shopify stores, returned errors.
- March 2026, AWS in the Gulf. Drone strikes hit two AWS data centres in the UAE and damaged infrastructure near another in Bahrain. Services in both regions were disrupted for days, and AWS advised customers to move their workloads elsewhere.
- Through 2026, power and weather. Outage trackers recorded power, cooling and lightning failures that took individual Azure and Google Cloud regions offline for most of a day at a time.
- 12 November 2026, Microsoft 365. A day without email, Teams and security consoles, as above.
What they have in common
- Most were self-inflicted. A configuration change, an update or a piece of automation, made by the provider itself. None of them was a cyberattack.
- One change reaches everywhere. These platforms are built as single global systems, which is what makes them efficient. It also means a mistake in one place arrives in every region at once.
- Everything shares a fate. Email, files, chat, sign-in and security monitoring often run on the same platform, so they fail together.
- Dependencies are invisible until they break. A crew app built on AWS stops when AWS stops, even if you have never heard of AWS. When Google failed, companies built on Google failed with it.
- The physical world still matters. Power, cooling, lightning and, now, conflict take data centres offline, and no software design fully avoids that.
- The customer can only wait. During every one of these outages, the people affected could do nothing but refresh a status page.
A yacht feels all of this twice: once through the provider, and again through a satellite link that is never guaranteed. A shore-side office has fibre and a cloud provider's best region a few milliseconds away. A yacht has a link that drops in heavy weather, behind a mountain or in a busy anchorage.
What stops on board when the cloud stops
Most yachts have moved more ashore than they realise. When the provider or the link goes down, these are the things that typically stop:
- Email, chat and shared documents, unless copies are kept on board.
- Crew software for rotas, hours of rest, maintenance and inventory that lives in a supplier's cloud.
- Management consoles for Wi-Fi, firewalls and security tools. The network usually keeps passing traffic, but nobody can see or change it.
- Smart-home and AV devices that need their maker's cloud to accept a command.
- Sign-in itself, where accounts are checked online every time.
- Security monitoring: alerts gather in a portal ashore, so a threat can be detected on a laptop and nobody on board will know.
None of this is exotic. It is simply what happens when software is designed for an office with permanent fibre and then installed on a vessel.
Systems under the owner's control
The alternative is not to refuse the cloud. It is to decide what the vessel must be able to do on its own, and to make sure that part runs on board. In practice that means:
- Runs on hardware aboard. The servers, storage and controllers that operate the yacht's systems are on the vessel and belong to the owner, not rented by the minute from a provider on another continent.
- Works with the link down. Lights, climate, AV, crew tools, documents and the internal network keep working without the internet, not in a reduced emergency mode but normally.
- The cloud is optional. Off-site backup, remote support and access from ashore are useful, and they should be added on top of systems that work locally, not required by them.
- The owner holds the keys. Administrator accounts, encryption keys and configurations belong to the vessel, and are documented, so a change of supplier or crew never locks the yacht out of its own systems.
- Updates arrive when you choose. The CrowdStrike lesson: test updates on one machine first, and do not let a vendor push changes to every critical system at once.
- Logs stay aboard. If a console ashore goes dark, the vessel still has its own record of what happened on its network.
That is how we build our own software. EdgeCrew runs on the vessel and keeps working offline. EdgeCore controls climate, lighting and AV from the yacht's own servers. EdgeOne manages Starlink, VSAT and 4G/5G from on board, and EdgeOne /NAS keeps the vessel's files and backups aboard. Email will always need the outside world, but a yacht can keep a local copy of mailboxes and documents so that crew can still read and work while the link is down.
The last resort: pull the plug
A yacht that runs on board gains an option a cloud-dependent one does not have: it can disconnect from the internet on purpose and keep working. That matters more than it sounds. Say a supplier announces a breach, a crew laptop is infected, or something on the network starts behaving strangely. Often the safest move is to cut the connection while you find out what is going on. On a cloud-dependent vessel that move switches off the yacht. On a vessel whose systems run aboard, guests notice nothing.
The way to find out which kind of yacht you have is to try it. Choose a quiet afternoon in port, disconnect every internet link at the router, and write down what stops working. Lights, climate, AV, the crew's tools, the files they need, internal calls and the local network should all carry on. Whatever does not is your list of dependencies, and the first draft of your plan.
Security when the cloud is down
Outages are not attacks, but they are when attacks work best. In the days after a big outage, phishing emails ask people to sign in again "to restore your mailbox", fake support staff call, and urgent password resets are requested, at exactly the moment users expect strange messages and administrators are busy.
- Endpoint protection keeps working locally. Microsoft Defender's antivirus engine, downloaded signatures and the firewall run on the device. What goes dark is the portal: alerts, investigation and remote actions such as isolating an infected machine.
- Review what changed during the outage: new inbox rules, mail forwarding, app consents and sign-in policy changes are the first things an attacker sets up after a successful phish.
- Look again at mail delivered from the backlog afterwards, particularly links and attachments.
- Never reset a password or a second factor on the strength of a phone call during an incident. Verify the person first, through a channel you already trust.
- Use a different password for every account, kept in a password manager, and a second factor everywhere it is offered. Change a password when there is a reason, such as a breach or a suspicious sign-in, rather than on a calendar; that is the current advice of both NIST and the UK's National Cyber Security Centre.
Our approach to segmenting a yacht's network keeps guest, crew and vessel systems apart, so that a mistake on one device does not reach the others.
Where to start
Start with a list: every system on board, what it needs from the internet, and what happens when it cannot get it. Then run the disconnect test, and decide, system by system, what must work on its own. Our network security survey gives you the first half of that list automatically: every device on the yacht's networks, what each one exposes, and what it talks to. If you would like help with the rest, from the dependency review to moving critical systems aboard, talk to us.