service outage
We speak of a service outage when an online service – such as an AI chatbot, online banking, or email – is temporarily unavailable or not functioning properly.
In plain terms
Service outage is the power cut of the digital world. Just as everything suddenly goes dark in an apartment, an application or website stops working: it does not load, gives an error message, or spins endlessly. The error is usually on the provider's machines, not on your phone or laptop.
This is especially noticeable with AI tools, because most popular large language models do not run on your own machine but in remote data centers. If something goes wrong there, the chatbot stops responding, the coding agent halts mid-task, and every application using that same model in the background becomes paralyzed.
Outages are not always total. Partial outages are common, where the service is only slower, some functions are missing, or it is unavailable only in certain countries. What distinguishes it from planned maintenance is that it happens unexpectedly and no one gave advance notice.
How it works
- Something goes wrong in the background: a faulty software update, a bad configuration, overload, a network error, a power outage at the data center, or an attack.
- The error spreads further. Today's services are built on each other, so the failure of a single cloud provider or AI model can drag down thousands of other applications.
- The provider's monitoring systems trigger alerts, and the on-duty engineers begin searching for the cause. In the best case, the error appears on the public status page, though this often happens with a delay.
- The fault is remedied by restoring the previous working version, rerouting traffic to backup machines, or restarting the affected systems.
- Later, serious companies publish a post-mortem analysis of what happened and what changes they will make to prevent recurrence.
An everyday example
On Monday morning, you ask the AI assistant to summarize the weekly report, but it only outputs: 'Something went wrong, try again later.' Your colleague faces the same situation, and the provider's status page shows a red alert. This is a service outage: the fault is not on your end, and you can only wait or switch to another tool.
Who is affected
A common misconception
Limits and risks
- 99.9 percent availability sounds good, but it still allows nearly nine hours of downtime per year.
- If you entrust critical work to a single AI tool, you have nothing to fall back on when it fails, so have a Plan B for time-sensitive tasks.
- An agent working on a multi-step task may stop halfway due to a failure, so afterward check what was completed and what was left unfinished.
- Compensation is usually minimal. For business subscriptions, providers typically only offer credits from the fee, and for free or consumer packages often nothing at all. Your lost revenue is generally explicitly excluded by the contract terms.
Similar, but not the same
- scheduled maintenance – Maintenance is a pre-announced, scheduled shutdown; an outage is unexpected and unintentional.
- data breach – In an outage, the service is unavailable, but your data is typically safe; in a data breach, the service may function while data reaches unauthorized parties.
- usage limit – If you reach your package's message or request limit, the service restricts only you while functioning flawlessly for others.
Related concepts
availability, operational safety, IT infrastructure
partial outage, complete shutdown, regional outage
downtime, service-level agreement, data center, cloud service, redundancia
A brief history
The English word "outage" was used mainly for power supply interruptions before being adopted by the IT industry. Major outages brought it into public awareness. On October 4, 2021, Facebook, Instagram, and WhatsApp became unavailable for hours. On July 19, 2024, a faulty CrowdStrike update disabled millions of Windows machines worldwide. On October 20, 2025, a failure in a major region of Amazon's cloud (AWS) shut down thousands of applications.
Where things stand
At the end of 2026, service outages are a normal part of digital life, and the major AI providers regularly report minor and major disruptions. More and more companies are preparing with a backup model or by using several providers in parallel.
In the news
- Nvidia's FOX agent-based factory management platform promises faster investigation of downtimeNov 4, 2026 · Ipar
- Digital twins could cut unplanned downtime by up to 30%, report saysOct 26, 2026 · Ipar
- NVIDIA to assess former German nuclear plant sites for AI data centersOct 22, 2026 · Ipar
- Hungary's EESZT health data platform partly offline this weekend for maintenanceOct 15, 2026 · Health
- Bank regulators in the US and Europe tighten scrutiny of AI risksOct 13, 2026 · Money
- Quality debt: AI-accelerated development brings production outages and testing gapsOct 12, 2026 · Code
- Analysis: outages have risen at several providers since coding agents launchedOct 5, 2026 · Code
- Worldwide Claude outage hits Claude Code on 29 September 2026Sep 29, 2026 · Code
Also known as
service outage · shutdown · malfunction · service outage · outage · service outage · downtime · downtime
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.