AOL 1996: routine maintenance took the world's largest online service down 19 hours

Before dawn on 7 August 1996, AOL technicians shut down their computer system for what the company called routine maintenance. The Washington Post reported that with the system down, so was the software that would normally detect glitches between AOL's computers and its network partner ANS — and a glitch hit ANS during exactly that window. The world's largest online service did not come back.

For 19 hours, America Online was simply gone. The error screen told users to try again in 1 hour and 45 minutes; the outage ran ten times longer, led national newscasts, and knocked NASA's announcement of possible life on Mars off the front page of the New York Times. Analysts told NPR that online services were growing too fast for their own systems — and that providers people rely on 'screw things up' because they lack experience with failure.

The decisions behind it were ordinary, which is what made them dangerous. Maintenance at AOL meant taking the whole service down because there was no online upgrade path. Monitoring was switched off as part of the same window. Growth was outrunning infrastructure. And when it was over, CEO Steve Case, citing the newness of the medium, could not guarantee it would not happen again. A strikingly similar outage in May 1996 — a single power phase cut at AOL's data centre, generators that never noticed, dead batteries — had never made a single column of news, so nothing changed.

Matt Korn, AOL's VP of Operations at the time, later recalled that the company eventually rebuilt the system so it no longer needed downtime for maintenance at all, silently fixing the original weakness. By then the blackout had already made history as the first internet outage the general public experienced as news — proof that reliability becomes a front-page topic the moment enough of daily life runs through your servers.

19 hours offline for the world's largest online service; it led national news, pushing NASA's Mars-life finding off the NYT front page

Maintenance procedure required a full shutdown of the entire service because there was no rolling or online upgrade capability.

The software that watched for network faults was powered down inside the same window, so the ANS glitch met no early detection.

Hypergrowth had outrun infrastructure; contemporaries already warned the online services were 'growing too fast for their own systems'.

A near-identical May 1996 infrastructure failure taught the organisation nothing publicly because it went completely unreported.

Leadership framing lowered the bar for recurrence instead of committing to a fix, treating outages as an unavoidable cost of a new medium.

Maintenance windows are outage rehearsals. AOL treated shutting the whole service as routine — and shut its monitoring off to do it. Reliability debt hides until the routine fails at national scale.

AOL rebuilt its systems so maintenance no longer required taking the service down, closing the specific weakness that caused the blackout, according to Korn. The company paid no lasting commercial price for the outage — it kept growing, was under separate SEC investigation at the time over its accounting, and three decades later the AOL brand was sold to Bending Spoons in 2025 for $1.5B. The 19-hour blackout remains the template case for what happens when critical infrastructure grows faster than its ability to be maintained.