On the morning of September 3, 2026, ChatGPT, Claude, and Grok all broke at roughly the same time. For a few hours, three of the biggest names in AI were struggling or completely unreachable, and nobody had a clear answer for why it was happening to all of them at once.
Around 5:40 PM local time, OpenAI started seeing a routing error that knocked ChatGPT and its coding tool Codex offline for a chunk of users. Logins failed. File uploads failed. Voice mode, search, deep research, and image generation all stopped working properly.
OpenAI said it had a fix in place by about 6:20 PM, more than half an hour later.
Anthropic’s Claude had it worse. Claude.ai, Claude Code, Claude Cowork, and the Claude API all went into a partial outage because of what the company called an infrastructure issue. The outage lasted about three hours, and Anthropic confirmed service was fully restored at 7:16 PM.
Grok, built by xAI, started acting up around 4:30 PM local time. People trying to use it on X got a message saying the model was overloaded and to try again later. xAI eventually traced the problem back to an outage at its data center in Memphis.
SpaceX, which shares infrastructure with xAI, later apologized for the disruption and said it had also affected some of its “compute partners,” a comment that hinted its Memphis problems might have rippled outward.
On top of all that, Microsoft Azure reported a networking problem in its East US region during the same window, and users started noticing issues with Google’s Gemini too, though Google never confirmed a full outage on its end.
Why This Is Odd
Normal outages happen constantly. Any single app going down for an hour isn’t news. But three competing AI companies breaking at almost the same moment is rare, and it happened at a time when millions of people use these tools daily for work, not just casual chatting.
The obvious question is whether something upstream broke and dragged all three services down with it. Microsoft Azure provides cloud infrastructure used by all three companies, so its East US incident became an immediate suspect.
Cloudflare, which also supports parts of all three companies’ systems, pushed back hard on any suggestion that it was the cause, telling The Register its services were running normally the entire time.
Nobody has confirmed a shared root cause. It’s entirely possible this was just an unlucky pileup of three separate failures. AI products run on layered infrastructure most users never see: cloud regions, networks, databases, authentication systems, and the servers that actually run the models.
READ: Cloudflare Outage Takes Down Multiple Websites Including ChatGPT, X and Spotify
A crack in one shared layer can take down several products that otherwise have nothing to do with each other. Yet, three unrelated failures landing in the same two-hour window would also be a pretty wild coincidence.
A few years ago, an AI outage meant people couldn’t ask a chatbot trivia questions for an afternoon. That’s not really the situation anymore. These tools are now wired into software development, customer support, research workflows, and everyday business operations.
When ChatGPT, Claude, and Grok all fail at once, it’s not just an annoyance but a reminder of how much day-to-day work now depends on infrastructure that most people never think about until it breaks.
OpenAI has been here before too. In January 2025, ChatGPT had a multi-hour outage tied to problems in its database layer, which pushed the company to work on making that infrastructure more resilient.
This latest incident suggests that even with a year of hardening behind them, the systems powering the biggest AI products are still capable of failing in ways that catch everyone off guard, sometimes all on the same morning.
As of now, all three services are back online. What actually connected the failures, if anything did, remains unconfirmed.
