On September 3, the three largest AI chatbots—ChatGPT, Claude, and Grok—simultaneously experienced outages lasting several hours. While not an internet-wide crash, it was enough to prevent many users from working, logging in, running code, or making API calls as usual.
For OpenAI, ChatGPT began returning errors at around 11 a.m. Eastern Time. The company’s status page recorded “elevated errors” across both ChatGPT and Codex. The impact went beyond standard conversations: logging in, file uploads, voice mode, search, deep research, and image generation also ran into issues. According to The Register, OpenAI later attributed the cause to a routing error that began at 7:43 a.m. Pacific Time, stating that a fix was rolled out approximately 34 minutes later.
Claude also experienced two waves of errors during the morning. Anthropic began investigating elevated error rates in Claude Sonnet 5 at 8:37 a.m. Eastern Time, then logged an additional issue affecting multiple models starting at 9:26 a.m. The company confirmed an “infrastructure issue” caused partial outages across Claude.ai, Claude Code, Claude Cowork, and Claude API; services were restored at 16:16 UTC. Downdetector recorded around 1,400 error reports at its peak, with Claude Chat and Claude Code accounting for the majority.
xAI’s Grok began experiencing errors around 9:30 a.m. Eastern Time across the web, iOS, Android, the X-integrated interface, and two US API regions. Users were greeted with messages stating the model was overloaded. Later, SpaceX AI apologized and linked the incident to a disruption at its Memphis compute center. The phrase “compute partners affected” made the overlapping timing with Claude noteworthy, though neither company has confirmed that Grok was the direct cause of Anthropic’s errors.
The simultaneous disruptions across three independent services initially fueled speculation about a shared infrastructure dependency. Cloudflare denied any major incidents, and status pages for AWS, Google Cloud, and Azure showed no related issues. The available evidence therefore points to isolated incidents overlapping in time, rather than a single provider causing an “AI crash.”
Whatever the ultimate cause, the incident serves as a practical reminder: with ChatGPT serving more than 1 billion weekly active users, Claude estimated at around 300 million monthly users, and Grok at about 117 million, chatbots are no longer just auxiliary tools. For both individual users and enterprises, contingency plans—such as saving work locally, using alternative models or tools, and avoiding routing entire workflows through a single API—are becoming basic operational necessities rather than over-preparedness.