ChatGPT, Claude and Grok go offline in morning outage

ChatGPT, Anthropic’s Claude and xAI’s Grok all experienced service interruptions Thursday morning. Users began reporting errors for Grok and Claude near 9 a.m. Eastern Time, and complaints about ChatGPT surged about 90 minutes later. Outage tracker data showed ChatGPT reports jumped from roughly 5,000 to more than 22,000 in ten minutes and peaked near 35,000; Claude and Grok reports each peaked near 1,500.

OpenAI’s status page logged elevated error rates across ChatGPT and its Codex coding agent and listed 19 affected components. Anthropic displayed an in-app message that read, “Due to unexpected capacity constraints, Claude is unable to respond to your message. Try again soon.” xAI reported no incident even as Grok returned an unavailable status to users.

No company publicly confirmed a cyberattack, and no major cloud provider reported a related fault. Some users pointed to a large GPU cluster in Memphis known as Colossus 1, where xAI and SpaceX consolidated capacity and where Anthropic has leased compute. Public posts noted Anthropic draws over 300 megawatts across roughly 220,000 Nvidia chips at that site and that xAI operates a larger GPU fleet. None of the labs or hosting firms linked the interruptions to a specific datacenter or confirmed hardware failure.

Cardano co-founder Charles Hoskinson suggested on public posts that a nation-state actor could be responsible, citing the affected systems’ reliance on Nvidia accelerators and noting that Google’s Gemini uses Google-built TPUs.

Developers and customers reported disrupted workflows and tried manual workarounds as services were strained. Google reported no incident affecting its Gemini model on its status pages, though users filed hundreds of reports and the company flagged an issue with issuing new API keys.

The outages follow earlier incidents at AI providers and internet infrastructure suppliers. Anthropic experienced a separate outage in March this year. OpenAI and Anthropic described elevated errors without naming a cause; xAI posted that it had no incident to report while users continued to see interruptions.