ChatGPT, Claude and Grok Go Down Together in Rare Outage

ChatGPT, Codex, Claude and Grok all stopped working properly within the same half hour on the morning of September 3, and this time it was not one company's bad day. Anthropic's status page had already been flagging elevated Claude errors since around 9:41 AM ET, but by roughly 11:07 AM ET, OpenAI's status page acknowledged issues across ChatGPT and Codex too, and xAI's page confirmed Grok was down. Three competing AI labs, three separate incident reports, all inside the same window.

That kind of overlap does not happen by coincidence very often. Reporters chasing the story quickly landed on the same theory: Microsoft Azure was seeing its own spike in outage reports at the exact same time, and all three companies lean on Azure for parts of their cloud footprint. Nobody has confirmed a single root cause publicly as of this writing, so treat that as the leading theory, not a fact.

RelatedAnthropic Ships Claude Opus 5 at Half of Fable 5’s Price

What actually broke, and when?

Claude's trouble came first. Anthropic's status page began reporting elevated errors on requests to multiple Claude models around 9:41 AM ET, later saying engineers had identified the cause and were working on a fix, though as of roughly 10:49 AM ET it was still "continuing to work on a fix." That is the same incident this site covered a few hours earlier, when Opus 5, Opus 4.8, Opus 4.6 and both Fable 5 versions were named as affected while Sonnet and Haiku stayed off the list.

What changed is the scope. By about 8:07 AM PT (11:07 AM ET), ChatGPT and Codex started throwing errors too, and Grok's own site began showing problems. DownDetector reports spiked for all three at once. Cursor, the AI coding editor, got dragged in as a side effect since it depends on Claude and other model APIs to run its agent features. Google's Gemini was not confirmed to be affected, despite some scattered user reports.

Why would three rival AI companies go down at the same time?

The obvious explanation, and the one multiple outlets converged on, is infrastructure. OpenAI, Anthropic and xAI all run parts of their stack on Microsoft Azure, and Azure's own status trackers showed a matching spike in outage reports during the same window. If a shared region, load balancer, or networking layer inside Azure had a bad morning, every tenant sitting on top of it would see symptoms that look identical from the outside: timeouts, failed completions, 500-series errors, even though the actual products (ChatGPT, Claude, Grok) share none of their model weights or training pipelines.

This is worth sitting with for a second, because it is easy to read three chatbots failing together as "AI is having a bad day" when the more accurate read is "a small number of cloud providers now sit underneath most of the AI industry." A single provider's infrastructure incident can look, from a user's chair, indistinguishable from three separate product failures.

Is this tied to OpenAI's Astra launch?

OpenAI is widely expected to ship its next flagship model, reportedly codenamed Astra and rumored to bring a jump from GPT-5.6 toward GPT-6, around this same period. Some users online floated the idea that the downtime was a deliberate maintenance window ahead of a big release, the way an Apple Store sometimes goes dark before a product drop. There is no evidence for that. Outages that coincide with Azure trouble, Claude trouble that started two hours earlier, and Grok trouble on an entirely separate stack are a strange way to stage a launch, and nothing on OpenAI's own status page frames this as planned.

RelatedCursor Builds 'Sand' Agent to Rival Claude Cowork

Who actually gets hurt by this?

Anyone with a paid seat riding on these APIs in production, not just people refreshing a chat window. Coding assistants like Cursor and Claude Code that call these models mid-task lose their agent loop, not just the chat UI. Startups that wrapped their entire product around a single model provider found out in real time what "single point of failure" costs when three of them go down together instead of just theirs. Customer support bots, internal tools, and anything scripted to call the API without a fallback provider simply stopped answering for the duration.

1. 9:41 AM ETAnthropic's status page reports elevated errors on Claude requestsOpus and Fable models named; Sonnet, Haiku unaffected
2. 10:49 AM ETAnthropic says it identified a cause, still working a fix
3. ~11:07 AM ETChatGPT, Codex and Grok begin reporting errors tooDownDetector spikes for all three simultaneously
4. Shortly afterOpenAI and xAI status pages both acknowledge active incidents
5. OngoingOpenAI reports some services recovering; Claude and Grok fixes still in progress

What it means for the market

Watch Microsoft here, not just OpenAI, Anthropic or xAI. If Azure infrastructure turns out to be the actual common thread, that is a reliability story for Microsoft's cloud business as much as it is an AI story, since Azure's pitch to enterprise buyers rests heavily on being the dependable backbone under exactly this kind of workload. For the AI labs themselves, an outage alone rarely moves a valuation, but a pattern of them feeds directly into enterprise contract negotiations, where uptime SLAs are now a real line item. The signal for investors watching MSFT, and for enterprise buyers evaluating any of these vendors, is concentration risk: a widening slice of the AI industry now depends on the same handful of hyperscale clouds, and today is a small, contained demonstration of what that looks like when it goes wrong.

Our take

None of this is unusual in isolation. Individual AI outages happen often enough that they barely make news on their own. What makes today worth writing up is the timing: three separate companies, three separate incident reports, converging inside roughly ninety minutes, with the most plausible explanation sitting one layer below all of them in shared cloud infrastructure rather than in any one company's code. That is the actual story here, more than "the chatbots are down again."

Original analysis by GenZTech Team. Sources: Anthropic status page, OpenAI status page, xAI status page.