ChatGPT, Claude, and Grok all crashed at once. What's going on?
If you tried to ask ChatGPT something Thursday evening and got a spinning wheel instead of an answer, you weren't alone — and it wasn't just OpenAI. On September 3, 2026, three of the biggest AI platforms on the planet — OpenAI's ChatGPT and Codex, Anthropic's Claude, and xAI's Grok — went down within roughly the same window, leaving millions of users staring at error screens.
The timeline
The trouble started on Anthropic's side first. Around 20:37 Beijing time, users began reporting errors on Claude Sonnet 5. That briefly recovered, but a broader spike in error rates soon hit multiple Claude models, including Opus and the Fable series.
xAI was next. At 21:30, Grok's status page marked the web service as an outright outage.
OpenAI joined the list at 22:58, when its status page added an incident for "elevated error rates" across both ChatGPT and Codex. By the time Chinese financial media picked it up at 23:11, all three companies had acknowledged problems and said they were investigating.
Downdetector, which aggregates user-submitted outage reports, recorded over 20,000 reports for OpenAI at the peak — some sources put the number closer to 30,000 or 40,000. Claude and Grok each drew roughly 1,000–1,200 reports. Developer tools like Cursor, which depend on these APIs, also showed issues.
Why three at once is the real story
A single AI platform going down is annoying. Three competitors going down simultaneously is genuinely weird.
OpenAI, Anthropic, and xAI run completely separate companies on separate infrastructure. Three independent failures lining up within a couple of hours would be a serious coincidence. The more likely explanation is a shared dependency somewhere lower in the stack — a cloud region, a CDN provider, a DNS service, or a network backbone that several AI platforms rely on. None of the three has confirmed that yet, and until someone publishes a post-mortem, that remains an inference rather than a finding.
The OpenAI twist: was this planned maintenance?
Here's where it gets interesting. Earlier the same day, OpenAI posted two cryptic messages on X. The official OpenAI account shared an image with no caption, and the ChatGPT account followed up with: "The stars are almost aligned."
That wording — and the timing — immediately sparked speculation that OpenAI was about to announce Astra, the model widely known as GPT-6. Some observers joked that the outage looked like the classic Apple Store routine: take the site down right before a product launch so the new stuff can go live.
It's a fun theory, but it doesn't explain why Claude and Grok would also be down. Unless the "stars aligning" comment was just a coincidence and the real cause is that shared infrastructure problem nobody has named yet.
Not the first time
This isn't even the first time this exact trio has gone dark together. Back on June 2, 2026, Claude, ChatGPT, and Grok all experienced outages in the same window, with all three recovering within roughly the same stretch. A pattern like that makes the shared-dependency explanation look even more plausible.
For teams building on these APIs, the lesson is the same both times: if your product's uptime is wired to a single AI vendor's health dashboard, you're one shared outage away from going dark alongside them.
Where things stand now
As of the latest updates, service appears to be coming back for some users after about half an hour of disruption, though recovery is uneven. All three companies are still investigating, and none has published a root cause.
The reliable way to check is each company's own status page — status.openai.com, Anthropic's status page, and Grok's site — which update faster than news coverage does.
If you're a user, there's nothing to fix on your end. Logging out and back in, clearing cookies, or switching browsers won't help when the problem is upstream. Repeatedly hitting retry just adds load to services already struggling.
Three giants, one blackout, no explanation yet. Whatever the cause turns out to be, it's a reminder of how much of the AI world runs on a surprisingly small set of shared pipes — and how quickly all of it can go quiet at once.
Sources: 9to5Mac, Business Upturn, 9to5Google, NDTV Profit, Downdetector, 财联社/CLS, Lookonchain, ChainCatcher, OpenAI/Anthropic/xAI status pages.
No comments yet