Updated 5 September 2026: SpaceXAI has confirmed that the Grok outage of 3 September 2026 began at its Memphis compute center, and an OpenAI engineer says the ChatGPT failure that morning was a separate routing error. Neither company has published a full technical postmortem.
The 3 September 2026 outage that took ChatGPT, Claude and Grok offline within 90 minutes was not one failure. It was two, and only one of them explains the pair that broke first.
What Did SpaceXAI Say About Memphis?
SpaceXAI blamed a failure at its Memphis compute center for the Grok outage on 3 September 2026. The company apologised to Grok users and added that it also wanted to "apologize to our impacted compute partners" — the first admission by any of the three labs that external customers were hit.
Grok was unavailable for roughly three and a half hours from 6:30 am PT (7:00 pm IST), according to SpaceXAI's status page. Elon Musk said corrective action was being taken. SpaceXAI has not named the fault or identified the partners.
Why Claude and Grok Broke Together, but ChatGPT Didn't
| Service | Incident opened (UTC) | IST | Gap |
|---|---|---|---|
| Claude | 13:26 | 6:56 pm | — |
| Grok | 13:30 | 7:00 pm | +4 minutes |
| ChatGPT | 14:58 | 8:28 pm | +92 minutes |
Anthropic announced a compute partnership with Musk's AI unit in May 2026 and leases capacity in Memphis, as Futurism reported. Claude and Grok opened incidents four minutes apart. ChatGPT opened 92 minutes later.
That gap is the whole argument. A shared facility fails at one moment, not in two waves an hour and a half apart. Anthropic has still not confirmed that Memphis caused its outage, so the Claude link remains an inference rather than a confirmed fact.
What Caused the ChatGPT Outage?
An OpenAI engineer who identified themselves as the incident commander wrote on Hacker News that "a routing error within our infra" caused problems for some OpenAI products, and said it was unrelated to the Astra launch. GPT-6 Astra rolled out the following day, 4 September 2026.
Treat this as a strong signal rather than a settled cause. OpenAI has not published it as an official postmortem, and the account is self-identified.
Was Microsoft Azure to Blame?
No. Microsoft logged no Azure incident for that window, and Cloudflare's chief technology officer said Cloudflare saw no disruption either. We flagged the Azure theory as unproven in our original report on the outage, and nothing published since has supported it.
The real finding is narrower and, for anyone buying AI capacity, worse. Anthropic and xAI are direct competitors who share a building. One site failure took out two rival products in four minutes.
Quick Answers
Why did ChatGPT, Claude and Grok go down on the same day?
There were two separate causes. SpaceXAI confirmed a Memphis compute center outage that hit Grok, and its compute partners and an OpenAI engineer said the ChatGPT failure was a separate internal routing error 92 minutes later.
Did the GPT-6 Astra launch cause the ChatGPT outage?
No. The OpenAI engineer who handled the incident said it was not related to the Astra launch. GPT-6 Astra rolled out on 4 September 2026, the day after the outage.
Live user reports for all three platforms are tracked on Downbits: ChatGPT, Claude and Grok.