ChatGPT, Claude, and Grok Go Down Simultaneously, Cause Unknown

On the morning of September 3, 2026, major foundation models went down at almost the same time. The three companies affected were OpenAI, Anthropic, and xAI. Connectivity failures hit the conversational AI services offered by each company. According to reporting by Wired’s Lily Hay Newman, explanations of the cause differed among the companies, leaving the full picture unclear.

The outages were concentrated in the 6 a.m. to 7 a.m. Pacific Time hours. They were simultaneous stoppages on a scale rarely seen. The impact on users was resolved within hours. Despite the simultaneity, however, no common cause has been identified.

Simultaneous Outages on the Morning of September 3

The sequence began with a status notice from Anthropic. At 6:23 a.m. Pacific Time, the company warned of a partial outage. Elevated error rates were confirmed for requests to Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5. It then notified users that it had identified the cause and rolled out a fix. It was logged as resolved by 9:16 a.m.

xAI also updated its status page at 6:30 a.m. It said Grok was experiencing an outage across all serving paths. At first, it only noted that the cause was under investigation. At 10:05 a.m., it said the situation had been resolved and traffic had returned to a healthy state.

OpenAI’s outage began at around 7:43 a.m. ChatGPT and Codex became unavailable for some users. It said a solution was implemented at around 8:17 a.m. and monitoring continued. The outage windows for the three companies overlapped, making it hard to accept as coincidence.

Details of the Routing Error Described by OpenAI

OpenAI cited a routing error as the cause. Spokesperson Kathleen Chaykowski told Wired that a routing error had disrupted use across multiple environments. The original explanation is as follows.

“A routing error starting around 7:43 am PT on Thursday, September 3, made ChatGPT and Codex unavailable for some users across platforms.”

She also referred to the implementation of a solution.

“As of about 8:17 am PT on Thursday, a solution was successfully implemented and is continuing to be monitored.”

The explanation is limited to routing controls within the company. It does not mention outages at outside providers. It is worded to say the impact was limited to some users. The specific nature of the error and measures to prevent recurrence were not disclosed.

Anthropic’s Partial Outage and Recovery Process

Anthropic declined to comment to the press. Status page notices are the main source of information. Starting with the partial-outage warning at 6:23 a.m., it reported identifying the cause and deploying a fix within a short time. It marked the issue as resolved at 9:16 a.m. A brief issue was also reported for Claude Sonnet 5 shortly after 9 a.m.

The range of affected models was broad. In addition to Claude Mythos 5.1 and Claude Opus 5, it included Claude Fable 5.1, part of the lineup that drew attention with Anthropic Declares Fable 5 Is Back. The outage came as a generational shift in foundation models progresses, raising concerns about spillover to development use. The company has not mentioned external factors. Its notices were structured to emphasize detection and recovery on its own side.

Anthropic also continues to face legal developments, including Round Hill Sues Suno and Anthropic for Over $1 Billion in Copyright Case. This outage is a technical operations issue. The two are not directly related. However, the transparency of outage response is drawing attention as a factor affecting trust in the company.

xAI and the Memphis Compute Site Outage

xAI’s explanation differs in nature from the other two companies. At first, the company posted formulaic language on its status page.

“Grok is experiencing issues. We are working on restoring service as quickly as possible.”

The post-recovery wording was also brief.

“We have resolved the situation, and traffic is healthy again.”

Later, its parent company SpaceX added context in the afternoon. It said there had been a stoppage that morning at a compute site in Memphis, in the United States. It said Grok’s outage stemmed from a problem at that site.

SpaceX did not respond to Wired’s request for comment. In a public statement, however, it apologized to affected compute partners.

“We’d also like to apologize to our impacted compute partners.”

The picture is one in which a stoppage of compute infrastructure led directly to a stoppage of conversational AI. The specific facility-level cause of the site outage has not been disclosed.

No Evidence to Support a

Shared-Infrastructure Failure Theory

Outages at the same time suggest a failure of shared third-party infrastructure. Normally, when multiple outages overlap in the same sector, a problem with cloud services, delivery networks, or outside contractors is suspected. This time, there is little material to support that hypothesis. OpenAI and Anthropic have not indicated external factors. Major infrastructure providers also have no outage records.

Cloudflare, Amazon Web Services, and Microsoft Azure all reported no outages that day. No large-scale cascading failure of delivery routes or compute resources has been confirmed. There were also sporadic reports of outages for Google’s Gemini. However, the company did not confirm an outage and left no record on its status dashboard. Google did not respond to Wired’s request for comment before publication.

As shown by YouTube Music Gemini Integration Transforms the Android Music Experience, Gemini’s use cases are expanding. Even if there had been a minor problem, user perceptions and the provider’s definition can easily diverge. Information on Gemini this time remains unconfirmed. There is no basis to conclude there was a four-company simultaneous outage.

SpaceX Partnership and Concentration of

Compute Resources

What stands out is the partnership around compute resources. Anthropic and xAI announced a compute partnership with SpaceX in May 2026. Under the plan, SpaceX would serve as a receiver for compute demand and drive expansion. This time, SpaceX mentioned its own site as background to the Grok outage and added an apology to partners. The existence of the partnership and its operational ties were made visible again.

However, the relationship between Anthropic’s outage and the Memphis site is unknown. Anthropic has not disclosed details of the cause. No link has been shown between OpenAI’s routing error and the site outage. There is no material tying the three incidents to a single cause. Only two facts remain: simultaneity and partnership ties.

This article is based on Wired (All Rights Reserved), relying on fair quotation under Article 32 of the Japanese Copyright Act. Quoted passages are shown verbatim in blockquotes. Company names, times, and target model names follow the wording of the original reporting.

Editorial Opinion

In the short term, securing alternatives during outages will become established as an operational requirement. Designs for switching across multiple models and monitoring of status pages may become standardized. Demands for vendor accountability will also intensify.

In the long term, concentration of compute resources and interdependence will become design issues. Partnerships such as the compute collaboration with SpaceX improve efficiency but can become single points of failure. Distributed configurations and ensuring transparency will become competitive conditions.

The remaining question is whether the simultaneous outages were coincidental. The scope of disclosure varies by company, and verification material is lacking. The challenge for enterprise users is how to audit outage information.

References

• “Nobody Is Saying Why OpenAI and Anthropic Had Outages Today”, by Lily Hay Newman — Wired, 2026-09-03T21:56:21.000Z (ARR)
• Source URL: https://www.wired.com/story/nobody-is-saying-why-openai-and-anthropic-had-outages-today/