ChatGPT, Claude and Grok simultaneously down: What happened on September 3

Lisa Ernst · 05.09.2026 · AI News · 10 min.

On September 3, 2026, three of the best-known AI services, ChatGPT from OpenAI, Claude from Anthropic, and Grok from SpaceXAI, reported problems almost simultaneously. For users, this looked like a joint ChatGPT-Claude-Grok outage . Internationally, the event was correspondingly often described with search queries like "chatgpt claude grok outage".

The temporal overlap is confirmed. However, a single common cause is not. OpenAI cited its own routing error, SpaceXAI pointed to an outage at its data center in Memphis for Grok, and Anthropic spoke of an infrastructure problem without publicly naming the same concrete trigger. It is precisely this distinction that is important when trying to understand what actually happened on that Thursday.

Source: simpleicons.org

Quick Facts

• Yes, the outages overlapped. Claude reported problems starting at 13:26 UTC, Grok starting at 13:30 UTC, and OpenAI dated its routing error to around 14:43 UTC.
• OpenAI confirmed a routing error. ChatGPT and Codex were therefore unavailable or error-prone for some users across platforms.
• Grok had a different publicly stated trigger. SpaceXAI attributed the outage to a failure at its Memphis Compute Center.
• Anthropic confirmed an infrastructure issue. However, the company did not provide evidence that the Claude outage was directly caused by Memphis or Colossus 1.
• A single, massive outage is not proven. Public status information from Cloudflare and major cloud providers provided no indication of a single global infrastructure failure that would explain all three services simultaneously.

ChatGPT-Claude-Grok Outage: The Confirmed Timeline

The status messages show how closely the events were timed. The following overview uses UTC and also converts to Central European Summer Time (CEST, UTC+2). At OpenAI, the start and mitigation times mentioned by a spokesperson align with the entries on the status page.

Thus, the officially reported incident windows for ChatGPT, Claude, and Grok overlapped for about 93 minutes, from 14:43 to 16:16 UTC. However, this does not mean that every user experienced a complete outage across all three services for this entire period. OpenAI had already activated its mitigation at 15:17 UTC, and status pages reflect aggregated availability.

What Happened with ChatGPT

OpenAI described the incident on its status page as "Elevated errors across ChatGPT and Codex". Affected were 15 ChatGPT components and four Codex components. OpenAI stated to the media that a routing error caused ChatGPT and Codex to be unavailable for some users across multiple platforms starting around 7:43 AM Pacific Time, which is 14:43 UTC.

A mitigation measure was active from approximately 15:17 UTC. OpenAI then entered the monitoring phase and later marked the incident as resolved. For individual Codex remote control users, there was a specific follow-up effect: they may have had to re-pair their mobile device after the incident.

A routing error does not necessarily mean that the actual AI models have failed. Simply put, the infrastructure that forwards incoming requests to the correct service, cluster, or backend can malfunction. For the user, the result is the same: requests fail, hang, or the service appears completely offline.

What Happened with Claude

Source: thesvg.org

Anthropic initially reported a brief Sonnet 5 incident on September 3, followed by a broader outage affecting multiple Claude models.

The situation at Anthropic was somewhat more complex. An initial incident with elevated errors for Claude Sonnet 5 began at 12:37 UTC and was already resolved by 12:56 UTC. Only half an hour later, Anthropic opened a new incident due to elevated error rates for Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5.

Later, Anthropic released a broader list of affected models, including Mythos/Fable 5.1, Mythos/Fable 5, Opus 5, Opus 4.8, and Opus 4.6. At 13:41 UTC, the status page reported that the cause had been identified; by 16:06 UTC, a fix had been rolled out. According to Anthropic, the impact ended at 16:16 UTC.

In statements to The Register, Anthropic described the event as an infrastructure problem that temporarily affected Claude.ai, Claude Code, Claude Cowork, and the Claude API. At that time, Anthropic did not release a detailed technical post-mortem explanation that clearly attributed the outage to a specific external data center or cloud provider.

What Happened with Grok

Source: thesvg.org

The official Grok status page documents the outage from 13:30 to 17:07 UTC, a duration of 3 hours and 37 minutes.

xAI documents the Grok incident most clearly: At 13:30 UTC, the investigation of a "Models outage" for Grok Web began. At 17:07 UTC, the status page reported that traffic was healthy again. The duration shown there is exactly three hours and 37 minutes.

The status page itself does not name a root cause. SpaceXAI later publicly explained that the Grok problems occurred after a failure at the Memphis Compute Center . At the same time, the company also apologized to affected compute partners. This second phrasing is particularly interesting for the question about Claude because Anthropic has indeed been obtaining compute capacity from SpaceX since May 2026.

Were Claude and Grok connected via SpaceX?

Source: models.dev

Anthropic and SpaceXAI confirmed a major compute partnership around Colossus 1 in May 2026. This makes a technical connection plausible, but it does not prove the cause of the Claude outage on September 3.

This is the most exciting, but also the most easily misrepresented part of the story. On May 6, 2026, Anthropic officially announced its intention to use the entire compute capacity of the SpaceX data center Colossus 1 . According to both companies, the facility includes more than 220,000 Nvidia GPUs; Anthropic spoke of more than 300 megawatts of additional capacity.

However, Anthropic does not run Claude exclusively on this infrastructure. The company also names AWS Trainium, Google TPUs, and Nvidia GPUs from other partnerships. Therefore, the compute agreement does not automatically mean that every Claude request runs through Colossus 1 or that an outage there would necessarily affect Claude.

The fact that SpaceXAI explicitly apologized to "Compute Partners" after the Memphis outage, and Anthropic is a publicly confirmed compute partner, makes a connection plausible. . However, it is not yet confirmed. Anthropic itself has not publicly attributed the September 3 outage to SpaceX, Memphis, or Colossus 1. Therefore, a plausible infrastructure connection should not be presented as a confirmed root cause statement.

Was Cloudflare or a Major Cloud Platform to Blame?

The nearly simultaneous problems initially suggested a common third-party provider: Cloudflare, AWS, Google Cloud, or Microsoft Azure would be typical candidates if several large online services fail at the same time. However, there is no reliable confirmation of this.

The Register reported that Cloudflare had explicitly stated it had no significant outage at that time. According to the report, the public status pages of AWS, Google Cloud, and Microsoft Azure also showed no comparable widespread outage. This is not mathematical proof that no common dependency existed: status pages can react with delays, and major providers share many smaller network, DNS, hardware, and transit dependencies. However, the available evidence supports no single global cloud outage as an explanation for all three AI services.

Why Did It Still Seem Like a Joint Mega-Outage?

From a user's perspective, the perception was understandable. If you opened ChatGPT, got an error, switched to Claude and also saw problems there before Grok also failed, you had effectively lost three independent fallbacks. In addition, the largest AI providers rely on very concentrated infrastructure: few GPU types, few hyperscalers, large data centers, global networks, and numerous common internet building blocks.

It is precisely here that one must distinguish between correlation and causation . The same time window is a strong signal to look for common dependencies. However, it is not proof that all outages had the same technical origin. As of September 5, 2026, public information is more likely to indicate at least two clearly distinguishable causes: OpenAI's routing error and SpaceXAI's Memphis outage. For Anthropic, the concrete cause remains publicly open above the description "infrastructure problem".

What Companies Should Learn from the Triple Outage

For companies that have integrated AI into productive workflows, the incident is more important than the question of which chatbot was offline the longest that day. It shows that a supposed multi-provider setup does not automatically mean real fault tolerance.

• Track dependencies down to the infrastructure: Two different model providers may rely on the same compute, network, or cloud partner.
• Actively test fallbacks: A second provider only helps if its health status is actually healthy. Blind switching can multiply errors.
• Limit retries: APIs should work with timeouts, exponential backoff, and jitter instead of generating additional load during an incident in aggressive retry loops.
• Cache work: Prompts, drafts, and intermediate results should not just exist in a volatile chat session.
• Maintain manual emergency paths: A business-critical process should not be completely blocked just because no frontier model is accessible.
• Secure agents particularly: Long-running autonomous systems need checkpoints, idempotency, and clean resumption after disconnections.

This applies particularly to persistent AI agents that work autonomously over longer periods. Our overview of Grok Bot and its always-on workflows shows why not only model quality, but also restart, permissions, and controlled execution become important for such systems.

What to do when ChatGPT, Claude, and Grok go down simultaneously again?

1. Check official status pages. For OpenAI, that is status.openai.com, for Claude status.claude.com and for Grok status.x.ai.
2. Do not immediately change local settings. If the provider itself confirms an incident, browser reinstallation, password change, or router reset usually won't help.
3. Handle errors cleanly for API workloads. 5xx errors and timeouts should be logged, retried with limits, and moved to a queue if necessary.
4. Test fallback providers separately. A health check should confirm that the substitute service is actually reachable before switching.
5. Save intermediate results locally. Long inputs, analyses, or generated results should be saved outside the chat before further attempts are made.
6. Only search locally if status pages are unremarkable. Only when no provider problem is visible are the company firewall, DNS, VPN, browser extensions, or your own network the more likely source of error.

FAQ

Conclusion

The unusual part of September 3, 2026, is not fabricated: ChatGPT, Claude, and Grok did indeed have overlapping outages. The headline "all three down simultaneously" therefore fundamentally describes the user experience correctly. However, it should not lead to the conclusion that there was a single confirmed mega-outage in the background.

OpenAI named a routing error as the cause of its problem. SpaceXAI pointed to the Memphis Compute Center for Grok. Anthropic confirmed an infrastructure problem but publicly left the specific underlying dependency open. The existing Anthropic-SpaceX compute partnership makes a connection between Claude and Grok technically conceivable; without confirmation from Anthropic, it remains a reasoned hypothesis and not a fixed fact.