ChatGPT, Claude, and Grok Outage: What Actually Happened

ChatGPT, Claude, and Grok all had outages within the same 90-minute window on September 3, 2026, and none of the three companies pointed to the same cause. OpenAI blamed a routing error, Anthropic cited an unspecified infrastructure issue, and xAI's Grok problems were later tied to an outage at a SpaceX compute facility. No shared root cause has been confirmed, and that's arguably the more interesting story for anyone building on top of these platforms.

Key Takeaways

• ChatGPT/Codex, Claude, and Grok all reported elevated errors or outages within roughly 90 minutes of each other on September 3, 2026.
• OpenAI attributed its outage to "a routing error"; Anthropic cited an infrastructure issue affecting Claude.ai, Claude Code, Claude Cowork, and the API; xAI pointed to a SpaceX compute center outage in Memphis.
• Cloudflare explicitly denied any disruption on its end, and AWS, Google Cloud, and Microsoft Azure status pages showed no incidents at the time.
• Anthropic signed a deal in May 2026 to use all compute capacity at SpaceX's Colossus 1 data center — a real, confirmed link between Anthropic and SpaceX's infrastructure, though Anthropic hasn't said its outage was caused by the Memphis incident.
• No single confirmed root cause ties all three outages together. Widely circulated "it was Azure" claims traced back to Downdetector and StatusGator crowd reports, not to Microsoft's own status page or any company statement.

What Actually Happened, and When?

On the morning of September 3, thousands of users on Downdetector reported problems with ChatGPT, Claude, and Grok in close succession. According to a company spokesperson who spoke to The Register, OpenAI's routing error began around 7:43 am PT and was resolved by about 8:17 am PT. Anthropic's status page logged a three-hour-six-minute outage with elevated error rates on Claude Sonnet 5 and other models, and a spokesperson confirmed service was fully restored at 16:16 UTC. xAI's status page showed it had begun investigating a Grok outage at 6:30 am PT the same morning.

What Did Each Company Actually Say Caused It?

Here's where the three accounts diverge. OpenAI told The Register its outage came from a routing error affecting ChatGPT and Codex specifically — no mention of a third-party cloud provider. Anthropic described its incident only as an infrastructure issue affecting Claude.ai, Claude Code, Claude Cowork, and the API, without naming a cause. Later the same day, SpaceX posted an apology for an outage at its Memphis compute center that morning, which lines up with the timing of Grok's problems. None of the three statements names Microsoft Azure, despite that becoming the most repeated explanation online.

Was Azure Actually the Cause?

This is the part worth being skeptical about. The Azure theory spread because OpenAI, Anthropic, and xAI all route meaningful production traffic through Azure infrastructure in various capacities, and Downdetector and StatusGator both picked up reports suggesting Azure's East US region was having ingress problems around the same time. But Microsoft's own Azure status page showed no incident that day, and neither OpenAI nor Anthropic named Azure in their statements. The Register also asked Cloudflare directly, since all three AI companies use it in different capacities — a spokesperson said flatly that Cloudflare's systems were running fine and that reports suggesting otherwise were incorrect. AWS and Google Cloud's status pages were likewise clean. In short: the shared-cloud-provider theory is plausible given how these companies are architected, but it isn't confirmed by any of the parties involved.

What's the SpaceX Connection?

This one has real substance behind it. Anthropic announced in May 2026 that it had signed an agreement to use all of the compute capacity at SpaceX's Colossus 1 data center, adding more than 300 megawatts and over 220,000 Nvidia GPUs to its infrastructure. That's a confirmed, on-the-record relationship between Anthropic and SpaceX compute. When SpaceX apologized for a Memphis compute center outage on the same morning Grok went down, it created an obvious hypothesis: if Claude and Grok both draw on infrastructure tied to the same physical location, a Memphis-area failure could plausibly touch both. Anthropic itself hasn't confirmed that connection for this specific incident, so treat it as a documented possibility rather than an established fact.

What Does This Mean for Multi-Cloud Reliance?

If you're building products on top of ChatGPT, Claude, or Grok's APIs, the practical lesson isn't "it was Azure" or "it was SpaceX" — it's that even well-resourced AI providers still have single points of failure somewhere in their stack, and multiple providers can go down in the same window without a shared cause. Falling back from one AI API to another during an outage doesn't guarantee resilience if the underlying infrastructure overlaps in ways that aren't visible from the outside. For anything latency- or availability-sensitive, that argues for building real fallback logic across genuinely independent providers and testing it before you need it, not assuming vendor diversity you haven't actually verified.

Wrapping Up

Three major AI providers went down within 90 minutes of each other, and eleven days later there's still no confirmed shared cause — just three separate incident reports and a plausible SpaceX-linked hypothesis for two of them. If your product depends on any single AI provider's uptime, this is a good prompt to test your actual failover path rather than assume it works.

Sources

The Register: original reporting with direct statements from OpenAI, Anthropic, and Cloudflare

Anthropic: official announcement of the SpaceX Colossus 1 compute agreement

Computing: coverage connecting Downdetector and StatusGator reports to Azure's East US region

Le VPN: timeline analysis questioning the single-cause narrative