Three major artificial-intelligence services suffered disruptions during overlapping hours on September 3, briefly making ChatGPT, Claude and Grok unavailable or unreliable for some users. The timing raised an obvious question about shared infrastructure, but the companies' public explanations did not identify a common failure.
OpenAI said a routing error began at about 7:43 a.m. Pacific time and affected ChatGPT and Codex across platforms. A company spokesperson told Wired that a solution was implemented around 8:17 a.m. and continued to be monitored. That explanation described an internal networking problem rather than a failure at a named cloud or content-delivery provider.
Anthropic began reporting a partial outage at 6:23 a.m. Pacific, citing elevated request errors involving Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5. The company later said it had identified the cause and deployed a fix, marking the incident resolved by 9:16 a.m. A separate brief issue appeared to affect Claude Sonnet 5 shortly after 9 a.m., according to the report. Anthropic declined to offer Wired additional comment on the episode.
XAI's status page began describing a Grok outage at 6:30 a.m. Pacific across its platforms and services. The incident was marked complete at 10:05 a.m., when the company said traffic was healthy again. SpaceX, xAI's parent, later attributed Grok's trouble to an outage at its Memphis compute center and apologized to affected compute partners.
The close timing initially made a shared supplier appear plausible. Large AI systems depend on overlapping layers of cloud computing, networking, authentication and content delivery, and a problem in one widely used provider can affect otherwise separate products. However, OpenAI and Anthropic did not point to an outside dependency. Major infrastructure companies including Cloudflare, Amazon Web Services and Microsoft Azure did not report corresponding incidents, Wired said.
There were also scattered reports of possible trouble with Google's Gemini service, but Google did not confirm an incident and its public status dashboard did not record one. That makes it difficult to distinguish a genuine related outage from the background level of user reports that accompanies any high-profile service disruption.
The available evidence therefore supports several concurrent incidents, not one demonstrated industry-wide failure. OpenAI identified routing, xAI identified a compute-center outage, and Anthropic disclosed only that it found a cause and applied a fix. The fact that the companies went down in a similar window remains notable, but coincidence alone does not establish shared infrastructure or coordinated activity.
For businesses that increasingly place AI assistants inside development, customer-service and research workflows, the episode is a reminder that model availability is an operational dependency. Redundancy across vendors may help, but only if those vendors do not rely on the same hidden services. Without fuller incident reports, customers cannot yet tell how independent these systems really are.
