ANALYSIS When OpenAI, Anthropic, xAI, and Google all suffered service interruptions within the same Thursday-morning window, the episode did more than inconvenience users: it forced a practical question about whether the industry's largest model providers share enough underlying infrastructure to fail in concert.
Cloud-based AI models operated by OpenAI, Anthropic, xAI, and Google suffered "a rare and overlapping set of significant service interruptions over a period of hours Thursday morning"1. Axios noted that while "outages are common," it is "unusual for all three to go down at once, especially as the world increasingly relies on AI models"4.
Why it matters
Enterprises, developers, and consumers now treat frontier AI models as always-on utilities. The September 3 incident showed that diversifying across providers may not protect against correlated downtime. ChatGPT, Claude, and Grok are built by separate companies, yet all three were simultaneously unavailable according to DownDetector. There have also been reported issues with Google's Gemini, though Google has not issued a confirmed outage notice.
The big picture
The timeline reveals how quickly the failures cascaded. Anthropic first reported a "partial outage" related to "elevated errors on requests to Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5" at 9:23 AM ET. By roughly 9:41 AM ET, Anthropic said it had "identified the cause" of the elevated errors and was working on a fix5. The affected model list soon expanded to include Mythos/Fable 5.1, Mythos/Fable 5, Opus 5, Opus 4.8, and Opus 4.6.
OpenAI's troubles surfaced slightly later. The company reported "elevated errors across ChatGPT and Codex" resulting in "degraded performance" as of 10:43 AM. The outage affected not just chat but logins, file uploads, voice mode, search, deep research, and image generation2. OpenAI said it had "applied a mitigation" and was "monitoring recovery," though its tools were still experiencing "degraded performance" even after the initial fix. The issue was marked "resolved" by 12:55 PM.
Anthropic deployed its own fix and reported the issue resolved by 12:16 PM, though a separate incident flagged "elevated errors on requests to Claude Sonnet 5" for a brief period just after noon. xAI's Grok was also down during the same window3.
ANALYSIS None of the companies publicly explained a root cause. The packet contains no statement from any provider attributing the outage to a specific shared dependency, a demand spike, or a common cloud-infrastructure failure. That silence is itself notable: Anthropic's status-page language moved from "investigating" to "identified the cause" to "fix deployed" without ever naming what broke. OpenAI's communications followed a similarly opaque pattern, limited to "applied a mitigation".
The near-simultaneous timing across four independent companies invites speculation about shared upstream infrastructure, whether cloud-compute providers, networking layers, or power grids. But the evidence in hand supports only the observation of temporal correlation, not causation. Anthropic reported its initial fix deployed by 12:16 PM. OpenAI's issue was marked resolved by 12:55 PM. ◆ Both resolutions fell within the same midday window, reinforcing the pattern of correlated behavior without explaining it.
The breadth of Anthropic's affected model list is worth noting. The outage touched six distinct model variants spanning three generations: Opus 4.6, Opus 4.8, Opus 5, Mythos/Fable 5, and Mythos/Fable 5.1. That breadth suggests the failure sat below the model layer, in serving infrastructure rather than in any single model's deployment.
The incident is likely to accelerate enterprise conversations about multi-provider redundancy and on-premise inference as a hedge. For the providers themselves, the open question is whether any post-mortem will address the coincidence publicly. As of the latest reports, Google had not issued a confirmed outage notice for Gemini despite user-reported issues, and no company had disclosed a shared technical root cause. A 9to5Google report confirmed the spike in DownDetector reports across all three platforms. Until the providers explain what happened, the correlation stands as an uncomfortable data point for anyone treating model diversity as a resilience strategy.