OpenAI, Anthropic and xAI Hit by Rare Same-Day Outages as Cloud Causes Remain Unclear

OpenAI, Anthropic and xAI all suffered rare service disruptions on Thursday morning, briefly knocking ChatGPT, Claude and Grok offline for some users and raising immediate questions about whether a shared infrastructure problem was to blame. By afternoon, each company had restored service, but none had publicly identified a common root cause.

The coincidence matters because these three companies sit at the center of the current AI boom. When their flagship chatbots go dark at roughly the same time, it exposes how dependent the industry has become on complex cloud, routing and compute systems that are mostly invisible to users until something breaks.

What happened during Thursday’s AI outages?

All three companies reported problems within a narrow window on Thursday morning, though the incidents did not line up perfectly and the severity varied by platform. OpenAI said ChatGPT and Codex were unavailable for some users after a routing error. Anthropic described a partial outage affecting several Claude models. xAI said Grok was experiencing issues across its services and linked the disruption to a problem at its Memphis compute center.

The timing made the outages look connected, at least at first glance. But the companies did not offer evidence that a single outside vendor or internet backbone failure was responsible. Instead, the available statements pointed to separate explanations, or at minimum to explanations that were still being investigated individually.

How did OpenAI explain the ChatGPT outage?

OpenAI said the problem was caused by a routing error that began around 7:43 a.m. Pacific time on Thursday, September 3. In a statement to WIRED, spokesperson Kathleen Chaykowski said the issue made ChatGPT and Codex inaccessible for some users across platforms.

The company said the problem had been fixed by about 8:17 a.m. PT and that the solution was still being monitored afterward. That kind of language suggests the incident was short-lived but not trivial: a routing problem can break access even when the underlying models and servers are still running.

OpenAI said a routing error beginning at 7:43 a.m. PT made ChatGPT and Codex unavailable for some users, and that a fix was in place by about 8:17 a.m. PT.

OpenAI did not say in its comment that a third-party provider was involved. It also did not publicly tie the incident to Anthropic’s or xAI’s outages, despite the overlap in timing.

Why does Anthropic’s outage stand out?

Anthropic’s incident was notable because it started earlier than OpenAI’s and affected multiple Claude model variants. The company began warning users about a partial outage at 6:23 a.m. PT and said requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 were producing elevated errors.

Anthropic later said it had identified the cause and deployed a fix, then marked the issue resolved at 9:16 a.m. PT. The company did not provide more detail to WIRED and declined to comment further on the episode.

One additional wrinkle was a brief period after 9 a.m. PT when Claude Sonnet 5 appeared to encounter similar problems. That suggests either lingering instability during recovery or a separate short-lived disruption in the same general time frame.

What users saw

For users, the practical effect was simple: requests failed, responses stalled, or the service became unreachable. For a product like Claude, which is often used through both consumer chat interfaces and developer-facing APIs, even a short outage can interrupt work across very different workflows.

Anthropic’s status updates implied the company moved quickly once it spotted the error spike. Still, the lack of a full explanation leaves open the question of whether the incident originated in Anthropic’s own systems or in a broader layer of cloud infrastructure.

What did xAI say about Grok going down?

xAI described a broader platform disruption that affected Grok across its services, beginning at 6:30 a.m. PT when it posted that it was investigating an outage. The company later said it was working to restore service as quickly as possible.

By 10:05 a.m. PT, xAI marked the incident complete and said traffic had returned to normal. In public comments, SpaceX — xAI’s parent company — said the issue with Grok stemmed from “an outage at our Memphis compute center this morning.”

SpaceX said the Grok disruption came from an outage at xAI’s Memphis compute center and apologized to impacted compute partners.

That phrasing is important because it points to an internal compute site rather than a general internet outage. It also arrives after xAI and Anthropic announced a compute partnership with SpaceX in May, which makes the mention of impacted partners especially relevant, even if the companies have not publicly spelled out the technical chain of events.

How likely was a shared cloud outage?

At first, a simultaneous outage across rival AI companies would naturally suggest a shared third-party provider, a cloud region problem or a content delivery network failure. That is often the most plausible explanation when multiple services in the same sector fail at once.

But Thursday’s episode resisted that easy answer. OpenAI and Anthropic did not point to an outside vendor in their comments to WIRED, and major infrastructure companies did not show public signs of simultaneous trouble.

• Cloudflare did not report a matching incident.
• Amazon Web Services did not report an outage.
• Microsoft Azure did not report an outage.

That does not prove the companies were operating on wholly separate stacks. AI platforms can still depend on overlapping internet routes, data pipelines, compute vendors and secondary services without all of them publicly reporting problems. But based on what was disclosed, there is no confirmed evidence of a single shared outage source.

Why the timing created so much speculation

The speculation was fueled by the narrow window in which the outages emerged. Anthropic started flagging issues shortly after 6:20 a.m. PT, xAI followed at 6:30 a.m. PT, and OpenAI’s problem appeared around 7:43 a.m. PT. To users, that sequence looks suspiciously coordinated.

In practice, internet failures often cluster in ways that are hard to interpret in real time. A noisy error in one layer can ripple into another, and companies may discover independent failures at nearly the same moment because traffic spikes or user reports rise together.

What this says about the AI industry’s infrastructure

Thursday’s outages are a reminder that the AI race is not only about models, benchmarks and product launches. It is also about the physical and logical infrastructure that keeps those systems reachable: routing, compute, load balancing, redundancy and vendor management.

In a sector where companies market near-constant availability and instant responses, reliability has become part of the product itself. A major chatbot outage does not merely inconvenience users; it interrupts coding, research, customer support, content production and internal enterprise workflows that now depend on AI tools functioning like ordinary software infrastructure.

The episode also highlights how opaque modern AI systems can be. Users see a branded interface — ChatGPT, Claude or Grok — but the failure could sit anywhere underneath: in the model layer, the API gateway, the data center, the network path, or a partner’s compute site. Without more detailed disclosures, it can be difficult to know where responsibility begins and ends.

How rare are outages like this?

These kinds of failures are uncommon enough to be newsworthy, but not unheard of. Large AI services are now scaled to serve millions of users and developers, which makes perfect uptime difficult. The more deeply these products integrate into daily work, the more visible even brief interruptions become.

For OpenAI, Anthropic and xAI, Thursday’s event was unusual not just because each company had a problem, but because all three ran into trouble within the same morning. That rarity is what made the incident stand out across the AI sector and beyond.

It is also worth noting that Google Gemini was the subject of scattered user reports that morning, though Google did not confirm a problem on its status dashboard and did not respond to WIRED’s request for comment before publication. Without an official confirmation, those reports remain anecdotal.

Timeline: Thursday morning’s AI service disruptions

The incidents unfolded over several hours, with each company restoring service on its own schedule. The table below shows the basic sequence as publicly reported.

Why companies were careful in their public statements

The limited detail from OpenAI and Anthropic likely reflects the caution companies often use when a problem is still being investigated. If a root cause is not confirmed, public statements tend to focus on impact, timing and remediation rather than speculation.

That caution is especially important in AI, where outages can be wrongly assumed to stem from model failures when the real issue may lie in routing, authentication, compute or load distribution. Companies also have incentives to avoid over-attributing blame to third parties before internal checks are complete.

xAI’s explanation was somewhat more specific because it pointed to the Memphis compute center, but it still did not spell out whether the issue was hardware-related, network-related or the result of a broader facility problem. SpaceX’s apology to compute partners suggests a wider operational impact, but not a complete public diagnosis.

What users and developers should watch next

For everyday users, the practical lesson is that even the largest AI chatbots can fail in brief but disruptive bursts. For developers, the lesson is more technical: resilience planning matters, and reliance on a single model endpoint or provider can turn a short outage into a business interruption.

Going forward, the most important question is whether any of the companies will publish a fuller postmortem. A transparent explanation could help distinguish between a one-off operational issue and a sign of larger fragility in the AI infrastructure stack.

The broader stakes for AI reliability

Reliability is quickly becoming a competitive feature in artificial intelligence. Customers increasingly judge model quality not just by output quality, but by uptime, latency and predictability. If a chatbot is smarter but unstable, users may still prefer a less capable but more dependable alternative.

That pressure is especially intense for companies positioning their models as essential tools for coding, office work and agentic workflows. When those systems go offline, the failure is not just technical — it affects trust, adoption and the perception of maturity across the entire sector.

Thursday’s multi-company outage did not produce a single confirmed explanation, and that uncertainty is itself revealing. The AI industry now runs on a stack so complex that even a cluster of outages can be difficult to trace publicly in real time. The result is a reminder that for all the advances in frontier models, the basic challenge of keeping them online remains a central part of the race.

Bottom line

OpenAI, Anthropic and xAI all experienced service disruptions on Thursday morning, temporarily taking ChatGPT, Claude and Grok offline for some users. OpenAI blamed a routing error, Anthropic said it found and fixed an issue affecting several Claude models, and xAI said the trouble came from its Memphis compute center, but none of the companies confirmed a shared cause.