Anthropic, OpenAI and xAI experienced overlapping service failures on Thursday morning, briefly making several prominent AI products unavailable or unreliable. The timing created an obvious suspicion that the incidents shared an infrastructure provider, but the information disclosed by the companies does not establish a common cause.
SpaceX attributed Grok’s problems to an outage at its Memphis compute center. OpenAI identified a routing error. Anthropic said it found and fixed the source of its partial outage but declined to give Wired more detail. The episode is therefore a cluster of confirmed disruptions with separate public explanations, not evidence of one industry-wide failure.
Three incidents overlap in the morning
Anthropic began reporting elevated errors at 6:23 am Pacific time. Its notice covered Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5, while Claude Sonnet 5 appeared to encounter similar trouble briefly after 9 am. Anthropic said a fix had been deployed and marked the incident resolved by 9:16 am.
xAI opened its own incident at 6:30 am, saying Grok was having problems across its platforms and services. It declared traffic healthy again at 10:05 am. SpaceX later said the cause was a morning outage at its Memphis compute center and apologized to affected compute partners.
OpenAI’s interruption was shorter. A company spokesperson said a routing error beginning at 7:43 am made ChatGPT and Codex unavailable to some users across platforms. OpenAI implemented a solution at about 8:17 am and continued monitoring it. The three timelines overlap, but they do not begin or end together.
A shared cause remains unverified
Simultaneous problems among competitors can indicate failure at a cloud company, content-delivery network or another supplier used by multiple services. That possibility was particularly notable because Anthropic and xAI announced a compute partnership with SpaceX in May, and SpaceX referred to affected compute partners in its apology.
Even so, neither OpenAI nor Anthropic cited a third-party provider in their comments to Wired. SpaceX did not answer the publication’s request for additional comment. Major infrastructure companies including Cloudflare, Amazon Web Services and Microsoft Azure did not report outages that morning.
Those facts leave several interpretations open. The services may have experienced independent failures that happened to coincide, or a relationship may exist that none of the disclosures identifies. The source provides no technical evidence that can distinguish between those possibilities. Treating the overlap itself as proof would go beyond what the companies confirmed.
Public status reports set the limit
The available account depends heavily on provider status pages and short company statements. Those records are useful for establishing when users saw errors, which products were affected and when operators declared recovery. They offer much less visibility into architecture, dependencies or why separate systems failed during the same period.
There were also scattered reports of possible disruption to Google’s Gemini service. Google did not confirm an incident, recorded nothing on its status dashboard and did not respond to Wired before publication. Without a verified company record or another exact technical account, Gemini cannot be added to the confirmed group.
All three documented incidents were resolved within the same morning. OpenAI’s stated interruption lasted about 34 minutes, Anthropic’s status window just under three hours, and xAI’s approximately three hours and 35 minutes. Those durations come from the posted times and do not measure every individual user’s experience.
The outages demonstrate how quickly coincident failures can produce a compelling infrastructure theory. They also show the limits of that inference when providers reveal only operational summaries. For customers, the practical event was real: multiple frontier AI services became less dependable at once. For investigators, the shared explanation remains absent, and the responsible conclusion is to keep the confirmed timelines separate until evidence connects them.
