Nobody Is Saying Why OpenAI and Anthropic Had Outages Today

Lily Hay Newman reports that frontier models from Anthropic, OpenAI, and xAI all saw rare outages on Thursday morning, taking down their chatbots in overlapping windows. SpaceX, xAI’s parent, said Grok’s issues came from "an outage at our Memphis compute center this morning" and apologized to impacted compute partners. Anthropic and xAI had announced a compute partnership with SpaceX in May.

OpenAI’s Kathleen Chaykowski said a routing error from about 7:43 am PT made ChatGPT and Codex unavailable for some users; a fix was in place around 8:17 am PT and still being monitored. Anthropic declined to comment beyond its status page: a partial outage alert at 6:23 am PT covering elevated errors on Claude Mythos 5.1, Fable 5.1, and Opus 5, with a fix deployed and resolution by 9:16 am PT; Sonnet 5 briefly had similar issues after 9 am. xAI marked Grok issues from 6:30 am PT through 10:05 am PT. Cloudflare, AWS, and Azure did not report outages; OpenAI and Anthropic did not cite an external shared cause.

Signal Desk’s read: coincidence is still on the table, and silence is doing real work. The classic shared-vendor signature — simultaneous multi-customer pain plus a CDN or cloud status incident — did not show up. That weakens the "one third party took everyone down" story. It does not prove independence. It only proves nobody with a microphone chose to name a shared dependency.

xAI’s Memphis compute explanation is the only causal narrative in public. The apology to "impacted compute partners" plus the May Anthropic–xAI–SpaceX compute partnership invites readers to wonder whether Anthropic’s morning sat on related capacity. Anthropic’s "identified the cause" status language without a public cause is corporate hygiene, not transparency. OpenAI’s routing-error account is specific enough to sound like an internal mistake — which, if accurate, makes the timing overlap with Grok and Claude look like bad luck rather than a common blast radius.

The industry lesson is not "find the smoking CDN." It is that three consumer-facing frontier surfaces can degrade in the same business morning without a joint postmortem culture. Users experience correlation even when engineering root causes differ. Enterprise buyers will treat this Thursday as a multi-vendor reliability sample: ask what failed, what was shared, and how fast each lab said so. Scattered Gemini reports without a Google confirmation are noise until a status page moves.

Who gains from clearer disclosure: buyers and anyone writing multi-model failover into production. Who gains from murk: labs that would rather not map their capacity partners in public. Expect more mornings like this as chat becomes infrastructure and compute partnerships thicken.

Overlapping AI chatbot outages have happened before and are often unrelated. What is new is the expectation that ChatGPT, Claude, and Grok stay up like email — and the thinness of public root-cause detail when they do not.