OpenAI, Anthropic, and xAI All Had Outages on the Same Morning

Three of the biggest AI providers, OpenAI, Anthropic, and xAI, all experienced service disruptions on the morning of Thursday, September 3. The timing raised immediate questions about a shared cause, but OpenAI cited only an internal routing error while Anthropic declined to comment on the cause entirely. xAI's parent company SpaceX pointed to an outage at its Memphis compute center. No major infrastructure provider, including Cloudflare, AWS, or Microsoft Azure, reported any issues that morning.

What happened

Company	Start time (PT)	Resolved (PT)	Official cause
Anthropic	6:23 am	9:16 am	Declined to say
xAI / Grok	6:30 am	10:05 am	Memphis compute center outage
OpenAI	7:43 am	8:17 am	Routing error

On the morning of September 3, users of ChatGPT, Claude, and Grok all ran into errors within roughly the same two-hour window. OpenAI spokesperson Kathleen Chaykowski told WIRED that a routing error starting at 7:43 am PT made ChatGPT and Codex unavailable for some users, and that a fix was in place by 8:17 am PT.

Anthropic began posting alerts about a “partial outage” at 6:23 am PT. The affected models were Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5. The company said it had identified the cause and deployed a fix, marking the incident resolved by 9:16 am PT. Claude Sonnet 5 also showed brief errors shortly after 9 am PT. Anthropic did not provide WIRED with any explanation of the root cause.

xAI posted “investigating outage” on its status page at 6:30 am PT, noting that Grok was having issues across all platforms and services. SpaceX, which owns xAI, later stated publicly that the disruption was caused by “an outage at our Memphis compute center.” The company also said: “We’d also like to apologize to our impacted compute partners.” The incident was marked resolved at 10:05 am PT.

There were scattered user reports of Google Gemini issues that same morning, but Google neither confirmed an outage nor logged any incidents on its status dashboard. Google did not respond to WIRED before publication.

Why does the timing raise questions?

When multiple services in the same sector go down at the same time, the usual suspects are shared infrastructure: a cloud provider like AWS or Azure, a content delivery network like Cloudflare, or another third-party vendor sitting underneath multiple products. None of those reported problems on September 3.

There is one potential thread. Anthropic and xAI announced a compute partnership with SpaceX back in May. SpaceX’s public apology to “impacted compute partners” during the xAI outage is the closest thing to a shared explanation, but it is far from a confirmed link. OpenAI has not indicated any connection to SpaceX’s Memphis facility, and Anthropic said nothing at all about a cause.

The result is an unusual situation: three coinciding outages, one partial explanation from xAI, and two companies staying quiet. Whether the incidents were related remains publicly unanswered.

Why it matters

Businesses and developers running production workloads on any of these models had a rough morning. A roughly three-and-a-half hour window (6:23 am to 10:05 am PT) covered the start of the US workday on the East Coast. Any team with automations, customer-facing chatbots, or AI-integrated workflows hitting these APIs would have seen failures with no clear ETA on a fix.

The lack of transparency is also notable. Infrastructure outages happen; what frustrates operators is when providers go quiet on root cause. Anthropic’s silence on the why, after a nearly three-hour incident affecting its flagship models, leaves teams with no information to improve their own resilience planning.

The SpaceX connection is worth watching. If a single compute partner underpins both xAI and portions of Anthropic’s infrastructure, a future failure at that facility could take down both at once again. Diversification of AI API dependencies is not just a theoretical concern after a morning like this.

Our take

We have seen this pattern before in other sectors: simultaneous outages, vague status pages, and post-incident silence. The AI infrastructure layer is maturing fast, but the operational transparency is still catching up. OpenAI gave a precise timeline and a named cause. That is the baseline standard. Anthropic’s non-answer is a red flag for any team treating Claude as a critical dependency.

If you are building anything that touches these APIs, the morning of September 3 is a good reason to audit your fallback logic now. Do you have a secondary model or provider that kicks in if your primary API returns errors? Do your automations fail gracefully rather than silently dropping data? These are solvable problems, and a brief outage is a cheap lesson compared to a production failure.

For teams keeping an eye on broader AI infrastructure trends, our AI news coverage tracks these developments as they happen.

What to do about it

1. Check your API error handling: confirm your code catches 5xx responses from AI providers and surfaces them clearly rather than silently failing.
2. Add a fallback model: if your workflow depends on one provider, configure a secondary (for example, route to GPT-4o if Claude returns errors, or vice versa).
3. Subscribe to status pages: bookmark status.anthropic.com, status.openai.com, and xAI’s status page so you get alerts before users do.
4. Test your fallback: run a fire drill by temporarily blocking your primary API endpoint and confirm the fallback actually activates end to end.

A three-hour outage affecting three providers on the same morning is a reminder that “cloud” infrastructure can still go down in ways nobody fully explains afterward.

Source: WIRED · AI