Analysis
ChatGPT, Claude and Grok all experienced outages within the same window on the morning of Sept. 3, an unusual simultaneous failure across three separate companies' flagship AI products, Axios reported. A routing error beginning around 7:43am PT made ChatGPT and Codex unavailable for some users across platforms, according to OpenAI. Claude's own status dashboard showed elevated errors specifically on its Opus 4.8 and Opus 5 models, with other Claude models reported at baseline error rates. xAI confirmed Grok was experiencing issues and said it was working on the problem.
One Shared Cause, or Three Coincidental Ones
The most interesting open question in this incident is whether it had one root cause or three. Microsoft Azure, which provides cloud infrastructure underlying all three companies to varying degrees and which Pulse has covered in prior Azure infrastructure reporting, was also reporting its own outages during the same window, and Ars Technica's own reporting noted Gemini appeared impaired too, with several tech outlets suggesting Azure's problem may have contributed to some or all of the AI outages. Cloudflare, by contrast, said its systems were running normally and not experiencing significant disruptions -- which rules out one commonly shared piece of internet infrastructure without confirming Azure as the definitive single cause across all three companies. OpenAI, Anthropic and xAI each described their own outage independently rather than jointly attributing it to a shared vendor, which is either appropriately cautious incident communication or evidence the companies genuinely don't yet know whether their failures shared a cause.
“## One Shared Cause, or Three Coincidental Ones The most interesting open question in this incident is whether it had one root cause or three.”
Why Simultaneous Outages Matter More Than Individual Ones
Individual AI-product outages have become routine enough that they rarely make news on their own -- every major lab has had multi-hour downtime at some point in 2026. What makes this incident notable is the correlation: as more of the economy routes through a small number of underlying cloud providers, an outage at one infrastructure layer can simultaneously take down products that look, from a user's perspective, like entirely independent competitors. ChatGPT, Claude and Grok compete aggressively against each other on model quality and pricing, but if a meaningful share of their reliability depends on the same handful of hyperscale cloud providers, that competitive independence is thinner at the infrastructure layer than it appears at the product layer.
What Users Actually Experienced
Decrypt and other outlets covering the incident in real time captured a specific, telling reaction: users across social platforms asking how to work without AI tools they'd apparently already built daily workflows around, in the space of a single weekday morning. That's a meaningful signal on its own about how quickly AI chat tools have moved from novelty to load-bearing infrastructure for a segment of knowledge workers -- a multi-hour outage across all three major providers simultaneously is now disruptive enough to generate its own news cycle independent of any single company's downtime record.
The Counterweight
All three services were back to normal by midday, a recovery time that's genuinely fast for a multi-company, cross-provider incident, and none of the companies has reported data loss or security compromise associated with the outage -- this was an availability failure, not a safety or integrity one. It's also worth noting that simultaneous outages across competing infrastructure happen periodically across every mature internet sector, from cloud hosting to CDNs, and AI products reaching that level of infrastructure maturity is arguably itself a sign of how mainstream the category has become rather than a unique AI-specific fragility.
What's still unconfirmed: none of OpenAI, Anthropic or xAI has published a formal post-incident report attributing a specific root cause, and until one does, whether this was genuinely one shared failure or three coincidental ones on the same morning remains an open question worth revisiting once postmortems land.
