OpenAI, Anthropic and xAI experienced overlapping service disruptions on September 3, temporarily degrading several of the most widely used hosted AI platforms. The providers have not identified a common cause, and available status information does not establish that the incidents resulted from a shared cloud or network failure.

Grok was the earliest confirmed failure. xAI’s status pages reported model outages beginning at 13:30 UTC across Grok’s web service, API, mobile applications and integrations, saying it was investigating and working to restore service.

Anthropic reported elevated errors beginning at about 13:23 UTC for several Claude models. The company later said it had identified the cause and deployed a fix. An Anthropic spokesperson described the disruption as an “infrastructure issue” affecting Claude.ai, Claude Code, Claude Cowork and the Claude API, without identifying an external infrastructure provider. Most services subsequently recovered, although some models remained affected for longer.

OpenAI’s problems followed later in the overlapping window. Its status service reported elevated errors affecting ChatGPT and Codex shortly before 15:00 UTC and subsequently said a mitigation had been applied while it monitored recovery. The incident affected multiple ChatGPT and Codex components.

There were also user reports of problems with Google’s Gemini during roughly the same period, prompting some reports to describe a four-provider outage. Google had not publicly acknowledged a Gemini incident, however, and its Cloud Service Health history showed no corresponding September 3 Gemini event. The Gemini disruption therefore remains less firmly established than the incidents acknowledged by OpenAI, Anthropic and xAI.

No shared upstream failure has been demonstrated. Microsoft Azure, Amazon Web Services and Cloudflare had not reported major concurrent incidents when Ars Technica checked their public status systems, and Microsoft’s public Azure dashboard showed no broad active event. That leaves open the possibility that the timing was coincidental or that a dependency not reflected on public status pages was involved.

The overlap is nevertheless notable as enterprises increasingly depend on remotely operated AI models through web services, APIs and coding tools. Separate failures occurring within the same window can produce much the same operational problem as a shared infrastructure outage for organizations whose fallback strategy consists largely of switching between a small number of external AI providers.