For a couple of hours on Thursday, a lot of people discovered how much of their day now runs through an AI chatbot — by watching three of them fail at once.

ChatGPT, Claude and Grok all went down in overlapping outages, confirmed on each company's own status page. OpenAI logged problems across ChatGPT and its Codex coding tool; Anthropic's status page reported "elevated errors" on Claude; Grok's site flagged issues too. DownDetector logged more than 12,000 reports early on and climbing sharply through the day, with around 80% of the OpenAI complaints pointing at ChatGPT. By late afternoon UK time recovery was partial and uneven: OpenAI had applied mitigations, Anthropic's smaller models were back but its Opus 4.8 and Opus 5 remained degraded, and Grok was still investigating with no confirmed fix time.

The interesting part is not the downtime

Outages happen. What makes this one worth a moment's attention is that three rival products, from three fiercely competitive companies, failed in the same window — and that is not supposed to happen to competitors unless they share something underneath.

The suspected something is cloud infrastructure. The leading AI labs do not run their own data centres end to end; they rent capacity from a small number of hyperscale clouds and route traffic through a smaller number of content-delivery networks. Early speculation pointed at a shared dependency — an Azure-level wobble that would take down anything sitting on top of it, regardless of whose logo is on the front end. None of the companies has confirmed a root cause, so treat the specific culprit as unproven. But the shape of the failure — competitors down together — only makes sense if the layer beneath them is shared.

Concentration is the risk nobody prices in

This is the part that outlasts today's outage. The AI industry looks diverse at the top — OpenAI, Anthropic, xAI, Google, all racing each other — and is far more concentrated at the bottom, where a handful of clouds carry almost all of it. As these tools stop being novelties and start being load-bearing — drafting the email, writing the code, triaging the tickets — a correlated outage stops being an inconvenience and becomes the kind of single-point failure that used to be reserved for a bank's core systems or a payments network.

For UK businesses that have quietly wired ChatGPT or Claude into real workflows over the past year, Thursday was a cheap lesson in a real dependency: the fallback for "the AI is down" cannot be "use a different AI" if they are all sitting on the same foundation. Google's Gemini — which runs on Google's own cloud — had a murkier day: some reports had it largely unaffected, others logged a spike in Gemini complaints too, and Google did not confirm an outage. Either way the lesson lands: "which cloud" is quietly becoming as strategic a question as "which model."

The chatbots are coming back. The dependency they exposed is not going anywhere — and the next shared-infrastructure stumble will land on an industry that leans on these tools more than it did this morning.