The Morning After Anthropic Filed to Go Public, Claude Went Dark for Everyone

On Monday, Anthropic filed the paperwork to go public. On Tuesday morning, Claude went dark for everyone on earth.

The global outage began at 06:04 UTC on June 2, according to Anthropic's status page. Within the hour the company had identified the problem and was deploying a fix. By 10:42 UTC it was monitoring the results. Four hours of silence, across every surface the company operates: claude.ai, the API that developers embed in their software, the Claude Code coding assistant, and the enterprise console. According to Downdetector, roughly 62% of users were completely locked out of the main website interface. The Independent documented the user experience: the chatbot page loaded normally, but any attempt to use it returned only an error message — "a bit longer, thanks for your patience" — and the system refused to respond to questions. Developers using Claude Code as a primary coding assistant lost their workflow tool mid-build. StatusGator recorded reports from India, Germany, Austria, Ireland, Indonesia, Pakistan.

Anthropic confirmed the incident and said it was working on a fix. That is the entirety of what the company has said publicly. No postmortem, no root cause, no explanation of what failed and why. The status page still says the incident is unresolved.

The timing is awkward in the way that only a $965 billion valuation can be. Anthropic announced in May that its revenue run rate had reached $47 billion, up from $10 billion the prior year. It closed a $65 billion Series H funding round days before the filing, valuing the company above OpenAI for the first time. The S-1 dropped June 1. By the time markets opened the following morning, the AI was offline.

Whether that sequence will be front of mind for institutional investors and underwriters reading the prospectus risk section remains to be seen — and it is the question the IPO roadshow will answer. The comparison to early cloud computing is instructive: AWS's April 2011 outage — which knocked out Reddit, Quora, and hundreds of other sites for hours — taught the industry that enterprise customers would demand SLA credits and multi-cloud contracts before signing blank-check infrastructure commitments. Whether AI infrastructure buyers draw the same lesson faster or give the industry the same benefit of the doubt is an open question. No regulatory body has yet issued formal guidance treating AI providers as critical infrastructure, but the structural logic is identical, and the pace at which regulators are catching up to AI's infrastructure role has not kept up with the pace at which enterprises integrated it.

The broader backdrop does not help. According to tracking data compiled from Anthropic's own status page by independent analysts, Claude experienced approximately 50 reported incidents in recent months, with 24 classified as major or critical. March 2026 alone accounted for the majority. One outage that month lasted nearly five hours. Another stretched across 19 hours of intermittent degradation. The company's 90-day uptime statistic sits at 98.87 percent for the main interface, which sounds fine until you consider that 1.13 percent of a service used by millions of developers and enterprises is a lot of broken workflows.

This is the infrastructure maturity problem the AI industry has been quietly deferring. As with early electrical infrastructure, every major technology transition involves a period where new systems become load-bearing before anyone has built the redundancy to match. Early internet backbone providers ran single Points of Failure that brought down half the network when a router hiccupped. AI is further along that learning curve than the industry's marketing suggests, and not in the direction investors want. Claude's simultaneous failure across all surfaces, on the morning after an S-1 filing, reveals a concentration risk that the reliability numbers have been quietly documenting. When a single provider going down means your coding assistant, your API integration, your enterprise chat tool, and your mobile app all stop working at the same time, that is an architectural fact, not an incident report.

What Anthropic says next will matter. The postmortem, if it comes, will show whether this was a novel failure or another iteration of the same capacity scaling problem the company has been wrestling with since March. The S-1 risk disclosures will show what the company told the SEC about this pattern. The IPO roadshow will show whether institutional investors apply the same tolerance they showed AWS after 2011.

For now, the status page is the record. It says the incident is under monitoring. It does not say why it happened.