SAN FRANCISCO — For about half an hour on Sunday, one of the most valuable private companies in technology could not get its own product to work. Anthropic’s Claude service suffered a widespread outage that left users unable to log in and disrupted multiple services, according to TechCrunch. Service was restored after roughly 36 minutes.
The incident, Anthropic’s second service disruption in recent weeks, landed at the worst possible moment: the company is preparing for an October IPO that could value it at $2 trillion, and its credibility with enterprise customers rests on the reliability of a product those customers now run core operations on.
Anthropic disclosed on Aug. 15 that second-quarter revenue exceeded $11.5 billion, roughly 14 times the year-ago figure, with the company’s first positive adjusted operating income. The growth has been driven in large part by enterprise contracts — companies that pay for Claude’s coding, customer-service and data-analysis capabilities and that measure uptime in contractual commitments. For that customer base, an outage is not an inconvenience; it is a breach of service-level agreements.
The 36-minute disruption will not move the revenue line. But it feeds a concern that has shadowed Anthropic’s rise: whether a company scaling as fast as any in history can keep its infrastructure ahead of its growth. Anthropic has been adding compute capacity at a furious pace, and its two largest suppliers — Google and Amazon — are also its shareholders, a structure that has raised questions about how much control the company actually has over its own reliability.
Downtime is the AI industry’s recurring vulnerability. OpenAI has suffered notable outages of its own, as have Google and Microsoft’s AI services. The difference for Anthropic is the timing. A public offering in October would put the company’s operational record under the most intense scrutiny of its existence, and every incident between now and then becomes a data point in investor due diligence.
The outage also highlights the concentration risk in AI’s supply chain. Anthropic’s services depend on cloud capacity from Google and Amazon, which means its reliability is only as good as the weakest link across multiple providers. When one leg of that chain stumbles — whether a cloud region, a power grid or a software deployment — the failure propagates to customers who have no visibility into where their requests are being routed.
Enterprise customers have begun to price this risk. Procurement teams evaluating AI platforms increasingly ask about multi-region redundancy, failover plans and uptime guarantees before signing contracts that commit their own operations to a model provider. Analysts who advise those teams said the recent string of outages across the industry has made reliability a first-order buying criterion, alongside price and model quality.
For Anthropic, the fix is expensive and slow. Adding redundant capacity across regions, insulating services from upstream cloud failures and testing failover scenarios all require engineering time and capital — two things the company has in abundance, but which cannot be rushed. The company has said it invests heavily in reliability, and its engineering organization has grown rapidly, but the record speaks for itself: two incidents in a month.
The market’s reaction, so far, has been muted. Anthropic is private, and its share sales happen in tender offers; no public price has reacted to the outage. But the signals from the IPO preparation are visible. Bankers preparing the offering are said to be assembling detailed risk disclosures, and investors conducting diligence have been asking pointed questions about infrastructure, incidents and mitigation plans.
What makes the outage especially awkward is the contrast with the company’s own numbers. A company that grew revenue 14-fold in a year, that posted its first profit, that is being valued at $2 trillion on the strength of its execution — that company spent 36 minutes unable to serve its own users. The gap between the scale of the ambition and the fragility of the infrastructure is the single biggest question facing the IPO, and it will not be answered by the balance sheet.
The incident also lands amid a broader infrastructure arms race. Anthropic has been signing compute agreements measured in tens of billions of dollars — roughly $40 billion with Google and $33 billion with Amazon — and has publicly described its goal of scaling Claude to serve hundreds of millions of users. Each agreement adds capacity, but capacity and reliability are different things: a system that is simply bigger can still fail in ways that a smaller, better-tested system does not.
For the company’s employees, the outage shows how unforgiving the moment is. Anthropic has grown from a research lab to a commercial juggernaut in under three years, and its operations teams are being asked to run infrastructure at a scale that did not exist for any AI company two years ago. The culture of urgency that produced the growth is the same culture that makes a 36-minute outage feel like a crisis, even when the customer impact is measured in minutes rather than days.
For now, Anthropic’s engineers have restored service, and the company has said it is investigating the cause. The next incident, whenever it comes, will be judged against the same standard: not whether it happens, but how fast it is fixed, and what it reveals about the systems underneath. In the AI industry’s hottest public offering window, every minute of downtime is a line item in the prospectus that investors are writing themselves.
