A ChatGPT outage struck thousands of users globally Monday, TechRadar reported, with Downdetector recording a spike from under 1,000 reports to over 5,000 within roughly 30 minutes of disruptions first appearing around 10:05 AM ET, as OpenAI confirmed on its status page it was investigating the issue across ChatGPT, Codex, and the API Platform.

A ChatGPT outage on Monday produced one of the sharpest Downdetector spikes the platform has recorded, going from near-zero reports to over 5,000 within approximately 30 minutes. The disruption was unusual in that it did not affect all users identically: some experienced complete login failures, others could log in but could not load conversations, and some found Codex affected while the main ChatGPT interface appeared functional for them. That variation is consistent with an infrastructure layer failure rather than a single-point outage.

OpenAI acknowledged the issue quickly, posting on its status page that it was investigating “degraded performance” across nearly every core feature. “We are investigating the issue for the listed services. Impacted users are currently unable to access ChatGPT, Codex and API Platform,” the company stated.

What OpenAI Confirmed and What Users Experienced

The status page showed a “Partial Outage” designation, though the scope of affected features and regions made it functionally closer to a full disruption for many users. OpenAI’s updates ran approximately 30 minutes apart during peak reporting, a cadence that users and enterprise clients have previously criticized as insufficient during active incidents.

The pattern on Downdetector, a flat line through Sunday night and early Monday morning followed by an almost vertical spike, is the signature of a sudden systemic failure rather than a gradual degradation. Business users on the API Platform were among those affected, with OpenAI noting issues for customers who had recently added new seats or upgraded accounts.

ChatGPT serves hundreds of millions of users. At that scale, any outage affects not just individual users but the downstream applications, business tools, and workflows built on the API. OpenAI’s enterprise revenue now accounts for 40% of its $2 billion monthly revenue, making uptime a direct factor in client retention and expansion contract negotiations.

The Regional Imbalance and Infrastructure Question

The UK bearing over 7,600 Downdetector reports against roughly 1,700 in the US is a data point worth examining. A geographic imbalance of that magnitude, European users seeing more than four times the disruption rate of American users at peak, suggests the failure may have originated in or propagated through a specific infrastructure region rather than being uniformly distributed across OpenAI’s global footprint.

OpenAI has not disclosed the geographic breakdown of its infrastructure or how load is balanced across regions. The company’s ongoing data center expansion, including a $60 billion facility in Abilene, Texas in partnership with Oracle, is focused primarily on US-based capacity. If European users route requests through US-hosted infrastructure, a localized US failure could manifest more severely in Europe due to latency and routing differences.

What This Means for AI Reliability and Crypto Markets

For the AI bubble debate, infrastructure reliability at scale is a material concern. OpenAI is seeking a $300 billion infrastructure partnership, has confirmed $2 billion in monthly revenue, and is expanding enterprise sales aggressively. A visible outage during a week already defined by major global market volatility produces exactly the kind of SLA pressure that enterprise contract renewals will be negotiated around.

For AI tokens and the broader AI infrastructure investment market, each high-profile outage confirms the scale of adoption that makes reliability so consequential, while simultaneously underscoring the engineering complexity that makes building at that scale more expensive and difficult than base-case investor assumptions. OpenAI restoring service within approximately one hour is a positive operational data point. The absence of any root-cause disclosure and the regional imbalance leave material questions open.