Early Thursday, users across the Asia-Pacific region began reporting that ChatGPT would not behave the way it usually does. Some could not load their conversation history. Others could send messages but watched them fail. A subset using ChatGPT Work and custom GPTs, the features businesses and developers rely on most, found them unusable for stretches of the morning.
OpenAI acknowledged the problem within the hour, posting a terse update to its status page: it was investigating errors affecting ChatGPT Work, conversations, and GPTs for some users in the region. The company said nothing about the cause, only that its team was working on it. Within roughly two hours, the affected services recovered, and OpenAI marked the incident resolved.
The outage was regional rather than global, concentrated in Asia-Pacific with scattered reports elsewhere. That matters because it points to a problem in a specific part of OpenAI’s infrastructure, a data center or a network path, rather than a broad failure of its models. But for the users caught in it, the effect was the same as any outage: a tool they had come to depend on simply stopped working for a while.
What made the incident harder to navigate was that OpenAI’s help center and support chat were themselves unavailable during part of the disruption. Users who wanted to ask what was wrong found the usual channels down, a compounding failure that turns a brief technical glitch into a customer-service problem.
OpenAI has dealt with a string of similar incidents this year. In early September, users in the Asia-Pacific region faced elevated errors across ChatGPT, image generation, file upload, and voice features, a disruption that lasted several hours. Each episode follows a familiar pattern: reports surface on social media, the status page flips to investigating, and the company restores service without disclosing what failed.
The frequency matters more now than it would have a year ago because of how OpenAI’s business has shifted. ChatGPT Work, the enterprise product, is the company’s main route to recurring corporate revenue, and businesses that pay for it treat availability as a contract term rather than a courtesy. A two-hour regional outage is a minor event for a consumer app and a business-continuity problem for a company that has routed its customer support through the tool.
Analysts said the incidents are unlikely to dent OpenAI’s growth in the region, which is among the fastest for AI adoption. But they accumulate. Every outage gives a procurement team a reason to ask hard questions about reliability and to consider keeping a second model provider on hand, which is precisely the multi-vendor world OpenAI’s rivals are hoping to encourage. Anthropic and Google are courting the same enterprise customers, and both have made reliability a selling point. A procurement officer choosing between providers tends to remember the morning the tool went down during a product launch, and that memory can cost OpenAI a contract even when the outage itself was brief.
The region at the center of the disruption is also the one OpenAI is counting on for growth. India, Japan, South Korea, and Southeast Asia have become among the fastest adopters of AI tools, and OpenAI has been staffing and partnering across the region to capture that demand. An outage that lands during the region’s working hours is therefore not an abstraction; it is a direct hit to the customers the company is trying hardest to win.
OpenAI, like most AI providers, reveals little about the cause of its outages. The status page says what failed and when it recovered, but not why. That opacity is common across the industry, yet it leaves customers unable to judge whether a disruption was a one-off or a sign of a deeper problem.
The status-page transparency is itself part of the modern contract between AI companies and their users. OpenAI, like its competitors, now publishes real-time updates on disruptions, a practice borrowed from cloud providers and payment networks. It is a way of signaling that the company takes outages seriously, even when it does not explain what happened.
For now, the Thursday incident looks like a blip. Service was restored quickly, and OpenAI has not suggested that any customer data was exposed or that the problem reached beyond a temporary error spike. The company did not say whether the Asia-Pacific errors were linked to any broader change in its infrastructure.
The larger question is whether these regional disruptions become more common as OpenAI expands its footprint. Running a model service that spans the globe, with enterprise customers on one side and hundreds of millions of consumers on the other, is an engineering problem that grows in difficulty with every new region and every new product. The morning’s errors were small in isolation. Whether they stay that way depends on how the company invests in the parts of its infrastructure that, this week, briefly stopped working.
