When Cloudflare went dark on November 18, 2025, it didn’t just slow things down—it pulled the plug on the modern internet. By mid-morning UTC, OpenAI’s ChatGPT and its cutting-edge video generator Sora vanished from the web, leaving millions of users stranded. The cause? A cascading failure in Cloudflare’s global network, the invisible backbone that holds up nearly 20 million websites, including some of the most trafficked AI platforms on earth. This wasn’t a glitch. It was a systemic collapse.
How One Company Broke the Internet
Cloudflare, founded in 2009 by Matthew Prince, Michelle Zatlyn, and Lee Holloway, processes over 101.5 million HTTP requests per second. Its 1.1.1.1 DNS resolver handles 1.2 trillion queries daily. When it failed, the ripple effect was immediate. Domains relying on Cloudflare’s CDN, DNS, and DDoS protection couldn’t resolve. Browsers returned blank screens. APIs timed out. Even basic web navigation became a gamble.At 08:00 UTC, users in Tokyo, London, and New York began reporting failures. By 09:15 UTC, OpenAI’s status page flashed “Incident in Progress.” By noon, Cloudflare’s own status page declared a “Major Outage” across all services. The timing was brutal—peak global usage hours. Students couldn’t access research tools. Developers lost API access. Enterprises relying on Sora for marketing assets were paralyzed.
OpenAI in the Crosshairs
OpenAI, co-founded by Sam Altman, Ilya Sutskever, and Greg Brockman, was uniquely vulnerable. With 100 million weekly active ChatGPT users and 50,000 enterprise testers on Sora, the outage wasn’t just inconvenient—it was economically catastrophic. Analysts estimated losses of $12,000 per minute across e-commerce and content platforms using Cloudflare. Within four hours, total damages exceeded $500 million, according to Cloudflare’s 2024 Service Disruption Cost Model.What made this worse? OpenAI had no backup DNS. Its infrastructure was deeply entwined with Cloudflare’s. No redundancy. No failover. “We assumed Cloudflare would be there,” one senior engineer at OpenAI told a source familiar with internal discussions. “We didn’t plan for the backbone to break.”
Engineering Teams in Overdrive
At 12:30 UTC, Cloudflare’s official Twitter account confirmed the issue: “Investigating a global outage impacting our network. Users may experience DNS resolution failures and origin connectivity issues. Engineering teams are engaged.” That was it. No timeline. No root cause. No reassurance.By 18:45 UTC, Tom’s Guide’s final update confirmed engineers were still working—no resolution in sight. The lack of transparency frustrated users. “I’ve seen outages before,” said a software developer in Berlin. “But this? This felt like the internet forgot how to breathe.”
Experts suspect a BGP routing corruption—similar to Cloudflare’s June 2022 outage that lasted nearly three hours. But this time, the scale was worse. Cloudflare had already suffered six major incidents in 2025, exceeding its own SLA target of four. The company had publicly boasted of 99.995% uptime, a claim Matthew Prince made at the May 15, 2025 SaaS Security Conference in New York. “We continuously refine incident response for the 0.005% scenarios,” he said. The November 18 outage was a brutal reminder: 0.005% can still bring the world to a halt.
Why This Matters Beyond the Tech World
This wasn’t just a tech problem. It was a societal one. Hospitals using AI for diagnostics lost access. Journalists relying on ChatGPT for research drafts were stuck. Small businesses using Sora to generate product videos couldn’t post content. Even TikTok and Shopify, though not officially named, likely felt the tremors—many use Cloudflare for edge security.The outage exposed a dangerous truth: the internet is more centralized than we admit. A single company, however well-intentioned, holds too much power. When Cloudflare stumbles, the entire ecosystem shudders. And unlike a power grid, there’s no manual switch to flip.
What’s Next?
Cloudflare has promised a post-mortem, but no date has been set. OpenAI, meanwhile, is quietly exploring multi-provider DNS strategies—something it avoided for years to maintain speed and simplicity. Industry watchers say this outage will force every major AI platform to reconsider its infrastructure dependencies.For now, users are left with one clear message: don’t rely on ChatGPT or Sora right now. And maybe, just maybe, we should all start asking: who else is holding the keys to our digital lives?
Frequently Asked Questions
Why did ChatGPT and Sora go down if Cloudflare wasn’t their host?
Cloudflare doesn’t host OpenAI’s servers, but it handles critical infrastructure like DNS resolution and content delivery. When Cloudflare’s network failed, browsers couldn’t find where ChatGPT and Sora lived online—even though the servers themselves were running. It’s like the address on a house being erased from every phonebook.
How long did the outage last, and when was service restored?
The outage lasted over 10 hours, from approximately 08:00 UTC to after 18:45 UTC on November 18, 2025. No official restoration time was announced. Cloudflare’s final update confirmed engineers were still working, with no ETA. Service gradually returned over the next 24 hours as DNS caches repopulated, but full stability took until November 19.
Was this Cloudflare’s worst outage ever?
Yes. While Cloudflare’s July 2024 outage lasted 27 minutes and affected 0.25% of traffic, the November 2025 event impacted nearly all services globally. It was Cloudflare’s seventh incident exceeding 1% of network traffic in 2025 alone—far beyond its SLA target. The scale, duration, and economic impact made it the most severe in the company’s history.
Could OpenAI have prevented this?
Possibly. OpenAI had chosen to rely entirely on Cloudflare for DNS and edge services, avoiding multi-provider redundancy to simplify architecture. While this improved performance, it created a single point of failure. Experts say most enterprise-grade AI platforms now use at least two DNS providers—a practice OpenAI is reportedly adopting after this incident.
What does this mean for everyday users of AI tools?
It means your favorite AI tools are only as reliable as the infrastructure beneath them. If you depend on ChatGPT for work, school, or creativity, you now know it’s vulnerable to third-party failures beyond OpenAI’s control. The lesson: diversify your tools, back up critical outputs, and never assume cloud services are bulletproof.
Is Cloudflare still trustworthy after this?
Cloudflare remains a critical player in internet infrastructure, but its reliability claims have been damaged. The company has a strong engineering reputation, but seven major outages in one year raise serious questions. Customers are now demanding transparency and redundancy options. Trust won’t be rebuilt until Cloudflare proves it can prevent—not just recover from—catastrophic failures.