ChatGPT went down. X stopped loading. Shopify, Discord, Coinbase, and hundreds of other sites threw error messages. Tuesday morning, a huge chunk of the internet just... broke. The culprit? Cloudflare, the infrastructure company that handles traffic for roughly 20% of the web.
What Actually Happened
The outage started around 7:20 AM ET and lasted about six hours. At its peak, nearly 5,000 users reported issues on Downdetector (which itself was partially down because it also relies on Cloudflare—irononic).
Cloudflare later explained the root cause: a database permissions change caused their Bot Management system to generate an unexpectedly massive configuration file. That file overwhelmed the system responsible for routing traffic, causing widespread failures across their network.
CEO Matthew Prince posted an apology: "On behalf of the entire team at Cloudflare, I would like to apologise for the pain we caused the Internet today." He called it the company's most significant outage since 2019.
The Scale of the Damage
This wasn't just a few websites. ChatGPT and Sora (OpenAI's products), X (Elon's social network), Claude (Anthropic's chatbot), the New Jersey Transit system, government sites including the Federal Energy Regulatory Commission, and countless e-commerce platforms all went dark simultaneously.
If you tried to access any Cloudflare-protected site, you got an error page explaining that your connection was fine, the website was fine, but Cloudflare's services were having issues. Which is accurate but deeply unhelpful when you're trying to get work done.
The outage affected multiple Cloudflare services: their core routing, the Access platform, the WARP service, and even their own dashboard and API. At one point, Cloudflare engineers couldn't use their own tools to fix the problem.
The Single Point of Failure Problem
Here's the uncomfortable truth: the modern internet relies on a handful of companies for critical infrastructure. Cloudflare, AWS, Azure, and a few others handle the vast majority of web traffic. When one goes down, huge sections of the internet become unavailable.
This isn't theoretical. We saw it with AWS in October. We saw it with CrowdStrike in July 2024 (that one grounded flights). We saw it with Microsoft Azure in November. And now Cloudflare in November again.
The companies that keep the internet running are themselves vulnerable to outages. And because so many services depend on the same providers, failures cascade spectacularly.
Why Cloudflare Specifically Matters
Cloudflare does several things that make websites work: it protects against DDoS attacks, speeds up page loads through caching, provides DNS services, and routes traffic efficiently. Sites pay them to handle the messy infrastructure work so they can focus on building products.
About 20% of the web uses Cloudflare. That's millions of sites, from small blogs to massive platforms like Discord. The consolidation makes sense from a business perspective—Cloudflare is really good at what they do—but it creates systemic risk.
When a configuration file grows too large and crashes Cloudflare's routing system, one-fifth of the internet stops working. That's not a bug, that's the architecture.
The Technical Breakdown
The issue started with a change to database permissions. This caused the Bot Management feature file to grow beyond expected size—essentially, the system started logging way more data than it should have. The configuration file used to route traffic ballooned until it crashed the proxy system.
Cloudflare identified the problem relatively quickly, but implementing a fix took hours because the outage affected their own ability to deploy changes. They had to disable features like WARP access in London just to stabilize the network.
Core traffic returned to normal by around 10:30 AM ET. Full restoration happened at 1:06 AM the next day. So technically a 6-hour major outage with a longer tail of intermittent issues.
What Cloudflare Is Doing About It
The company outlined several prevention measures: hardening configuration file ingestion, adding global kill-switches for certain features, eliminating the risk of error reports overwhelming system resources, and reviewing failure modes in core proxy modules.
They're also conducting a full post-mortem, which is standard practice after major incidents. The goal is to ensure that this specific failure pattern can't happen again.
But here's the thing: there will always be new failure patterns. Systems this complex have emergent behaviors that nobody predicted. You can't design for every edge case because you don't know all the edge cases until they happen.
The Broader Pattern
This is the third major infrastructure outage in about a month. AWS, Azure, now Cloudflare. These aren't isolated incidents—they're symptoms of how dependent we've become on a few centralized providers.
The cloud promised resilience through distribution. In practice, we've created new single points of failure that are bigger and harder to fix when they break. Multi-cloud strategies help, but they're expensive and most companies don't actually implement them properly.
There's no easy answer here. Decentralization sounds great in theory but makes services slower and more complex. Centralization is efficient but creates systemic risk. We've collectively chosen efficiency, and incidents like this are the cost.
What This Means for You
If you build anything on the internet, you need redundancy. Don't rely solely on Cloudflare, or AWS, or any single provider. Have fallbacks. Monitor external dependencies. Build systems that degrade gracefully when upstream services fail.
For everyone else: this is just a reminder that the internet isn't magic infrastructure that always works. It's maintained by companies with their own problems, vulnerabilities, and occasional configuration mistakes that take down huge chunks of the web.
Tuesday was a reminder that the internet is simultaneously incredibly robust—most of it kept working—and surprisingly fragile when a key player goes down.
My Take
I was in the middle of a ChatGPT conversation when it died. Switched to Claude, that was down too. Tried to check Twitter to see if it was just me, and X wouldn't load either. That moment of realization—oh, this is big—was mildly terrifying.
We've built so much of modern life on infrastructure we don't control and barely understand. When it works, it's invisible. When it breaks, suddenly you can't access work tools, social media, banking, transit schedules, or AI assistants.
Cloudflare fixed it relatively quickly, all things considered. Six hours is fast for an outage this widespread. But it's a sobering reminder of how centralized and vulnerable our digital infrastructure really is.
The internet isn't resilient because it's distributed anymore. It's resilient because a handful of very competent companies work really hard to keep their systems running. When they mess up, the cracks show.