When Cloudflare crashes: The other invisible pillar of the internet collapses 🌐💥

Cliquez ici pour lire en français

Tuesday, November 18th, 2025, 12:20 PM. You open X (formerly Twitter) to check the latest news. Blank page. You try to log into ChatGPT for an urgent question. Error 500. Canva refuses to load your designs. League of Legends kicks you out of your game. You start to panic: another outage? Not quite the same as a month ago, but almost as spectacular.

This time, it wasn’t AWS that buckled, but Cloudflare. Another name you’ve probably never heard of, yet one that protects and accelerates a massive portion of the global web. And when this invisible shield falls, chaos ensues.

Less than a month after the massive AWS outage that paralyzed Snapchat, Netflix, and thousands of services, the internet has once again proven its fragility. But before understanding what happened, we need to grasp who Cloudflare really is and why it’s so crucial.

The invisible guardian of the web 🛡️

If AWS is the foundation that hosts websites, Cloudflare is its security and acceleration system. Founded in 2009, this California-based company set itself an ambitious mission: to make the internet faster, safer, and more reliable for everyone.

Concretely, Cloudflare plays three essential roles. First, it acts as a shield against cyberattacks. Every day, millions of DDoS (Distributed Denial of Service) attack attempts try to overwhelm websites with artificial traffic to make them inaccessible. Cloudflare filters this malicious traffic before it reaches the websites’ servers. It’s a digital bodyguard that discreetly protects millions of sites.

Second, it drastically accelerates page loading. Thanks to its global network of over 270 data centers spread across all continents, Cloudflare temporarily stores copies of websites. When you visit a Cloudflare-protected site, you’re actually accessing the copy closest to you geographically. In Paris? You access the Paris server. In Tokyo? The Tokyo server. Result: loading times cut in half, sometimes even by two-thirds.

Finally, Cloudflare manages DNS (Domain Name System) for millions of domains. DNS is the internet’s phone book: it translates site names like « google.com » into IP addresses that computers understand. Without DNS, you’d have to memorize number sequences for each site. Cloudflare is one of the world’s largest DNS managers.

Image generated with AI

Approximately 20% of global web traffic passes through Cloudflare. Giants like Discord, media outlets like Frandroid, platforms like Canva, games like League of Legends, and even government e-visa portals depend on its services. It’s colossal.

The parallel with AWS is striking: like AWS for hosting, Cloudflare has become a single point of passage for a huge share of the web. And as we saw on November 18th, this centralization comes at a price.

Black Tuesday: When the shield collapses 💥

Back to that Tuesday, November 18th, 2025. At 11:20 AM UTC (12:20 PM French time), something breaks in Cloudflare’s infrastructure. Very quickly, the first reports flood in. X (Twitter) stops responding. ChatGPT displays errors. Canva refuses to load. League of Legends boots out its players.

The irony of the situation? Even Downdetector, the website that tracks internet outages, became inaccessible itself. It’s a bit like firefighters calling the fire department because their own station was on fire.

At 12:48 PM (Paris time), Cloudflare finally publishes a terse first statement: « Cloudflare is currently experiencing internal service degradation. Some services may be affected intermittently. » The kind of sentence that’s about as reassuring as « stay calm » shouted during a fire.

Image generated with AI

Testimonies start flooding social media. A corporate YouTuber reports having to postpone recording several podcasts. A copywriter explains that even his decentralized open-source software like Anytype was blocked because some encoding went through cloud applications hosted behind Cloudflare. A manager at a major French bank confides anonymously: « In moments like these, you realize our dependence on the Cloud and American providers… even just to respond to customers via chatbots. »

The impacted services are mind-boggling in their diversity: OpenAI (ChatGPT), X (Twitter) with over 5,600 reported incidents, Canva, Spotify, Facebook, e-visa portals for Saudi Arabia, Kenya, and Thailand, news sites like Numerama and Les Numériques, and even Cloudflare’s own internal services including its admin dashboard.

The outage hits all continents simultaneously, demonstrating the extent of global dependence on Cloudflare.

The origin of chaos: A simple configuration error 🔧

Contrary to initial fears, this wasn’t a massive cyberattack, sabotage, or natural disaster. Just a human error, amplified by a technical domino effect.

The origin of the outage is almost embarrassingly mundane: a change in a database generated an oversized file. Imagine a recipe that should fit on one page but suddenly becomes 50 pages long. Cloudflare’s machines weren’t prepared to handle such a large file.

Result: a bug that spread like wildfire across Cloudflare’s 270 data centers worldwide. Servers crashed one after another. Worse still, the system oscillated: functional for a few minutes, then down again, then functional… A real nightmare for engineers trying to understand what was happening.

At 2:09 PM (Paris time), Cloudflare announces it has identified the problem. At 3:30 PM, the fix is deployed and traffic gradually returns to normal. At 6:06 PM, it’s officially resolved. Total duration: approximately 3 to 4 hours for most services.

Overnight, Cloudflare’s CEO Matthew Prince apologizes and promises measures to prevent recurrence.

Déjà vu: The chilling parallel with AWS 🔄

For those following tech news, this scenario has a whiff of déjà vu. Less than a month ago, on October 20th, 2025, AWS experienced a catastrophic outage that paralyzed Snapchat, Fortnite, Netflix, Signal, and thousands of other services for nearly 14 hours.

The cause was different (a malfunction in the load balancer monitoring system in the US-EAST-1 region), but the consequences were similar: a devastating domino effect, millions of impacted users, paralyzed businesses, massive financial losses.

These two major outages in less than a month reveal an uncomfortable truth: the internet rests on a few key players whose failure can trigger global chaos. AWS controls roughly 30% of the cloud computing market. Cloudflare protects and accelerates 20% of global web traffic. Together, they form a critical part of internet infrastructure.

The problem isn’t so much these companies’ competence—AWS and Cloudflare are technical references with engineering teams among the world’s best. The problem is excessive centralization.

Imagine a highway where 50% of traffic passes over a single bridge. Even if that bridge is the strongest in the world, built with the best materials by the best engineers, if it collapses, it’s total catastrophe. That’s exactly what’s happening with the internet today.

Both outages also revealed the same fragility: mundane internal errors (permissions change for Cloudflare, subsystem malfunction for AWS) can have planetary consequences within minutes.

Internet centralization: A fragile house of cards 🏰

These successive incidents raise a crucial question: why did we allow the internet to become so centralized?

The answer is simple: economics. AWS, Cloudflare, Microsoft Azure, and Google Cloud offer exceptional services at competitive prices. For a startup, paying a few hundred euros per month rather than investing millions in its own servers is obvious.

Result: massive concentration. AWS, Azure, and Google control 70% of the cloud market. Cloudflare and its competitors dominate web protection. This oligopoly creates systemic risks: when one falls, it’s a global domino effect.

Image generated with AI

Solutions theoretically exist. Companies can distribute their services across multiple providers (« multi-cloud »). But it’s more expensive, more complex, and requires rare expertise. In Europe, initiatives like Gaia-X attempt to create alternatives, but the lag is colossal compared to American giants with 15 to 20 years’ head start.

Ignored lessons: Why it will happen again 🔁

Here’s the frustrating part: these outages aren’t surprises. AWS has experienced several major incidents (2021, 2023, 2025). Cloudflare too. Each time, promises of « corrective measures » abound. Yet it continues.

Why? Because perfection doesn’t exist in computing. These systems are dizzyingly complex: millions of lines of code, thousands of servers, hundreds of interconnections. A single error can topple everything.

Cloudflare promised stricter validation, new emergency mechanisms, complete revision. Exactly what AWS promised after its October outage. Let’s be realistic: another outage will occur. The question isn’t « if » but « when. »

What if next time, instead of a few hours, it lasted several days? Would our digital world survive?

Accept the risk or change the system ? ⚖️

November 18th, 2025, will remain a brutal reminder of our dependence on a few tech giants. Less than a month after AWS, Cloudflare demonstrated that even the most robust infrastructures can collapse.

Two options lie before us. The first: accept this risk as the price of cloud convenience. This is the option we’ve implicitly chosen so far.

The second: fundamentally rethink internet architecture to make it more distributed and resilient. Invest massively in alternatives. Require critical services to maintain redundancy. This option is expensive and complex, but perhaps necessary.

Because imagine: what if the next outage occurred simultaneously at AWS, Azure, and Cloudflare? What if it lasted 3 days instead of 3 hours? Would our entirely digitized economy survive?

Food for thought before your next post on X or your next League of Legends game. Behind every click lies an invisible, fragile, and perhaps over-centralized infrastructure.

Were you impacted by this November 18th outage? How did you experience it? Did you notice that two major outages hit the internet in less than a month? Share your experience and opinion in the comments!