ChatGPT, Claude, and Grok All Go Down: Why Is Everyone Suspecting Cloudflare?

Author: Frank, MSX Maitong

As both the scapegoat for internet outages and an often-overlooked "shovel" in the AI era.

On the evening of September 3rd, ChatGPT, Claude, and Grok, three leading AI platforms, went down almost simultaneously.

++A little complaint: Now even outages have a "new trio," has Gemini been pushed out of the lineup?++

Interestingly, after the incident, many people's first reaction was to check the status pages of Cloudflare, AWS, and Azure, especially Cloudflare.

However, this time it seems that the blame does not fall on Cloudflare, as its status page did not report any major global incidents, and there is currently no evidence indicating that the failures of the three platforms originated from the same source.

This raises a more interesting question: Why is it that whenever there is a widespread service disruption on the modern internet, Cloudflare is always one of the first names to be suspected?

Especially since this time it clearly wasn't their fault, the fact that everyone thought of them first says a lot.

1. Why Does Everyone Think of Cloudflare When There’s an Outage?

Many people first became aware of Cloudflare due to its CDN, website acceleration, or DDoS protection services.

But if today one still only understands it as a CDN company, that view is somewhat outdated.

When we open a website, it seems to be a simple "User → Website" interaction.

In reality, the chain in between is much more complex, closer to "User → Domain → DNS → Cloudflare → CDN / WAF / DDoS / Authentication / Edge Computing → Web Server."

In other words, Cloudflare may not own the website you are ultimately accessing, but it could be positioned between you and the website:

• Websites want to load faster, they need to go through it;
• If under DDoS attack, they rely on it to block the threats;
• API requests, bot traffic, and internal system access by corporate employees must go through it;
• Developers are starting to run programs directly on edge nodes, utilizing Workers;
• In the AI era, with model APIs, AI Gateways, Agents, and MCPs, Cloudflare is also extending into this layer;

Cloudflare's current positioning is as a unified platform serving apps, agents, and workforce, and I believe understanding Cloudflare as the "water, electricity, and gas" of the internet might be more accurate.

Usually, you don’t feel its presence; websites load normally, APIs respond correctly, and malicious traffic is blocked in the background. But as soon as something goes wrong, all websites can suddenly go down together.

For an infrastructure company, this is a very luxurious position.

Follow Official X for US stock information

2. When Cloudflare Goes Down, Does Half the Internet Report Errors?

Of course, the more important the position, the greater the cost of errors.

Looking back at Cloudflare's outage history, it is almost a cautionary tale of modern distributed networks.

In 2019, a regex in a WAF security rule malfunctioned, quickly maxing out the CPU handling HTTP traffic, ultimately affecting core proxies, CDN, and WAF, leading to a massive number of 502 errors across the internet.

In 2022, a network configuration adjustment went awry, impacting 19 data centers, particularly the busiest nodes, affecting about 50% of request volume at one point.

By 2025, as Cloudflare's products increased, incidents began to exhibit "internal contagion."

First in June, the underlying storage infrastructure relied on by Workers KV failed, causing WARP, Access, Gateway, Workers AI, Turnstile, and a series of services to go down for over two hours.

Even more dramatically, in November, a database permission change caused a configuration file used by Bot Management to suddenly double in size and be synchronized across numerous machines, ultimately causing issues with core network traffic.

Cloudflare initially thought it was facing a massive DDoS attack, only to find out it had generated a file that was too large...

In the end, what truly magnifies small errors is Cloudflare's position; it has embedded itself into the essential pathways of too many websites, applications, and services.

Join* Official Community *to discuss hot topics

3. NET: The "Shovel" Easily Overlooked in the AI Era

However, viewing Cloudflare merely as a "very important internet infrastructure company" misses another layer.

In recent years, its business trajectory has become increasingly clear: expanding from CDN, DNS, and DDoS protection to Zero Trust, network security, and Workers development platforms, and now incorporating AI Gateways, Workers AI, and Agent infrastructure.

And AI may be making its already critical position even more valuable.

The reason is simple—the more Agents there are, the more network requests, identities, and security needs arise.

This logic differs from traditional SaaS; previously, the internet was primarily about humans accessing machines—one person opening a webpage, clicking a few links, and exiting, resulting in a limited number of requests throughout the day.

Agents are entirely different; to complete a task, they might continuously search dozens of pages, call several APIs, leading to potentially dozens or hundreds of machine calls.

If in the future, more internet traffic truly shifts from Human-to-Machine to Machine-to-Machine, then Cloudflare will be the platform they operate on and the network they need to traverse, theoretically becoming busier.

The latest financial report confirms this trend, with Q2 revenue reaching $696 million, a 36% year-on-year increase, and the growth rate continuing to rise from about 34% in Q4 last year and Q1 this year.

For a company with annual revenue exceeding $2 billion, achieving such growth is not easy, which is why after the financial report was released in early August, NET briefly surged above $330.

However, while the fundamentals are strong, the market is already aware of this, which is why after the August peak, the stock price fell significantly.

Next, we will see if the machine traffic, security needs, and development platform revenue brought by Agents can grow fast enough to continue to meet the already high expectations.

Conclusion

Last night, when ChatGPT, Claude, and Grok collectively malfunctioned, it was not due to Cloudflare.

But this blunder itself is quite suitable for understanding Cloudflare.

What we usually see is only the top layer; today it’s ChatGPT, tomorrow it’s Claude, and in a few years, it may be replaced by a new batch of Agents.

Applications will change, and entry points will change.

But as long as they still need to connect to the internet, access data, verify identities, call APIs, and require someone to block attacks, that underlying layer of infrastructure will not simply disappear.

In fact, the more machines there are, the busier this layer may become.

What Cloudflare is betting on is likely this very fact.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.