From Honey Pots to the Global Edge: The Rise of Cloudflare

From Spam Tracking to the Cloud

Matthew Prince and Lee Holloway started Project Honey Pot in 2004 to track email spam. They later met Michelle Zatlyn at Harvard Business School in 2009 and planned a service to stop internet threats rather than just monitoring them. Lee Holloway built the first working prototype in his spare time. The team launched Cloudflare at TechCrunch Disrupt in 2010, where early beta users noted that sites loaded 30% faster due to the caching layer. Before Cloudflare, Prince and Holloway worked on Unspam Technologies, where Holloway served as chief software architect. In 2009, they secured $2.05 million in Series A funding from Pelion Venture Partners and Venrock. Following a massive DDoS attack in May 2016, the company grew into an entity with 1,200 employees and 83,000 paying customers by its 2019 IPO. Prince and Zatlyn rang the bell at the New York Stock Exchange in September 2019.

Dominating the Programmable Edge

Cloudflare now handles over 60 million HTTP requests per second at peak. Its network covers 335 data centers in more than 100 countries. I find its move into edge compute through Workers smarter than the static models used by legacy providers. Akamai still holds significant media delivery volume, but Cloudflare wins the programmable edge battle by integrating security and performance into a single fabric. Fastly focuses on its Wasm-based runtime, yet Cloudflare’s scale and the integration of R2 storage create a stronger pull for developers. The company expanded beyond a niche security tool to become a fundamental piece of internet infrastructure that handles massive amounts of traffic through its distributed network of 335 data centers across more than 100 countries. You should remember that the number of Points of Presence often matters less than the actual capacity of each deployment. Cloudflare and Amazon CloudFront collectively serve 55% to 60% of all CDN-fronted HTTP requests. Akamai uses DNS-based mapping to steer traffic, while Cloudflare utilizes an anycast topology. This performance supports a 22x EV/Forward Sales valuation.

AI and the 2025 Outage

The rollout of the Infire Engine in 2025 allowed LLMs to run on edge GPUs without cold starts. This innovation places Cloudflare at the center of the Agentic Internet. However, the company failed its users on November 18, 2025, when a database permission change caused a feature file to double in size and crash the Bot Management module. A ClickHouse query behavior change generated duplicate rows, causing the feature file to exceed the size limit of the software routing traffic across the network. This error caused significant 5xx errors and hindered authentication for many customers. Beyond technical failures, Italian regulators fined Cloudflare €14 million in January 2026 over DNS-level blocking refusals. Can the company maintain its neutral stance while facing increasing regulatory pressure from the EU?