Cloudflare Hits Fastest in 60% of Top Networks With Rust-Based FL2!

Hey everyone, it's me, Shiichan! Today I've got some exciting numbers for you: a story about how much faster a network got. We're talking milliseconds here, but for the people using the sites and apps you build, every one of those milliseconds is the experience itself.

Cloudflare Blog blog.cloudflare.com

What was announced?

This one comes from the Cloudflare Blog. By migrating their request handling layer to a Rust-based architecture called FL2, Cloudflare grew the share of the world's top 1,000 networks where it is the fastest from 40% to 60%. And this isn't a number from a lab, it's real speed based on measurements from actual users.

The story so far

In their last performance post (published during Birthday Week in September 2025), Cloudflare was the fastest across 40% of the largest 1,000 networks in the world. Even back then they noted they were close, often by tiny margins, in many more networks, but they weren't satisfied with 40%. By December 2025 (their most recent analysis), that number reached 60%, and that's today's news.

What changes

Between September and December, Cloudflare became the fastest in 40 more countries and 261 more networks. The biggest jump was in the United States, where it took the top spot in 54 additional ASNs.

And on average throughout December, it was 6 ms faster than the next-fastest provider. That may sound small written down, but it means real users all over the world can connect to your site or app that much sooner.

Dive Deep

First, "how do they measure it?" Cloudflare picks the top 1,000 networks by estimated population (the source is APNIC's data) and uses connection time, how long it takes a device to complete a handshake with the endpoint, as the metric. It's the closest thing to the "Internet speed" users actually feel (they explain why they favor this metric in an earlier post).

For rankings they use the trimean of connection times. It's a weighted average of the first quartile (25th percentile), the median (50th percentile), and the third quartile (75th percentile), which smooths out noise and outliers so the "typical user experience" shows through cleanly.

The way they gather the data is neat too: it's called Real User Measurements (RUM). When a user hits a Cloudflare error page, a tiny speed test runs quietly in the background, and the browser pulls small files from Cloudflare, Amazon CloudFront, Google, Fastly, and Akamai, recording the round-trip time for each. That gives data straight from real-world network conditions.

Then, "how did they make it faster?" One part is physical: new points of presence (PoPs) in Constantine (Algeria), Malang (Indonesia), and Wroclaw (Poland). In Wroclaw, free users' round-trip time (RTT) went from an average of 19 ms to 12 ms (a 40% improvement); in Malang, Enterprise traffic went from 39 ms to 37 ms (a 5% improvement).

The other part is software, and that's the real star behind the 40%-to-60% jump. By leveraging protocols like HTTP/3 and rethinking congestion window management, they shaved milliseconds off processing right in the code. They also improved CPU and memory usage in the fundamentals, such as establishing connections, SSL/TLS termination, traffic management, and the core proxy that every request flows through, so the global hardware fleet runs more efficiently.

Cloudflare compares incoming connections to highway toll booths. If there aren't enough booths, or they're slow, lines build up, so they keep speeding up each booth while also routing cars to open ones to keep the lines short.

Wrap-up

• Cloudflare grew its "fastest" share of the top 1,000 networks from 40% to 60% (September to December 2025)
• The main driver was software: migrating the request handling layer to the Rust-based FL2 architecture. New PoPs (Constantine, Malang, Wroclaw) helped too
• 40 more countries and 261 more networks between Sept and Dec; 54 more ASNs in the US; 6 ms faster than the runner-up on average in December
• Measurement uses APNIC-based top 1,000 networks, the trimean of connection times, and real-user RUM data

If you're an infrastructure or SRE fan who wants to know how a network actually gets faster, right down to how it's measured, this one's for you!