Cloudflare Network Update: Climbing From 40% to 60% of Top Networks
For internet infrastructure, latency isn't just a metric; it is the foundation of user satisfaction. Every millisecond trimmed from a connection directly shapes the experience of the people using your applications and websites. At Cloudflare, performance is a constant focus, and we regularly share our progress to remain transparent.
In our latest analysis, we saw a significant jump in our standing. When we reported results during Birthday Week 2025, we were the fastest provider in 40% of the 1,000 largest networks. By December 2025, that figure had climbed to 60%. This suggests that our recent efforts—combining new hardware deployments with deep software optimizations—are making a tangible difference for real users.
A Methodology Focused on Real-World Connections
To understand this progress, it helps to know how we establish our benchmarks. We define "largest" using population estimates from APNIC for the top 1,000 networks globally. This gives us a sample that spans diverse geographies and reflects where actual users are connecting from.
Our primary comparison measure is connection time—the duration required for a user’s device to complete a handshake with a target endpoint. This metric more closely tracks what users perceive as their internet speed than abstract alternatives, as it inherently includes real-world network conditions like congestion and geographical distance. To ensure our rankings are stable and not skewed by outlier events, we rely on the trimean of connection times, a weighted average of the 25th, 50th, and 75th percentiles.
The data itself is collected via Real User Measurements (RUM). Through our error pages, we quietly run small speed tests in the background. The user’s browser retrieves files from multiple providers—including ourselves, Amazon CloudFront, Google, Fastly, and Akamai—and logs the duration of each exchange. This approach captures performance in living conditions rather than in controlled labs.
The Drivers: New Locations and Software Efficiency
Last year, we detailed how expanding our physical presence reduces latency by bringing hardware closer to users. We continued that strategy with deployments in Constantine, Algeria; Malang, Indonesia; and Wroclaw, Poland. These sites had measurable impacts on connected traffic:
- In Wroclaw, free users saw average RTT drop from 19ms to 12ms—a 40% gain.
- In Malang, Enterprise traffic improved from a 39ms average RTT to 37ms.
However, deploying new locations alone cannot account for the leap from 40% to 60% market leadership. A substantial part of that gains comes from code-level refinements in how we handle connections. By leaning on protocols like HTTP/3 and optimizing how we manage congestion windows, we save milliseconds directly in the software layer. We have also focused on reducing CPU and memory usage in our core processes—including connection establishment, SSL/TLS termination, traffic management, and the central proxy that every request traverses. These efficiency gains allow us to squeeze more performance out of our existing hardware fleet.
One way to visualize this is through a toll booth model. Network congestion is like lines forming at a road toll. You fix it either by building more booths or by making each booth process cars faster. We are improving both aspects: accelerating the per-car processing at the booth (software) and improving how we route cars to available booths (traffic management).
Seeing the Results in the Data
By December, we had overtaken the competition in 261 additional networks compared to our September figures, spanning 40 new countries. The most dramatic change was observed in the United States, where we are now the fastest in 54 more ASNs than before.
Throughout December, we maintained a decisive margin over the competition. On average, our connection times were 6ms faster than the next-quickest provider during that period, demonstrating that the gap is not just about the number of wins, but the consistency of our performance over time.
A Commitment to Chasing the Remaining Margins
Those figures translate into real users connecting to their favorite sites and services faster. Still, we view a 60% majority as a launching point rather than a summit. There are many networks where we rank second by a hair's breadth. We can see those sub-second distinctions clearly in our data, and we are actively working to close those final gaps. Our commitment remains to deliver the fastest connectivity across the world’s networks. In the interim, we will continue to post updates as we make progress.
