Why We Keep Score

Cloudflare's core proposition is speed: improvements in our network translate directly into a faster experience for the end users of applications built on our platform. To hold ourselves to that standard, we run a continuous, public benchmarking program against our main competitors. Our latest data, collected from August 6 to September 4, shows that we currently hold the number one spot in 40% of the top 1,000 global networks by estimated population. This is a shift from our September 2024 update, where we reported being the fastest in 44% of those networks.

Measurement here is not about abstract technical superiority; it is an attempt to recreate the real-world conditions of an internet user. We use TCP connection time as our core metric because it most closely approximates the perceived speed of a connection, accounting for real-world factors like congestion and distance that other, more laboratory-bound metrics ignore. To reduce statistical noise, we rank providers using the trimean, a weighted average of the first quartile, median, and third quartile of our connection time data. This gives us a balanced view of overall performance rather than focusing on extreme outliers.

How the Data Is Gathered

We capture this performance data directly from the browser. On a selection of Cloudflare-branded error pages, we run Real User Measurements (RUM). This involves a user's browser retrieving a small file from multiple networks—Cloudflare, Amazon CloudFront, Google, Fastly, and Akamai—to compare real-world performance. This provides us with industry-standard metrics like TCP Connection Time, Time to First Byte (TTFB), and Time to Last Byte (TTLB), allowing for an objective comparison of load times from the user's perspective. The measurement pool consistently focuses on the top 1,000 networks by estimated population, excluding those that aren't last-mile ISPs to best reflect the end-user experience.

Close Margins at the Country Level

The competitive landscape is not just about the global average; it's a country-by-country fight where results are often determined by razor-thin margins. In many regions, the fastest provider is often ahead by less than 5%. For instance, in India, Cloudflare currently ranks second, trailing the leader CloudFront by roughly 5 milliseconds.

These rankings are fluid and can shift day by day. As an example, in Mexico, Cloudflare was the second-fastest provider by 0.73 ms on August 5th, but just a week later, we were the fastest by 3.72 ms. Such frequent reordering underscores the intensity of the competition across the globe.

Our Performance Where We Aren't #1

Rather than solely focusing on the wins, we analyze the data to understand where we need to improve. In networks where we rank second, we are often extremely close to the top provider—in over 50% of those cases, the difference is less than 5% (10 ms or less). Similarly, where we rank third, half of those networks are within a 10% difference of the leader. These slim gaps indicate that we are consistently competitive and that top rankings can change rapidly based on minor optimizations by any provider.

Analyzing individual large networks, like AS701 (Verizon Business), reveals a more nuanced story than a single rank can tell. When considering the P95 value (the 95th percentile, which highlights the worst-case connection scenarios), Cloudflare may rank fourth. However, when looking at the 75th and 50th percentiles of that same data, we are only 2 milliseconds slower than the leader. This indicates that for the vast majority of users, our performance is on par with the fastest provider, even if our worst-case scenarios have more room for improvement.

To better capture this full spectrum of user experience, we adopted the trimean for ranking providers back in July 2025. Summarizing user experience with a single number is always difficult, but the trimean provides a more balanced representation than a single percentile. Still, the P95 remains a key metric for its own merits: looking across all networks in our dataset, we rank first in roughly the same number of networks as our closest competitor, but we hold that top spot in 103 more networks than the next-fastest provider. This suggests that our network shows particular strength in handling the "long-tail" of challenging, worst-case conditions, which points to the inherent resilience of our infrastructure.

Translating Data into Network Upgrades

This performance data serves as a roadmap for network improvements. When we identify a network or region where we consistently underperform, we can locate the cause and implement a fix.

A clear example of this is our work with the network AS8966. Before June 13, 2025, our performance for users on that network was suffering; we were the slowest provider. By analyzing our measurement data, we identified the issue and prioritized deploying equipment to a new partner data center in the region. Partnering with local service providers allows us to host our hardware closer to users, reducing latency substantially. In this case, the new point of presence took our rank from last to first, improving latency by roughly 150 ms in a single day.

Our approach remains a transparent cycle of continuous improvement: we identify performance bottlenecks using this data, roll out internal and infrastructure fixes, and then report the results. We will continue to share our progress as we work to optimize the network and expand our lead in the countries and networks where we are already fastest, while closing the narrow gaps in the rest.