Connectivity’s Uneven Floor
When the pandemic forced the world indoors, a reliable Internet connection became the difference between participating in daily life and being cut off from it. For those with fast, affordable access, remote work, online schooling, telehealth, and video calls with distant family were all manageable. For those without it, even scheduling a vaccination or finding reliable health guidance could become a near-impossible task.
That split is not a niche concern. Roughly 60% of the global population has some form of Internet access, which means the remaining 40%—billions of people—have none. And among those who are connected, the quality of that connection varies enormously: some enjoy low-latency, high-bandwidth fiber; others pay premium prices for slow, unreliable service that makes a simple video call an exercise in frustration.
This connectivity gap is not merely a technical annoyance. It is a societal one. Without good access, communication, education, employment, and access to essential services all suffer. The problem is not about building better routers or faster protocols; it is about ensuring that the benefits of the networked world are not reserved for a fortunate subset.
Cloudflare’s response is Project Pangea, an initiative aimed at community networks in underserved areas. The goal is straightforward: provide those networks with cheap, secure, high-bandwidth transit to the broader Internet, reducing the cost and complexity of getting online. It does not solve every infrastructure challenge—last-mile wiring, spectrum availability, and hardware cost remain—but it targets a key bottleneck by making the backbone more accessible to the ISPs that serve these communities.
Why Macro View Hides the Problem
Look at Internet performance on a national or regional scale and the picture often appears healthy. Average speeds are decent, and coverage maps look full. Zoom in, though, and the averages hide sharp disparities. A county might have excellent service in one district and near-dial-up speeds in another. A city can boast fiber-to-the-home in some neighborhoods while adjacent blocks struggle with spotty, oversubscribed connections.
Cloudflare is in a position to see these variations clearly. With servers in more than 200 cities across over 100 countries, and tens of trillions of requests processed monthly, the company’s network data spans the full breadth of the Internet’s actual performance. That vantage point reveals that poor or nonexistent connectivity is not a remote, developing-world-only issue; it exists in the wealthiest nations as well. If you are reading this on a solid connection, the nearest dead zone or underserved community may be much closer than you expect.
Measuring Internet Quality Beyond a Single Number
Internet performance can't be reduced to one figure. The quality of a connection depends on both how much data can flow through it and how quickly that data travels. Bandwidth, measured in megabits per second (Mbps), determines how fast large files or streams download. Latency, measured in milliseconds (ms), is the round-trip time for data to reach a server and return—critical for real-time applications.
To understand where connectivity excels and where it fails, we examined both metrics across eight countries: the United States, Brazil, United Kingdom, Germany, France, South Africa, Japan, and Australia. The results reveal patterns that don't always match expectations about urban and rural divides.
Latency below 30ms delivers near-imperceptible delay for gaming or video; performance degrades noticeably around 60ms. For bandwidth, 5–10Mbps suffices for video conferencing, and the US Federal Communications Commission (FCC) sets the "Advanced Service" bar at 25Mbps. One indication of problems in the US: 9% of ZIP codes show average latency above 150ms, the threshold where videoconferencing becomes unreliable.
Eight Countries, Eight Different Profiles
United States: Inequality in the Wealthiest Markets
US median throughput sits at 50.27Mbps with median latency of 46.69ms. The country's connectivity problems are visible even at county level, though they don't break down neatly by geography. Yes, remote areas struggle, but large cities including Los Angeles, Milwaukee, San Francisco, and Florida's Orange County contain ZIP codes with uniformly poor performance adjacent to well-served neighborhoods.
The FCC has acknowledged its broadband maps don't accurately identify underserved areas. Our data suggests the problem persists despite new data collection efforts. Even where performance is strong, costs vary wildly—President Biden's 2021 Executive Order noted that more than 200 million US residents have only one or two reliable providers, paying up to five times more than markets with genuine choice.
New York City tops the list for bandwidth among US locations, a reminder that excellent numbers at the aggregate level can hide severe local disparities. Cities like Fairfax, Virginia, and Knox County, Tennessee also show drastic variation within small geographic areas.
Australia: Bandwidth Gaps Beyond the Rural Divide
Australia's median throughput is just 33.34Mbps, with median latency of 42.04ms. Suburban areas suffer more than the headline suggests: regions like Illawarra perform no better than rural Wagga Wagga. Fiber-to-the-premises connections in city centers create stark contrasts, leaving many households without adequate bandwidth for multiple simultaneous video calls or HD streaming.
Interestingly, Australia's latency profile is respectable—the country performs well on responsiveness even where throughput disappoints.
Japan: Low Latency, Uneven Speed
Japan shows how a country can maintain consistently low latency across vast distances—even Okinawa, roughly a thousand miles from Tokyo, shares the national profile of 31.89ms median latency. The problem lies elsewhere: median throughput is 61.4Mbps nationally, but several prefectures on Kyushu Island, Okinawa, and Western Honshu substantially lag behind. Tokyo, as the concentration of population and data centers, predictably shows the best performance.
United Kingdom: Following the Hubs
Latency across the UK is generally strong at 34.12ms median, but bandwidth tells a different story. Inner London and Manchester lead, and these cities also host the country's main Internet exchange points. Efforts to push content and data centers into other population centers, such as Edinburgh, could meaningfully improve regional performance. Median throughput is 53.8Mbps.
Germany: Unifying Through Frankfurt
Frankfurt am Main stands as one of the world's major Internet hubs, and Germany's best performance radiates from there. However, the former East Germany shows both higher latency and lower speeds—median throughput is 48.79Mbps overall, with median latency of 42.1ms. The disparity tracks historical boundaries, creating a noticeable performance split.
France: Hubs Beyond Paris
Paris has long anchored French Internet, but Marseille is emerging as a second hub thanks to submarine cable landings. Lyon and Bordeaux are positioned for similar growth. These four cities deliver the highest throughput (median 48.51Mbps nationwide) and lowest latency (median 54.2ms). Our data shows performance tied closely to geographic proximity to these interconnection points.
Brazil: Strong Core, Disconnected Periphery
Most of Brazil demonstrates solid latency—49.25ms median—because of proximity to major hubs in São Paulo and Rio de Janeiro. The Amazon region stands out for its lack of service, with many connected areas posting both low bandwidth and high latency. Campinas deserves special mention: it delivers some of the country's best performance and recently gained a new regional data center. Median throughput nationwide is just 26.28Mbps.
South Africa: Three Cities Carry the Load
South Africa's Internet evolution has multiplied from a single hub in Johannesburg to include Cape Town and Durban, and performance clusters around these cities. Still, median throughput of 6.4Mbps means much of the country cannot sustain HD video conferencing or streaming. Latency (59.78ms median) also degrades steadily with distance from the three urban hubs.
Case Study: ISP Concentration in Alabama
Does a market with fewer providers result in worse performance? We examined Alabama across all 65 counties, comparing average download speed against the largest ISP's traffic share.
The pattern is stark. In most counties, the largest ISP carries 80% or more of traffic, with competitors operating at trivial scale. Only three counties—Marion, Escambia, and Etowah—have a dominant carrier with less than half the traffic. Etowah County is among the state's best performers.
Plotting download speed against the combined traffic share of all non-dominant ISPs reveals a strong positive correlation: as smaller providers gain meaningful market presence, speeds rise. Henry County is the clearest illustration—where one ISP carries 98% of traffic, performance is the worst in the state. The pattern argues that investment alone doesn't improve a region's Internet; genuine ISP competition is essential for better throughput.
When Distance Isn’t the Real Problem
For many underserved regions, the challenge isn't just a lack of fiber or bandwidth. A separate but equally damaging issue is poor routing, often called "tromboning." This occurs when data packets travel far outside their logical path, only to loop back to a destination that was nearby all along.
A concrete case is Tuskegee, Alabama, home to a local university. Off-campus students have only a single choice for high-speed broadband. Even with that local access, traffic bound for campus can take a bizarre detour: routing all the way through Atlanta—two hours to the northeast—before returning to its destination.
This happens because the largest ISPs exchange traffic with other networks in only a small set of major hubs: Seattle, San Jose, Los Angeles, Dallas, Chicago, Atlanta, Miami, Ashburn, and New York City. If you live in one of those metros, you likely never notice. If you don't, your traffic may be shuttled across state lines—or international borders—before completing a short hop.
The economic impact is real. Tromboning represents wasted bandwidth and increased cost, as a growing amount of data is hauled over long distances instead of being exchanged locally. This inefficiency is passed down to consumers in the form of slower performance and higher access costs.
Tools like traceroute and the RIPE Atlas platform can reveal these paths. Testing from Alabama to Facebook produced an extreme display of the problem: a route that travelled from AT&T's network via Atlanta to Telia (an IP transit provider), across the Atlantic through Paris and Amsterdam, before finally reaching Facebook and turning around. Even though Facebook's content is certainly cached in Atlanta, the request travelled overseas because of the network path negotiated between providers.
Traceroute to 157.240.201.35 (157.240.201.35), 48 byte packets
1- 192.168.6.1 1.435ms 0.912ms 0.636ms
2- 99.22.36.1 99-22-36-1.lightspeed.dctral.sbcglobal.net AS7018 1.26ms 1.134ms 1.107ms
3- 99.173.216.214 AS7018 3.185ms 3.173ms 3.099ms
4- 12.122.140.70 cr84.attga.ip.att.net AS7018 11.572ms 13.552ms 15.038ms
5 - * * *
6- 192.205.33.42 AS7018 8.695ms 9.185ms 8.703ms
7- 62.115.125.129 ash-bb2-link.ip.twelve99.net AS1299 23.53ms 22.738ms 23.012ms
8- 62.115.112.243 prs-bb1-link.ip.twelve99.net AS1299 115.516ms 115.52ms 115.211ms
9- 62.115.134.96 adm-bb3-link.ip.twelve99.net AS1299 113.487ms 113.405ms 113.25ms
10- 62.115.136.195 adm-b1-link.ip.twelve99.net AS1299 115.443ms 115.703ms 115.45ms
11- 62.115.148.231 facebook-ic331939-adm-b1.ip.twelve99-cust.net AS1299 134.149ms 113.885ms 114.246ms
12- 129.134.51.84 po151.asw02.ams2.tfbnw.net AS32934 113.27ms 113.078ms 113.149ms
13- 129.134.48.101 po226.psw04.ams4.tfbnw.net AS32934 114.529ms 114.439ms 117.257ms
14- 157.240.38.227 AS32934 113.281ms 113.365ms 113.448ms
15- 157.240.201.35 edge-star-mini-shv-01-ams4.facebook.com AS32934 115.013ms 115.223ms 115.112msThis isn't a critique of the specific companies involved. The networks themselves didn't fail; the structure of their interconnection did. A variety of technical and business conditions between the involved parties can cause this behavior. For the user, the outcome is the same.
The remedy is straightforward on paper: more interconnection points in more cities, and more cross-network peering in each. This carries an upfront capital cost, but the long-term operational savings would justify it.
Mapping the Digital Divide
The data shows that internet quality isn't a simple function of geography, income, or urbanization. It varies dramatically from region to region, including in economically developed areas and dense city centers where you would expect robust infrastructure.
The Alabama case, however, makes the underlying drivers clear. Performance tracks competition. Markets with multiple ISP choices consistently outperform those with limited options. And lack of interconnect points compounds this, producing lossy, indirect routes that cheat users out of speed no matter what they pay.
Expanding access to the last 40% of the world's population offline requires more than fibre in the ground. It demands increasing the number of independent networks that provide service at the local level and creating more locations where those networks can meet and exchange data cheaply and directly. These structural changes are a necessary counterpart to any effort aimed at pulling new users into the connected world.
