Understanding how global events shift traffic patterns helps engineers build resilient systems. This data illustrates how human behavior dictates network load, emphasizing the need for robust observability and the ability to handle both massive spikes and unexpected usage drops.
For 96 years, the World Cup has been a global phenomenon, uniting nations and communities through a shared love of sportsmanship. While its popularity is nothing new, what is novel today is how rare a truly collective global experience has become. In an era defined by microtrends and algorithmic bubbles, it is increasingly uncommon for people across most countries to engage in the exact same event.
That is precisely the unifying power of the World Cup. Fans from all over the globe reshape their daily routines around these once-in-a-lifetime matchups and storylines — and because Cloudflare operates a global network with 330+ points of presence worldwide, we are in a unique position to see exactly how this global ritual reshaped the world’s online activity throughout June and July 2026.
Cloudflare Radar tracks HTTP traffic, DNS, security, and more to highlight global Internet trends. In this blog post we’ll use that data to explore how the World Cup impacted global traffic patterns throughout the tournament’s run.
To understand how traffic changes throughout a match, we had to establish what it is “normally.” One way to do this is by looking at raw request volumes, or the amount of traffic we see on our network per country. But these amounts vary per country (the amount of daily traffic in the United States is always a larger number than the traffic in Portugal), which makes it difficult to establish a globally applicable baseline. Instead, we defined "normal" using the median traffic of the four preceding weeks: a month-long window that provided a stable, per-minute reference and smoothed out day-to-day noise.
We also wanted to know whether traffic rose or fell relative to that baseline, but a plain difference wouldn't let us compare a high-volume country against a low-volume one. Instead, we used the ratio of current to baseline traffic, expressed as a log₂ value: the log makes increases and decreases symmetric around zero (+1 = twice normal, −1 = half). In other words, a score of zero means traffic is perfectly normal, a positive number shows a spike, and a negative number shows a drop.
One factor shaping how traffic changes is simply what time the match kicks off locally. The largest changes in activity happen when a match is played in the overnight and early-morning hours — roughly midnight to 8am local time. These are the hours when very few people are normally online, so fans staying up (or waking early) to watch push traffic well above its usual level, more than doubling it in some cases. As the graph shows, this is where the deviation peaks on both workdays and weekends.
By contrast, matches played during normal daytime and working hours — around 9 a.m. to mid-afternoon — don’t show such an impact: traffic stays close to its usual level, likely because the people watching would already have been online anyway. In the early evening there's a smaller, second lift, most visible on weekdays, as a match keeps people connected at a time when usage would normally start to wind down. Weekends follow a similar shape, with the strong early-morning rise but a gentler evening bump.
The impact of kickoff time is easiest to see when comparing matches within a single country that take place at very different hours. Bosnia and Herzegovina provides a clear example. As seen in the graph shown above, when Bosnia played at 2 a.m. local time, people stayed awake to watch and traffic during the game jumped to well above its normal level, at times more than doubling. When Bosnia played in the evening, the opposite happened: traffic dipped below normal (falling to about 70% of typical value), as people put their devices aside and focused on the match itself.
When Brazil played Japan in the Round of 32 (Brazil won 2–1 on June 29, 2026), the two countries watched the very same game 12 hours apart: kickoff in Brasília (GMT−3) fell during normal waking hours in Rio de Janeiro (GMT−3), while in Tokyo (GMT+9) it landed in the dead of night.
The result is two nearly parallel curves for the same 90 minutes: one higher than normal, one lower. Japan's traffic (red) sits well above normal, around +1, roughly double its usual level, because the match aired in the small hours, when almost no one would ordinarily be online. Brazil's traffic (green), by contrast, runs below normal, around −0.4, as the game fell in the middle of an ordinary active day. In this case, watching the match pulled people away from their usual browsing rather than adding to it.
One of the most compelling aspects of the World Cup is seeing which storylines and teams capture the attention of fans across the world. We’ve discussed how regional traffic patterns change as a result of matches. But who are they watching? Which matches made the most impact on Internet traffic?
Here's how we calculated this: for each match, we took the two-hour window after kickoff and, for every country with enough baseline traffic to give stable measurements (small, noisy markets are excluded), computed how far traffic strayed from normal. We then took the absolute value of each country's deviation, so we're measuring how much traffic changed, not in which direction (a surge and a drop both count as impact), and for each match we took the median of those absolute deviations across all countries. Because several group-stage matches were played simultaneously, making it impossible to attribute a country's traffic swing to one game or the other, we dropped those concurrent matches to avoid ambiguity.
The result is this ranking of the matches that moved the Internet most, worldwide. And there's a surprise: the very top spot wasn’t snagged by a final or semifinal. It was Argentina vs. Switzerland on July 11, a quarterfinal that saw Argentina win 3-1 — and that moved Internet traffic by a factor of about 1.26. That put it ahead of the France vs. Spain semifinal, which had a factor of 1.21. The rest of the top matches were a mix of quarterfinals, round-of-16 and even round-of-32 ties.
To decide which team the world watched most, we looked at each team's matches and aggregated the median worldwide impact across all countries. In other words, when a given team took the field, how much did the typical country's traffic move away from normal? Not surprisingly, Argentina topped the list at 1.17x, meaning that when Argentina played, the typical country's traffic swung about 17% away from its normal level, the strongest global pull of any team. This comes as no surprise, since they were the defending champions and each knockout game could have been Lionel Messi's last dance for his national team. Love them or hate them, people were watching them.
Not far behind were nations packed with superstars such as France, Brazil, Portugal, Morocco, Spain — and Norway, fueled by the Erling Haaland phenomenon. Haiti and Iraq appear in the top as outliers due to their high deviation scores relative to their typical traffic, suggesting matches against major teams drove disproportionate engagement.
Compared to HTTP request data in the month preceding the World Cup, there was an overall increase in requests to gambling industry websites since the opening game. Additionally, whereas pre-tournament traffic followed a clear weekly pattern, after the Cup’s opening game, the trend flattened into a more constant profile, likely a consequence of the high, near-daily regularity of matches.
Because Cloudflare is present in 120+ countries and handles traffic from Internet users worldwide, we can see distinct behavioral patterns across the globe. For example, when examining the deviation trends during the Algeria vs. Austria group stage game on June 28, we noticed something peculiar: Austria’s traffic (in red) increased during halftime, while Algeria's (in green) decreased. The former follows the pattern described above of people spending more time online while not watching the game, while Algeria’s is the complete opposite — and they’re not the only ones.
Algeria, in green and denoted as DZ, saw a much higher uptick in Internet traffic during the match than Austria, in red.
To understand patterns in behavior across countries we grouped every country's match-day behavior by the shape of its traffic curve and let the patterns cluster together.
Grouping match-day traffic shapes this way, three distinct patterns emerge. The largest group (44 countries playing 101 matches) shows Internet usage rising during hydration breaks and halftime, the natural pauses in play, as people reach for their phones. A second, smaller group (8 countries playing 18 matches)) is its near mirror image: traffic falls at exactly those same moments, dipping during the breaks instead of climbing. The third group is a clear outlier, made up entirely of Iran's three matches. The explanation is simple: the May baseline was measured while Iran was still coming back online after the shutdown, so its match-day traffic sits far above that depressed reference, producing a deviation unlike any other country's. You can read more about Iran’s Internet shutdowns and partial restoration throughout 2026 on our blog.
To better understand the second cluster, which included Algeria, Tunisia, Jordan, Egypt and DR Congo, we looked more closely at the traffic mix for these countries. We broke down traffic patterns by Multipurpose Internet Mail Extensions, or MIME type, and grouped it in families to easily distinguish clusters of content types. MIME types act like digital labels that tell browsers exactly what kind of file they are receiving, whether it's an HTML page, a JPEG image, or an MP4 video stream. By tracking these labels, we can infer what kinds of content users are consuming.
Our hypothesis was that this behavior could be explained by a disproportionate amount of people watching the games via streaming in those countries. To test this, we compared traffic pattern distribution in games with teams of both clusters. In the following example, we see traffic distribution of Algeria and Austria respectively in the match between both countries.
In Algeria, traffic was far above normal, then dipped at halftime. Note the large increase in streaming traffic, in orange.
In Austria, where streaming services were used less, Internet traffic increased at halftime.
In the Algeria graph above, we can see that the bulk of the increase during the match window indeed was driven by requests to multimedia and streaming services. This supports our hypothesis that the traffic trendlines correlate with use of streaming to watch the match.
In Algeria, traffic rose sharply at kickoff, dropped during half-time, and returned to elevated levels once the second half began. Hydration breaks, by contrast, had little to no visible effect, which suggests that viewers don't meaningfully change their Internet or social behavior for short, in-play pauses, but do so during the longer halftime interval. Other countries in this cluster show similar behavior. This might be because a viewer is unlikely to close a stream for a three-minute cooling break, but a fifteen-minute halftime is long enough to close the stream and step away.
A minority of countries, including Tunisia and Algeria, disconnect during halftime, with traffic dropping below its in-play level (the blue boxes, sitting under the 1.0 line). The majority of countries go the other way: traffic rises during the break as people pick up their phones the moment play stops, then settles again when the second half begins. Croatia and Bosnia and Herzegovina show this most strongly, with halftime traffic running well above their in-play baseline.
Halftime is a long, familiar pause, but what about the much shorter hydration breaks? These last only about three minutes, taken midway through each half. Is three minutes really enough to change how people behave online? It turns out it is. Just as at halftime, the moment play briefly stops, traffic in most countries ticks up before falling again when the game resumes.
We measured this by taking, for each match, the peak number of requests in a window around the middle of each half (where the hydration break falls) and comparing it to the five minutes immediately before the break. The pattern lines up with everything we've seen so far: the same audiences that surge at halftime also spike during these brief pauses, while the few countries that tend to disconnect during breaks show little or no lift. Even a three-minute gap in play is enough for a large share of viewers to glance back at their phones.
At the scale of Radar’s global HTTP requests, it is genuinely hard for any single event to leave a visible mark. Even so, the 2026 World Cup Final, which pitted Europe’s and American football champions against each other, had enough social impact to affect the Internet’s footprint. When looking at the volume of HTTP bytes from June 20 to June 22 we can immediately identify the final match kicking off at 21:00 UTC on June 19, as well as England vs. France for the Bronze medal at 23:00 UTC on June 18.
During the final match, Argentina and Spain's traffic volume increased up to 20 percentage points when compared to a similar period. The bronze medal match also coincided with a traffic increase, although at a smaller scale.
HTTP request volume during the final match also appeared correlated not only with the timeline of the game, but with each individual stage as well. Looking at the graph, we can roughly pinpoint moments such as the kickoff, halftime break, hydration breaks, as well as the final whistle.
Across every time zone, match, and goal, Cloudflare Radar provides a front-row seat to how the world connects during landmark cultural moments. To explore more interactive traffic insights and track how major worldwide events shape internet activity every day, visit Cloudflare Radar or follow us on social media at @CloudflareRadar (X), https://noc.social/@cloudflareradar (Mastodon), and radar.cloudflare.com (Bluesky).
Continue reading on the original blog to support the author
Read full articleTraditional consensus like Raft struggles in global networks due to leader-dependency. Meerkat uses QuePaxa to provide leaderless, high-availability consensus, ensuring control-plane stability across hundreds of data centers despite unpredictable WAN conditions.
Engineers building global systems often struggle with Raft's sensitivity to network latency and leader failures. Meerkat demonstrates a leaderless approach that improves availability in unpredictable wide-area networks, offering a more robust foundation for critical control-plane data.
This case study demonstrates that even logically sound architectural changes can trigger hidden internal bottlenecks at scale. It highlights the importance of profiling query planning and shows how massive part counts in ClickHouse can lead to unexpected lock contention.
This migration demonstrates how moving from eventually consistent stores to transactional databases and specialized container infrastructure can drastically improve performance and scalability for high-concurrency workloads like headless browsers and AI agents.
