DEV Community

Cover image for How Many Websites Use Nginx? 72.9M Sites, Counted From One Response Header
webatla
webatla

Posted on

How Many Websites Use Nginx? 72.9M Sites, Counted From One Response Header

As of November 16, 2026, webatla counts 72,864,210 websites running Nginx, 22.36% of the 325.8 million websites it tracks. webatla's Nginx technology page carries the live figure and updates it daily.

Nearly all of that figure rests on one field of the HTTP response, and you can read that field yourself with a single curl call. One limit applies to everything below: webatla has no history series for Nginx, so these Nginx usage statistics describe a single day, and none of them is a trend.

One header line behind 72.9 million sites

webatla's pages on web servers are open about the method. The Tengine, Angie and Caddy pages all say that identifying a server depends on the name it sends in a response header, and that operators can strip or change that name. Asking how many websites use Nginx therefore comes down to asking how many servers put that word in one header. A domain whose response names Nginx is counted once under Nginx, whatever HTTP status it returned.

In webatla's catalog of 3,083 technologies, only three entries reach more sites than Nginx: Favicon (a marker in the page HTML that points to the site icon), Cloudflare and HTTP/2. Favicon describes the page itself, so the chart below leaves it out and keeps the CDN, protocol, security header and server entries.

Lollipop chart of websites per entry: Cloudflare 83.7M, HTTP/2 75.9M, Nginx 72.9M, HTTP/3 66.7M, HSTS 55.4M, Apache HTTP Server 39.7M, OpenResty 22.0M, LiteSpeed 5.77M

Websites per entry for the largest CDN, protocol, security header and server items in the webatla catalog. Nginx sits third among them, behind Cloudflare and HTTP/2.

Cloudflare (83,716,411 sites) and HTTP/2 (75,872,858) are the only delivery entries above it. HTTP/3 follows on 66,687,954 sites and the HSTS header on 55,396,680. The nearest web server, Apache HTTP Server, has 39,735,437 sites, so Nginx is about 1.8 times larger. As Nginx market share of the whole index, that is 22.36%, against 12.20% for Apache and 6.76% for OpenResty.

Check one host from your terminal

You do not need a scanner to see the same signal for a single site. A HEAD request, filtered down to the status line and the server line, is enough.

curl -sI https://example.com | grep -iE '^(HTTP|server)'
Enter fullscreen mode Exit fullscreen mode

Each common answer also tells you whether a header-based count can see the site.

  • server: nginx with no number is Nginx with its release hidden, usually through server_tokens off. The site still counts as Nginx and only drops out of the version table.
  • server: nginx/1.24.0 (Ubuntu) is a distribution build. Ubuntu's package appends its own name after the release.
  • server: openresty or server: Tengine is a server built on Nginx code that reports its own name, so it is counted under that name.
  • server: cloudflare means you reached Cloudflare's edge. The site is filed under Cloudflare, and whatever runs at the origin stays out of sight. webatla detects Cloudflare on 83,716,411 websites, 25.70% of the index.
  • A name you do not recognize, or no server line at all, means the field was changed or removed somewhere in the path, or the server never sent one. Stock Nginx can only hide its release number. Renaming the server or clearing the field takes a third-party module such as headers-more.

In browser DevTools the same value is under the Network tab, on the document request, in the response headers. Treat it as a claim and not as proof. The field is free text, and whoever runs the box can set it to anything.

One case never reaches your terminal at all. An Nginx instance that works as a load balancer or TLS terminator behind another public layer does not answer outside requests, so no scan records it. All of these gaps leave Nginx sites out of the count, and none of them can be corrected from outside.

Nginx as the front door for an application

The catalog files Nginx under two categories, Web servers and Reverse proxies, and the second one explains a lot of the count. A common layout puts Nginx on ports 80 and 443 to terminate TLS and serve static files, then hands everything else with proxy_pass or fastcgi_pass to an application server such as Node.js, gunicorn, PHP-FPM or Kestrel.

That layout shapes what a scan sees. By default the Nginx proxy module does not pass the upstream's own Server header to the client, so the response names Nginx and the application stays hidden. It is one reason application servers look small in header data. Node.js appears on 3,863,899 sites and Kestrel, the ASP.NET Core server, on 3,755,298. Kestrel's page on webatla says it only becomes visible when something in the response path lets its signature through.

A detection means Nginx answered the request. It does not tell you whether Nginx serves a static page by itself or forwards traffic to a Django app three hops behind it.

What the release number gives away

Stock Nginx prints its release in the Server header unless told otherwise. Even so, most sites do not show one. webatla's Nginx page reports version shares only for sites with a known version, and it lists 1.28.3 on 7,391,017 sites at 43.0%. Dividing one by the other gives roughly 17.2 million Nginx sites that print a release, about 23.6% of the total. The remaining 55.7 million or so send the bare name.

Lollipop chart of the estimated share of sites that print a version: IIS 98.8%, OpenResty 42.0%, Nginx 23.6%, Apache HTTP Server 19.0%, Angie 3.0%

Estimated share of each server's sites that print a release number, worked back from the top release and its share on each webatla page.

Among the five server pages checked here that carry a version table, Nginx sits in the middle. IIS prints a release on nearly all of its sites (about 98.8%), OpenResty on about 42%, Apache HTTP Server on about 19% and Angie on about 3%. These are estimates built from rounded shares, so read them to the nearest point or two.

Which Nginx sites print a number is not random either. Turning the number off is a standard hardening item, so the servers that keep printing it lean toward the ones that get less care. webatla's Nginx page flags the same bias in its own version table, so treat that table as a tilted sample of the installed base.

Even and odd minor numbers

On nginx.org, an even minor number (1.26, 1.28, 1.30) marks a stable branch, and an odd one (1.27, 1.29, 1.31) marks mainline, where new features land first. nginx.org itself maintains only two branches at a time, the newest even line and mainline. Distribution packages get their fixes from the distribution instead. Grouping the 25 most common version strings by minor line shows which numbering the installed base sits on.

Ranked list of Nginx minor lines by sites printing a version, even lines in green and odd mainline lines in grey: 1.28 8.43M, 1.18 1.37M, 1.24 1.32M, 1.20 901K, 1.30 871K, 1.29 760K, 1.14 689K, 1.31 564K, 1.22 433K, 1.26 422K, five smaller lines 518K

Nginx sites per minor line from the 25 most common version strings, split by whether the minor number is even (stable type) or odd (mainline). Even lines hold about nine sites in ten.

Stable-numbered lines account for 14,711,082 sites in the top 25 and mainline-numbered lines for 1,567,434, about nine to one. That ratio says more about packaging than about anyone picking a branch: about a third of the even-numbered sites carry a distribution's default number (next section). The 1.28 line has 8,430,194 sites by itself, more than the other fourteen lines in the top 25 put together (7,848,322). Behind those 25 strings sits a long tail: webatla sees 412 distinct version strings in total, and the other 387 share roughly 0.9 million sites.

Old numbers can sit on patched binaries

Several of the most common strings match the default nginx package of a Debian, Ubuntu LTS or RHEL release.

Version string Sites printing it Same number as the default package in
1.18.0 1,370,772 Ubuntu 20.04 and 22.04, Debian 11
1.24.0 1,319,429 Ubuntu 24.04
1.20.1 804,179 RHEL 9
1.14.0, 1.14.1, 1.14.2 688,973 Ubuntu 18.04, RHEL 8, Debian 10
1.22.1 432,769 Debian 12
1.26.3 315,937 Debian 13
1.6.2 131,118 Debian 8
1.10.3 71,298 Debian 9

Together that is 5,134,475 sites, about 30% of those that print a release. A matching number does not prove the distribution, since anyone can compile 1.18.0 from source. For strings this common, though, the package is a fair first guess.

It matters because distributions backport security fixes into their packages and keep the upstream number, for as long as the release is supported. An Ubuntu 22.04 server with current updates and a forgotten one that has not been touched in years both answer with the same string. The header cannot tell them apart, and neither can any scan that reads it. The same caution applies to raw counts of old releases. Among the 25 most common strings, at least 2,334,103 sites print a number below 1.20, and 2,262,161 of them carry one of the distribution numbers in the table. For those hosts, the question worth asking is whether the operating system still receives updates and whether they are installed.

Skills that carry over to other servers

The configuration language reaches further than the Nginx count. OpenResty is the Nginx core with a Lua runtime added, Tengine is a fork maintained under Alibaba and Taobao, and Angie is a fork started in 2022 by engineers who had worked on the original. All three read the same style of configuration, and the Angie page on webatla notes that an operator can swap binaries and keep the existing config files.

Four stat cards: 72.9M sites answer as Nginx, 22.0M as OpenResty, 1.77M as Tengine, 117K as Angie, 96.8M detections in all against 39.7M for Apache HTTP Server

Sites that answer as Nginx or as a server built on its code. The 96.8 million total counts detections, and one site can carry two of these names.

The four names add up to 96,766,060 detections, about 2.4 times the 39,735,437 for Apache HTTP Server. That sum counts detections and overstates distinct sites. In the British Virgin Islands and in Slovakia, the Nginx and OpenResty counts together exceed all websites hosted in the country, so some sites carry both labels.

If you write tooling, modules or config generators for this family, expect a long tail of releases. No single Nginx release covers more than 43.0% of the sites that print a version, while 1.31.1.1 alone covers 80.9% of versioned OpenResty sites.

Two location fields, two different answers

webatla keeps two kinds of location for each site: the server's country, by IP address, and the domain's extension. For Nginx they often disagree, and the one you filter on changes the result.

Dumbbell chart of the Nginx share by server country and by ccTLD: Russia 61.5% and 61.6%, Sweden 58.9% and 42.8%, Netherlands 41.3% and 24.6%, United Kingdom 38.4% and 19.1%, China 36.9% and 45.4%, Canada 10.1% and 16.2%

Nginx share of websites counted by server country and by country code domain. Russia matches, while Sweden, the Netherlands and the UK split widely.

Russia is the control case. Nginx runs on 61.5% of websites hosted there and on 61.6% of .ru domains. Elsewhere the two fields split. Servers in Sweden, the Netherlands and the United Kingdom run Nginx far more often than domains on .se, .nl and .uk do, for example 41.3% against 24.6% in the Dutch case. China and Canada go the other way, with .cn at 45.4% against 36.9% for servers located in China.

The data does not say why each gap exists, and explaining them one by one would be guesswork. A sample of "Swedish Nginx sites" built from IP addresses and one built from .se domains are different populations, so say which one you used. For scale, the largest single slices are servers in the United States (22,978,763 Nginx sites, 35.7% of those with a known country) and .com domains (30,040,058, 41.2% of all Nginx sites). The full Nginx statistics article has the country and TLD tables.

A checklist for your own Nginx hosts

You can do all of this from a shell. If you maintain Nginx, a few habits follow from how these numbers are produced.

Read your own header from outside. Run the curl line above against your public hostname. Whatever it prints is what a header-based scanner records about you.

Decide on the release number on purpose. Hiding it takes one directive in the http block of nginx.conf, after which the header says just nginx.

server_tokens off;
Enter fullscreen mode Exit fullscreen mode

Run nginx -t before you reload. The directive also drops the release from Nginx's default error pages.

Do not confuse hiding with patching. A hidden number changes what outsiders see and nothing else. On the host, nginx -v shows the release the binary reports even when the header hides it. For a distribution package, the package version from apt policy nginx or rpm -q nginx and the vendor's security notices say more than the upstream number.

Test the origin as well as the edge. If a CDN sits in front, its header may describe the CDN rather than your origin. The server behind it needs its own checks.

FAQ

Is the Nginx count complete?

No. It covers sites where Nginx answers the request and names itself. Sites behind a CDN that replaces the Server header (Cloudflare does), internal Nginx instances and edited headers are missing, so 72,864,210 is a lower bound.

Which Nginx release number turns up most often?

1.28.3, on 7,391,017 sites. That is 43.0% of the Nginx sites that print a version and about 10% of all Nginx sites, since roughly three in four print none.

Does nginx/1.18.0 in a header mean the server is vulnerable?

Not by itself. 1.18.0 is also the default package in Ubuntu 20.04 and 22.04 and in Debian 11, and those packages receive backported fixes under the same number for as long as the release is supported. Only the package version on the host, and whether that release still gets updates, settles it.

Does server_tokens off hide Nginx completely?

No. It removes the release number, but the header and the default error pages still say nginx. In the open-source build, changing or clearing the name itself takes a third-party module such as headers-more, or a patched build.

Sources

Top comments (1)

Collapse
 
lin7c profile image
xcl •

The version-number caveat is the most useful part: 1.18.0 being the most common version does not mean those servers are unpatched, since Debian and Ubuntu backport fixes without bumping the reported version. That also makes any count of "old" versions hard to interpret — it counts reported version strings, not actual patch state. Is there a way to separate distro-packaged Nginx from self-built installs in the header data, for example via distro-specific version strings or build markers? The server_tokens off recommendation follows from that uncertainty as well, since a version header that cannot be trusted is arguably worse than no version at all.