Human Written & Fact Checked · Updated November 2026

100 Web Server Statistics for 2026

The answer to which web server leads depends on what you count. Responding sites, active websites, domains and computers each expose a different part of the delivery infrastructure.

This page keeps those measurements separate. It combines original surveys with observed protocol and connection data, without treating a visible server as proof of the underlying hosting provider.

Headline figures

Global survey responses: Surveyed sites: 1,470,435,351 (September 2026). (Netcraft, Sep 2026)
Global survey responses: Web-facing computers: 15,463,120 (September 2026). (Netcraft, Sep 2026)
Global survey responses: Cloudflare active-site share: 25.57% (September 2026). (Netcraft, Sep 2026)
Global survey responses: nginx computer share: 44.00% (September 2026). (Netcraft, Sep 2026)
Global measured website population: Cloudflare Server was detected on 33.5% of websites with a known web server (9 November 2026). (W3Techs, Nov 2026)
Global measured website population: LiteSpeed was detected on 13.7% of websites with a known web server (9 November 2026). (W3Techs, Nov 2026)
Global measured website population: Cloudflare Server usage among the top 1,000 websites with known server software: 54.2% (24 October 2026). (W3Techs, Oct 2026)
Global measured website population: Unix was detected on 92.2% of websites with a known operating system (9 November 2026). (W3Techs, Nov 2026)
Global measured website population: Mobile CDN HTML requests using HTTP/3: 29% (July 2025 dataset). (HTTP Archive, Jan 2026)
Global measured website population: Mobile Cloudflare document requests using HTTP/3: 49.9% (July 2025 dataset). (HTTP Archive, Jan 2026)
Global measured website population: Mobile CDN requests using Zstandard: 12% (July 2025 dataset). (HTTP Archive, Jan 2026)
Global measured website population: Mobile third-party CDN requests using TLS 1.3: 96% (July 2025 dataset). (HTTP Archive, Jan 2026)

How do website and software surveys describe server footprints?

A survey's denominator determines what its ranking means. Responding sites, domains and detected software describe related infrastructure, but none is a direct count of hosting customers.

These September observations come from Netcraft's responding-site survey. Counts and shares describe the same developer footprint instead of independently corroborating it. Multiple sites can belong to one organisation, and visible software can be an intermediary. Neither the count nor the percentage establishes customer revenue, active publishing or origin-server location. Survey totals can also change as responding hostnames appear, disappear or change their detected technology.

How do the domain counts and shares compare?

Surveyed domains: 323,147,961. (Netcraft, Sep 2026)
OpenResty domain share: 17.91%. (Netcraft, Sep 2026)
nginx domain share: 17.44%. (Netcraft, Sep 2026)
Cloudflare domain share: 14.25%. (Netcraft, Sep 2026)
Apache domain share: 13.94%. (Netcraft, Sep 2026)
OpenResty: 57,881,995 domains. (Netcraft, Sep 2026)
nginx: 56,342,998 domains. (Netcraft, Sep 2026)
Cloudflare: 46,044,142 domains. (Netcraft, Sep 2026)
Apache: 45,038,770 domains. (Netcraft, Sep 2026)

The domain denominator reduces the effect of multiple responding sites under one domain. These Netcraft measurements therefore answer a different question from the all-site table. They are not verified counts of unique customers or physical machines. A domain can use shared infrastructure and several technologies; the figures do not prove an exclusive commercial relationship.

Which other software and operating systems are detected?

Nginx 29.9% of known-server sites. (W3Techs, Nov 2026)
Apache 21.0% of known-server sites. (W3Techs, Nov 2026)
Node.js 7.5% of known-server sites. (W3Techs, Nov 2026)
Microsoft-IIS 2.9% of known-server sites. (W3Techs, Nov 2026)
Caddy 1.2% of known-server sites. (W3Techs, Nov 2026)
Google Servers 1.1% of known-server sites. (W3Techs, Nov 2026)
Envoy 0.5% of known-server sites. (W3Techs, Nov 2026)
Railway edge proxy 0.1% of known-server sites. (W3Techs, Nov 2026)
Tengine 0.1% of known-server sites. (W3Techs, Nov 2026)
IdeaWebServer 0.1% of known-server sites. (W3Techs, Nov 2026)
ArvanNginx 0.1% of known-server sites. (W3Techs, Nov 2026)
Kestrel 0.1% of known-server sites. (W3Techs, Nov 2026)
Windows 8.0% of known-OS sites. (W3Techs, Nov 2026)

W3Techs measures known-server websites; its operating-system observation instead uses websites with a known OS. These November snapshots have different detection and sampling rules from Netcraft's September survey. Multiple technologies can coexist, so shares need not be exclusive. They describe detected software, not software quality or hosting revenue.

What changes across activity, popularity and computer measurements?

Activity filters and popularity thresholds select different website populations. Computer observations introduce an infrastructure unit, so comparisons need both the selection rule and the measurement date.

What do active sites and the busiest sites show?

nginx active-site share: 15.93%. (Netcraft, Sep 2026)
Apache active-site share: 12.14%. (Netcraft, Sep 2026)
LiteSpeed active-site share: 5.70%. (Netcraft, Sep 2026)
Cloudflare: 55,639,579 active sites. (Netcraft, Sep 2026)
nginx: 34,649,380 active sites. (Netcraft, Sep 2026)
Apache: 26,403,710 active sites. (Netcraft, Sep 2026)
LiteSpeed: 12,411,998 active sites. (Netcraft, Sep 2026)
Cloudflare busiest-million share: 29.10%. (Netcraft, Sep 2026)
nginx busiest-million share: 19.65%. (Netcraft, Sep 2026)
Apache busiest-million share: 15.01%. (Netcraft, Sep 2026)
LiteSpeed busiest-million share: 3.66%. (Netcraft, Sep 2026)

Netcraft's active-site classification and busiest-million selection are different filters. Counts and shares of active sites describe the same population. A higher share of busy sites does not prove superior performance: workloads and engineering resources differ. Neither filter identifies every technology behind the visible delivery layer or the number of machines serving a website.

How do detected-server shares vary by popularity threshold?

Cloudflare Server: top 1,000,000 known-server sites: 47.6%. (W3Techs, Oct 2026)
Cloudflare Server: top 100,000 known-server sites: 53.4%. (W3Techs, Oct 2026)
Cloudflare Server: top 10,000 known-server sites: 55.1%. (W3Techs, Oct 2026)
Nginx: top 1,000,000 known-server sites: 27.2%. (W3Techs, Oct 2026)
Nginx: top 100,000 known-server sites: 27.1%. (W3Techs, Oct 2026)
Nginx: top 10,000 known-server sites: 30.2%. (W3Techs, Oct 2026)
Nginx: top 1,000 known-server sites: 31.8%. (W3Techs, Oct 2026)
Apache: top 1,000,000 known-server sites: 16.7%. (W3Techs, Oct 2026)
Apache: top 100,000 known-server sites: 14.4%. (W3Techs, Oct 2026)
Apache: top 10,000 known-server sites: 11.8%. (W3Techs, Oct 2026)
Apache: top 1,000 known-server sites: 9.8%. (W3Techs, Oct 2026)

These W3Techs observations use its October ranking snapshot and known-server denominator. The rank thresholds are nested, not separate segments to add together. They differ from Netcraft's busiest-site classification and reporting month. A shift across thresholds describes the selected sample, not a measured improvement in software speed, reliability or customer satisfaction.

What do web-facing computer observations measure?

Apache computer share: 17.70%. (Netcraft, Sep 2026)
Microsoft computer share: 5.41%. (Netcraft, Sep 2026)
nginx: 6,803,521 computers. (Netcraft, Sep 2026)
Apache: 2,737,362 computers. (Netcraft, Sep 2026)
Microsoft: 836,263 computers. (Netcraft, Sep 2026)
Web-facing computer count increased by 208,114 during the month. (Netcraft, Sep 2026)

A web-facing computer is Netcraft's infrastructure observation, not a hardware sales unit. Virtualisation, shared hosting and network-facing services complicate translation into physical servers. The developer shares use this computer denominator instead of websites or domains. The monthly change is a survey observation, not evidence of equipment purchases. Site counts can move differently when the number of sites served by each observed computer changes. These measurements therefore help describe infrastructure scale while leaving ownership, physical location and workload capacity unresolved.

Which HTTP protocols do measured servers actually serve?

Observed protocol use is narrower than technical support. A server can support a newer protocol even when a particular crawl negotiates an older one.

How do protocol shares differ between origins and CDNs?

Mobile CDN HTML requests using HTTP/2: 69% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile CDN HTML requests using HTTP/1.1: 2% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile origin HTML requests using HTTP/2: 79% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile origin HTML requests using HTTP/1.1: 21% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile third-party CDN requests using HTTP/3: 45% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile third-party origin requests using HTTP/3: 7% (July 2025 dataset). (HTTP Archive, Jan 2026)

HTTP Archive measures browser requests under its crawl conditions. Initial HTML requests and later third-party resource requests have different opportunities to negotiate a protocol. The origin and CDN categories are determined by the study's detection method and do not constitute randomised experimental groups. These values therefore describe the measured traffic, instead of global internet traffic or every server's supported feature set. They should not be interpreted as a direct estimate of the performance gain a website would receive by enabling a protocol.

What document-request rates appear for Cloudflare, CloudFront and Sucuri?

Desktop Cloudflare document requests using HTTP/3: 44.0% (July 2025 dataset). (HTTP Archive, Jan 2026)
Desktop Amazon CloudFront document requests using HTTP/3: 3.2% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile Amazon CloudFront document requests using HTTP/3: 3.8% (July 2025 dataset). (HTTP Archive, Jan 2026)
Desktop Sucuri Firewall document requests using HTTP/3: 0.5% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile Sucuri Firewall document requests using HTTP/3: 0.6% (July 2025 dataset). (HTTP Archive, Jan 2026)

A document request is the page document instead of the collection of subresources subsequently fetched. That distinction counts because connection reuse and protocol discovery can affect later requests. The mobile and desktop observations also reflect the test configurations and site populations used in the crawl. They do not establish an intrinsic provider speed ranking. The largest mobile document observation is retained in the takeaways. Compare these results only within their stated request type and dataset, without substituting a provider's total network traffic as the denominator.

What document-request rates appear for other measured platforms?

Desktop Akamai document requests using HTTP/3: 0.2% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile Akamai document requests using HTTP/3: 0.2% (July 2025 dataset). (HTTP Archive, Jan 2026)
Desktop QUIC.cloud document requests using HTTP/3: 0.1% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile QUIC.cloud document requests using HTTP/3: 0.1% (July 2025 dataset). (HTTP Archive, Jan 2026)
Desktop Google document requests using HTTP/3: 0.1% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile Google document requests using HTTP/3: 0.1% (July 2025 dataset). (HTTP Archive, Jan 2026)

Low observed use does not prove that a provider lacks protocol support. Initial connection behaviour, site configuration and the measurement approach can all affect the negotiated protocol. Identical rounded values across device classes are also not evidence of identical underlying workloads. These rows are an original crawl observation, not a certification test of each platform's features. They should remain separate from subresource adoption figures, which use another request population. Website owners need direct tests of their own delivery path before drawing an operational conclusion.

How do compression and TLS measurements compare?

Compression and connection security concern separate layers of delivery. Their adoption and timing measurements can describe observed infrastructure without proving a causal performance advantage.

Which compression formats appear in encoded responses?

Mobile CDN requests using Brotli: 46% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile CDN requests using gzip: 42% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile origin requests using Brotli: 39% (July 2025 dataset). (HTTP Archive, Jan 2026)
Mobile origin requests using gzip: 61% (July 2025 dataset). (HTTP Archive, Jan 2026)

The comparison concerns the measured compressed-response categories. It should not be relabelled as the percentage of all websites using a compression algorithm. A website can deliver different encodings for different resources or clients. The newer encoding observation is kept in the takeaways. Its use does not independently measure bytes saved or rendering speed: content type, compression settings and the original payload also matter. The source distinguishes CDN and origin responses, but these are observed deployment groups instead of equivalent workloads assigned randomly to different delivery systems.

What do TLS versions and desktop negotiation timings show?

Mobile third-party origin requests using TLS 1.3: 78% (July 2025 dataset). (HTTP Archive, Jan 2026)
Median desktop CDN TLS negotiation: 57 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
Median desktop origin TLS negotiation: 177 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
90th-percentile desktop CDN TLS negotiation: 89 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
90th-percentile desktop origin TLS negotiation: 277 ms (July 2025 dataset). (HTTP Archive, Jan 2026)

These timings come from simulated browser connections under controlled datacenter conditions. Median and upper-percentile values describe different points in a distribution and should not be added or averaged into an undocumented summary. The source's browser cannot successfully negotiate obsolete TLS versions that it no longer supports, so the successful-request data is not an inventory of all legacy servers. The third-party origin version share has a separate request denominator. These measurements cannot predict the connection time for every user, geography or device.

What do mobile TLS negotiation timings show?

Median mobile CDN TLS negotiation: 183 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
Median mobile origin TLS negotiation: 302 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
90th-percentile mobile CDN TLS negotiation: 216 ms (July 2025 dataset). (HTTP Archive, Jan 2026)
90th-percentile mobile origin TLS negotiation: 416 ms (July 2025 dataset). (HTTP Archive, Jan 2026)

The mobile results retain the study's simulated conditions and first-connection context. They are not measurements from a representative panel of mobile users in every country. Network distance, device capabilities, session resumption and deployment configuration can change actual experience. The CDN and origin groups also contain different websites, so the observed difference is not a controlled estimate of a migration benefit. Report the median and upper percentile separately to preserve the distribution. A site-specific performance decision should use measurements from the intended audience and delivery path.

Glossary

Methodology and sources

Every figure on this page comes from the organisation that produced it: regulators, statistics offices, public bodies, research institutes, and companies reporting their own data or surveys. Each one links to the original report or dataset and carries its release date. Survey results keep their population and fieldwork period, and financial figures keep the measure the producer used. Figures from different producers are not added together or ranked against each other. The page is reviewed monthly and updated when a producer releases newer data.

HTTP Archive · Jan 2026
Netcraft · Sep 2026
W3Techs · Nov 2026
W3Techs · Oct 2026
W3Techs · Nov 2026