Web server shares in 2026: why nginx is 14% and 44% at the same time

Netcraft’s July 2026 survey publishes five populations, and nginx looks different in each — inside one report:

From 14.45% to 43.92% — a threefold spread, and every figure is correct. They answer different questions: on how many names, on how many live sites, on how many physical machines. Any lone "nginx share is X%" without a named population is meaningless.

Worth remembering separately: in the all-names view the "other" bucket holds 34% — more than any named vendor. A four-slice pie chart implies a completeness that does not exist.

W3Techs as of 27 August 2026:

The causes are methodological, not errors:

• Different universes. Netcraft probes 1.49 billion hostnames, most of them parked or machine-generated. W3Techs deliberately excludes those and works with 20M+ sites that carry content. Apache and LiteSpeed sit mostly on live shared-hosting sites and therefore roughly double in W3Techs.
• LiteSpeed is the sharpest divergence (3.88% vs 14.8%). It concentrates in cPanel hosting: dense in W3Techs’ population, near-invisible across Netcraft’s sea of names.
• W3Techs allows several servers per site — its column sums to about 111%, so it is not a partition of a whole at all. Netcraft assigns one server per name.
• Both count reverse proxies as servers. An Apache site behind Cloudflare is recorded as Cloudflare. What runs at the origin is measured by nobody.

No single honest "web server market share" can be derived from these sources. The correct presentation is both, side by side, with populations named — not a winner.

W3Techs offers two Russian cuts as of 27 August 2026, and they are different populations rather than two estimates of one:

A .ru domain on German hosting lands in the first column and not the second; a .com run from Moscow, the reverse. The denominator for both is sites whose server could be identified, not all sites in the segment.

Source limitation: only the top three positions per segment are published free. There is no open figure for LiteSpeed, IIS, Cloudflare or Angie in Russia — not from W3Techs, not Netcraft, not Cloudflare Radar.

On Angie, the Russian nginx fork: W3Techs tracks it but shows it in the tail below 0.1%. No more precise public figure exists, despite the project’s visibility in the trade press.

Jumps that are not growth. In the domains view OpenResty gained 9.01 percentage points in a single month — from 15.1 to 42.7 million domains. Organic growth does not look like that: almost certainly one large host changed a header. Narrating it as a trend is a mistake.

Both figures move. W3Techs updates daily, Netcraft monthly. Any share without a retrieval date goes stale silently. That is exactly how statistics turn wrong: a figure from two years ago keeps circulating through listicles labelled "2026".

And one note on what these data do not contain at all: contrary to common belief, HTTP Archive and the Web Almanac have no web server chapter. Raw Server headers exist in their dataset, but no published statistic does.

To see what your own site sends, use the response header check.