Download websites that use HTTP/3

About HTTP/3

HTTP/3 is young enough that its numbers still move for a single reason. A provider changes a default. The protocol arrived in 2022 and runs on QUIC over UDP instead of the TCP foundation its predecessors used, which removes a class of stalling on lossy networks and shortens connection setup. Site owners did not choose any of that. Edge networks did, and the adoption curve is really a record of their release schedules.

As of October 2026 we complete a HTTP/3 handshake with 65,982,998 of the sites we crawl, which puts a recent standard in the same order of magnitude as things that have been ordinary for years. The speed is explained by concentration rather than enthusiasm. When a few large edge platforms enable the protocol for every customer at once, entire customer bases acquire the marker in a week without a single administrator doing anything at all.

The Miscellaneous classification hides how commercial the field is. For a delivery network sales team, the list of domains that do not yet negotiate the protocol is a prospect list, because the domains that do are already served by somebody who ships it. For a performance consultancy, the same split identifies sites still paying a latency cost on mobile networks. For a researcher measuring standards diffusion, the file is the raw material for the only honest version of that measurement, which is a count across the whole population rather than a sample of well known sites.

United States accounts for 43,228,088 of the domains where the handshake succeeds, and the suffix weight sits on .com. The tilt has the same cause as everything else here. New transports appear first on platforms that operate their own edge, and those platforms are registered in a small number of markets while serving the rest of the world from one configuration. A country column derived from serving infrastructure is a fact about where connections terminate, so any national reading needs the domain suffix and the language of the page beside it.

A lookup on one domain tells you whether one server answered. The interesting questions are all comparative. Which hosting providers have flipped the default and which have not. How the adoption rate differs between suffixes and between markets. Whether a portfolio of sites moved uniformly or in pockets, and how any of that changed between one snapshot and the next. Comparative questions need the whole population in a file, with the transport column sitting next to the server, delivery network and country columns so the crossings can be made locally and repeated later.

Detection here is an observation and not a capability audit. We record that our crawler negotiated the protocol from our network at a point in time, so a server that advertises support but fails to complete the exchange for us, or that offers it only to clients we do not imitate, is recorded as absent. The advertisement browsers use to discover support is a hint rather than a guarantee, and a site can speak the new transport at the edge while the origin behind it speaks something much older. Absence in the file means unobserved, which is a weaker and more honest claim than unsupported, and treating the two as the same thing will overstate every gap you find.