2025 Self-Host User Survey

The annual reader survey run by selfh.st (Ethan Sholly), published 21 November 2025: 4,081 completed responses, with the full response counts published as JSON in the project's own CDN repository, so every figure below is countable rather than relayed.

It is the first source in this wiki that answers who runs this stuff with numbers instead of GitHub stars. It is also the first to make plain how narrow the answer is.

Who answered

84% self-host for personal purposes only (3,417 of 4,081); 16% for a mix of personal and work. Eight respondents — 0.2% — described their setup as enterprise. Among the personal respondents, 1,977 of 4,069 setups serve one or two people, and the single largest bucket is one user (1,123).

The reasons given are not the ones the marketing uses. Multi-select, 18,041 selections across 4,081 people: hobby 3,862 · privacy 3,520 · cost 2,596 · convenience 2,584 · education 2,474 · flexibility 2,374 · business 525. Hobby beats privacy, and both beat cost. That reorders the pitch self-hosting records: this population is not mainly escaping a subscription, it is running a hobby that happens to produce a service.

The sample is the finding's main limit and the survey does not hide it. Respondents came from a self-hosting newsletter's own audience — 81% work in a technology field, 94% are male, 88% are aged 20–50, and they are more educated than the general population. So the numbers describe an enthusiast community, not the households and small businesses the open-source-alternative pitch is aimed at. Nothing here samples a person who installed one container because a friend set it up.

What they run it on

Linux-based and appliance operating systems dominate. 10,488 selections: Linux 3,317 · Proxmox 1,832 · Home Assistant OS 1,199 · Raspberry Pi OS 979 · TrueNAS 774 · Unraid 412 · Synology DSM 398 · Windows desktop 373. Three of the top seven are purpose-built appliance platforms rather than general OSes, which says the tooling has moved a long way from "a spare PC under the desk".

Hardware skews cheap and pre-built: consumer hardware and low-powered devices (the Raspberry Pi class) lead, ahead of custom builds, cloud VPS and enterprise gear.

Containers are effectively universal. 96% deploy containers (3,932 of 4,081); given the choice, 69% prefer a container to bare metal and only 4% prefer bare metal. Docker 3,637 and LXC 1,256 carry the field, Podman 446 behind them. 77% use no orchestration at all — Kubernetes and friends are a 23% minority even in a population this technical.

Half of container users add management software on top, and among those, three tools take 82%: Portainer 1,336 · Dockge 425 · Komodo 308.

Updates are done by hand. 45% update manually, 40% "depends", 15% automatic — the survey attributes the reluctance to breaking changes. That is a direct measurement of the maintenance burden self-hosting argues arrives after the decision, and it lands on the pessimistic side: the people best equipped to automate updates mostly decline to.

Exposure and backup

88% use a custom domain. Remote access is 7,122 selections: VPN 2,902 · reverse proxy 2,716 · forwarded ports 721 · LAN only 510. Reverse proxies split five ways with no winner — NGINX Proxy Manager 842 · Traefik 751 · Caddy 598 · Cloudflare Tunnels 563 · NGINX 528 · Pangolin 290.

Backups: onsite 3,054 · cloud 2,100 · offsite 1,029 · none 542. The survey's own note is that 34% keep their data in exactly one place. And 17% deliberately keep sensitive data off their own hardware, citing security and backup worries — a self-hoster's vote of no confidence in self-hosting, from inside the sample.

86% do not self-host e-mail. The one service everybody agrees is not worth it.

What they run

Media streaming, 6,011 selections: Jellyfin 2,487 · Plex 1,219 · AudioBookshelf 840 · Navidrome 656 · Kodi 237 · Emby 149. Files: Nextcloud 1,236 · Syncthing 1,058 · Synology 550 · FileBrowser 481. Git: Gitea 849 and Forgejo 613 take 78% between them, GitLab 302 behind. Home automation is deployed by 63%.

Authentication is where the fragmentation shows: 2,261 selections and the largest single entry is "other" at 737 — Authelia 390, Pocket ID 310, authentik 268, Pangolin 241, Keycloak 158 follow. Nothing has won.

Asked for a favourite, 3,678 people named 340 different applications; 313 of them drew fewer than 20 votes each. The head is short — Home Assistant 588, Jellyfin 522, Immich 429, Plex 208, Vaultwarden 152 — and the tail is nearly everything.

59% have not donated to any self-hosted project in the past year.

What this closes, and what it does not

Open question 3 — who actually runs any of this — moves from unanswerable to partly answered. The corpus previously had stars, forks and commits, and project-quality-score showed why: download counts do not exist for software shipped as an app or a compose file. A survey routes around that by asking people. The cost is that it counts respondents, not installations, and the respondents are a self-selected slice of one newsletter's readers.

Open question 2 — what happens when the maintainer stops — gets nothing. The survey asks what people run now. It has no tenure question, no churn question, and no question about a service that was abandoned or a project that died under its users. The one adjacent number is that 542 people have no backup at all, which says something about what an abandonment would cost them and nothing about how often it happens.

Two smaller readings:

• The enterprise number is 8. Whatever ever-gauzy-class business platforms are for, this population is not it, and a survey of hobbyists cannot be used as evidence about open-source-alternative adoption in business.
• Proxmox at 1,832 and Home Assistant OS at 1,199 put two appliance platforms ahead of Raspberry Pi OS. The trend in this community is towards prebuilt hosting substrates, not towards hand-rolled machines — the opposite of what "self-hosting" connotes.