# Self-hosting in 2026: which software is worth running and which isn't

> A guide to deciding which services are worth self-hosting in 2026, which ones don't pay off, and how much maintenance a home server of your own demands.
> By José Maldonado · Technology · 2026-10-19

The return to local is no longer just a hobby. Fast Company published an analysis on October 14, 2026 about repatriating workloads from the public cloud, and a day earlier a developer documented on dev.to how he replaced eleven subscriptions with alternatives on his own server. A company reviewing its bill and a person who wants to stop paying: opposite ends, same direction.

The [Flexera State of the Cloud 2025 report](https://www.flexera.com/about-us/press-center/new-flexera-report-finds-84-percent-of-organizations-struggle-to-manage-cloud-spend), with 759 respondents, puts the share of workloads that have already come back from the cloud at 21% and the share citing spending as their top problem at 84%. At home the phenomenon is smaller: Lea Gröber and her team estimated in [USENIX Security 2024](https://www.usenix.org/conference/usenixsecurity24/presentation/gr%C3%B6ber-private-clouds) that 8.4% of the US self-hosts at least one service, with the web as the most common use. And the uncomfortable finding is the profile: people who self-host stand out not for their privacy, but for their systems training.

A server of your own doesn't raise a debate about the cloud, but a question about the work you accept and what happens the day something breaks. The services that justify that cost justify it comfortably; the ones that don't just trade one dependency for another.

## What counts as self-hosting?

Self-hosting means the service runs on infrastructure you control: a mini PC in the living room, a container on a cheap VPS or a rented machine. What decides the label is who keeps the copy and who decides when it gets updated.

A manufacturer's NAS with the vendor's cloud enabled is out. So is a provider's plan where the software is managed for you, even if the code is free: there you delegate what defines home self-hosting.

The category includes hosting managed by you on someone else's server: paying for a VPS and deploying your instance there is not the same as buying the provider's plan, where you purchase support in exchange for those tasks.

In the first case, you update, you back up and you answer for it; in the second you buy support in exchange for those tasks. Free software without a server isn't self-hosting either, even if it shares the philosophy.

## Is it worth setting up a server of your own?

The answer depends on three variables and their order. The first is the cost of the subscription versus keeping the service for three years. The second is the volume and growth of the data. The third is tolerance for an outage.

The usual mistake is buying the server because it's fun and then looking for tasks for it: like renting a premises before you have a business. It's good for learning, and that counts, but it isn't efficiency.

Criterion

Question it answers

Signal in favor

Cost

Does the subscription exceed keeping it for three years?

High, sustained monthly fee

Data

How much does it take up today and in two years?

Photos, video, your own files

Dependency

What happens if it goes down for a weekend?

You can wait without harm

Learning

Do you want to administer systems even if you don't save?

It's a goal in itself

The first two are worked out with paper and calculator. The third separates the projects that survive from the ones that get abandoned, because it forces you to decide in advance what you do when the service doesn't respond.

## What pays off, based on the work it takes?

The catalog runs past two thousand projects, so the list doesn't help: what helps is the maintenance load of each category. The author of the dev.to thread installed eleven services on [Proxmox VE](https://www.proxmox.com/en/) with LXC containers and noted the consumption of each one, figures that work as an order of magnitude because they come from his own machine. He also keeps a directory of self-hosted apps, so he has an interest in the list looking attractive.

Replaces

Alternative

Observed RAM

License

Google Photos

Immich

~4 GB

AGPL-3.0

Plex

Jellyfin

~2 GB

GPL-2.0

Bitwarden Cloud

Vaultwarden

~128 MB

AGPL-3.0

Todoist and Trello

Vikunja

~512 MB

AGPL-3.0

Feedly

FreshRSS

~256 MB

AGPL-3.0

Private GitHub

Gitea

~512 MB

MIT

Two lessons come out of that table. Almost everything is small: Vaultwarden drops below 100 MB at rest and the rest fits in 512 MB, so an expensive computer isn't the bottleneck. And the weight isn't in the application but in what it drags along: Immich needs PostgreSQL and a machine learning container, and Jellyfin burns resources transcoding video, not serving it.

Passwords, notes, RSS and your own repositories almost always go well: modest data and a one-day outage breaks nothing. The photo library is in between: big benefit and also a big cost if you get it wrong. Email and identity sit at the opposite extreme. We've already discussed some of this in our article [5 free software alternatives to leave Big Tech behind.](https://observatorioblockchain.com/en/technology/5-alternativas-software-libre-dejar-atras-big-tech/)

## How much does it cost to keep it running?

Hardware stopped being the barrier. The end of Windows 10 support in October 2025 flooded the second-hand market with desktops with 16 GB for 100-160 dollars, and new mini PCs with an N100 go for around 150. For light services, anything is more than enough and draws little.

Scenario

Hardware

Monthly cost

Light apps on a machine you already have

Raspberry Pi or old laptop

0 to 5 dollars in electricity

5 to 15 mixed services

N100 mini PC with 16 GB or a mid-range VPS

5 to 15 dollars

Savings figures should be read with care: selfhosting.sh calculates between 7,700 and 9,200 dollars a year replacing all subscriptions, and between 1,380 and 1,880 for the ten usual services. It has its own directory: these are a ceiling, not a promise. And the cost no table captures is your time.

## How much maintenance does it demand?

Maintenance has four fronts and only one is technical. The first is updates. selfhost.directory counted 150 stable releases between October 8 and 15, 2026, across 2,668 live projects. Ignoring them isn't comfortable with the service exposed to the internet.

The second front is backups, and that's where most people fail: a password manager without a tested backup is a risk, not a feature. Rebuilding the service from scratch should be a written, rehearsed procedure, and that's where it's decided whether the project lives three years or three months.

The third front is exposure: every service open to the internet widens the attack surface. The prudent rule is to start on the local network or a VPN and open only what's necessary to the outside.

The fourth front is the abandonment of the project you rely on: 17% of the census hadn't been updated in a year and 171 were archived. And the sustainability paradox: in the 2024 selfh.st survey, with 3,700 responses, 60% hadn't donated anything in a year. Choosing software with a team behind it is part of the job.

## What isn't worth self-hosting?

Email tops the list, and not because of installation but because of deliverability: getting your messages into the inbox requires a domain reputation that takes months. Identity and single sign-on come right after: with ten or twenty services, Authentik or Authelia show up, good but still specialist territory. If that piece fails you don't lose one service, you lose access to all of them.

The third temptation is migrating everything the first weekend. The pattern that works is the opposite: one service you already pay for, a month checking that the backup restores, and only then the next one. And what [already works for free, with data you don't care about and no cost to grow](https://observatorioblockchain.com/en/technology/5-programas-software-libre-simplifican-vida-digital/), stays where it is.

## The movement behind it, without exaggerating

Without exaggerating: Flexera's 21% is about companies, not living rooms. What did change in 2026 is maturity: [Nextcloud launched its Academy on October 1](https://nextcloud.com/blog/nextcloud-academy-free-online-courses-for-contributors/), still in beta, and the selfh.st survey includes local AI for the first time.

The route takes up less than the shopping list: pick a service that replaces something you pay for and would hurt to lose; deploy it with Docker Compose on the cheapest thing you have; put a proxy with a certificate in front; schedule an external backup and test restoring before trusting it with data; wait a month before the second one.

That order has a logic: starting with the backup rather than the application separates a home server from a summer experiment. The hard part isn't installing, but recovering when the disk fails on a Sunday night.

## A server of your own is a decision about your time

Anyone who enjoys administering systems will find a project that pays for itself in learning. Anyone who just wants to stop paying fees will do better to pick two services, set them up well and not expand the list.

The 2026 data points in a direction that's short on epic but solid: better software than five years ago, cheap second-hand hardware and standardized tools. What hasn't changed is your part: backups, updates and not opening more fronts than you can handle.
