I have been self-hosting services for eight years. In that time the argument for it has changed considerably. The early-2010s case was mostly ideological: privacy from surveillance capitalism, distrust of centralised platforms, a general preference for understanding the systems you depend on. Those arguments still apply, but in 2024 there are several more pragmatic ones worth making.
The cost argument has become real
Cloud storage prices have increased substantially since 2021. Google One, iCloud, and Dropbox have all raised prices, sometimes multiple times. At the 2 TB tier you are now looking at roughly £100–£130 per year across most providers, per account. A household with two adults and shared storage needs is spending £200–260 annually for cloud storage alone.
A 20 TB NAS with two consumer hard drives in a mirrored configuration costs around £400–500 to build and will likely run for five to seven years. The five-year total cost, including electricity and a spare drive, is around £600–700. For comparison, five years of cloud storage at current prices for the same household is over £1,000. The numbers now clearly favour self-hosting at any meaningful data volume.
The full analysis depends on your storage volume — see our NAS vs. cloud cost comparison for the numbers at different scales.
The privacy argument is better supported
The practice of mining user data for advertising purposes is now well documented rather than speculative. More practically, there have been several high-profile cases of cloud providers scanning user content and taking action on it — sometimes correctly, sometimes incorrectly, almost always without meaningful recourse for the user. Self-hosted services do not report to anyone.
This matters most for data that is sensitive in aggregate even when individual items seem innocuous: location history, health data, communication patterns. A self-hosted Nextcloud does not phone home. A self-hosted Immich photo library does not train a model on your family photos. A self-hosted Vaultwarden does not have a server-side copy of your password vault.
Reliability is more nuanced than people assume
The usual objection is that cloud services have better uptime than a machine you run yourself. This is true in aggregate, but it is the wrong comparison for most use cases. Cloud services have outages too — major ones, affecting millions of users simultaneously. When iCloud is down, every Apple device in the household stops syncing. When Google Drive has an incident, work stops for everyone who depends on it.
A self-hosted service, when it fails, fails for you alone. You can fix it. You understand what happened. The blast radius is contained. For services that are genuinely critical, you can add redundancy: a second host, synchronised data, a failover. That option is not available when your data lives on someone else’s servers.
The real reliability comparison is between “a well-maintained homelab” and “a commercial service”. A neglected homelab with no monitoring, no backups, and aging hardware will absolutely have worse availability than a commercial service. A maintained one, with proper backups tested regularly and basic monitoring in place, compares favourably for most personal use cases.
The learning value compounds
Running your own services is an education that is hard to replicate any other way. When something breaks, you debug it. You read man pages, understand kernel messages, learn how networking actually works. Over years this compounds into genuine expertise that is directly applicable to professional work in systems engineering, security, and infrastructure.
This is not a justification for doing something that does not otherwise make sense — but it is a real benefit that tends to get left out of cost comparisons.
When self-hosting does not make sense
Honest self-hosting advocacy requires acknowledging the limits:
- If you cannot or will not maintain it. Software needs updates. A self-hosted service with a two-year-old CVE in its web interface is a liability, not an asset.
- For email. Running your own mail server is technically possible but practically difficult: deliverability depends on IP reputation, and the operational burden of dealing with spam filtering and blocklists is significant. Most people are better served by a privacy-focused mail provider than by a self-hosted MTA.
- When availability genuinely cannot slip. If a service being down for four hours while you troubleshoot a drive failure would have serious consequences, think carefully about redundancy costs.
The practical starting point
If you are considering starting: begin with something low-stakes but immediately useful. A local DNS resolver with Pi-hole requires minimal maintenance and has a tangible, visible effect on your network. A WireGuard VPN lets you reach your home network securely from anywhere. Neither of these requires always-on hardware — a Raspberry Pi handles both comfortably.
The homelab grows from there, at whatever pace your interest and time allow.