I kept seeing "just self-host it, it's basically free" in threads, and I wanted to know what "basically free" adds up to when you write the real numbers down.
So I did the boring version of the exercise. I took 11 open-source apps people actually run, wrote down what each one costs to host, then opened the pricing page of the commercial tool it replaces and wrote that down too. Every SaaS number below was read off the vendor's own pricing page in August 2026. No estimates, no "starting from" hand-waving where I could avoid it.
Full disclosure before the numbers: I work on InstaPods, which is one of the places you can host these. I have used our plan prices in the self-hosted column because they are the ones I can quote exactly. You can substitute a $5 VPS and do the ops yourself, and I make that case honestly further down, because for a lot of people it is the right answer.
What does self-hosting 11 apps cost per year?
Running all 11 on managed pods came to $107/mo, or $1,284/yr. The commercial equivalents at their entry paid tiers came to $385/mo, or $4,620/yr, plus a one-time $1,500 onboarding fee on the CRM. The gap is real, but roughly a third of it comes from two line items.
Here is the whole table.
| App | Replaces | Self-hosted | Per year | SaaS entry paid tier | Per year |
|---|---|---|---|---|---|
| Beszel | Datadog | $3/mo | $36 | $15/host/mo | $180 |
| Uptime Kuma | Pingdom | $3/mo | $36 | $15/mo, 10 monitors | $180 |
| Fider | Canny | $3/mo | $36 | $79/mo billed yearly | $948 |
| n8n | Zapier | $7/mo | $84 | $20/mo, 750 tasks | $240 |
| Umami | Plausible | $7/mo | $84 | $9/mo, 10k pageviews | $108 |
| Instatic | Webflow | $7/mo | $84 | $25/mo annual for a CMS | $300 |
| NexoPOS | Loyverse | $7/mo | $84 | $34/mo for one store | $408 |
| Postiz | Buffer | $15/mo | $180 | $5/channel/mo, 6 channels | $360 |
| LiteLLM | Portkey | $15/mo | $180 | $49/mo, 100k logs | $588 |
| AutoMem | Mem0 | $15/mo | $180 | $19/mo Starter | $228 |
| CRM | HubSpot | $25/mo | $300 | $90/seat/mo, 1 seat | $1,080 |
| Total | $107/mo | $1,284 | $385/mo | $4,620 |
Two of those rows deserve a caveat rather than a victory lap, and I would rather flag them than let the total do the arguing.
Canny at $948/yr is the entry price past its free tier, not a typical bill. Canny is free up to 25 tracked users, and a tracked user is anyone who posts, votes or comments. Plenty of small products never cross that line. If you never cross it, Fider saves you nothing.
HubSpot's $1,080 assumes one seat. Sales Hub Professional is $90 per seat per month and carries a one-time $1,500 onboarding fee, so year one for a single seat is $2,580 and a five-person team is $6,900. That row is doing a lot of work in the total, and it is the row where self-hosting looks most dramatic precisely because per-seat pricing is what it is.
What are you actually renting when you self-host?
A VPS is a slice of a physical server: some CPU, some RAM, some disk, a public IP, and root on a Linux install nobody else shares with you. That last part is the whole reason self-hosting works. You can install anything, open any port, and run a long-lived process. Shared hosting cannot do that, which is why the "$2/mo hosting" you see advertised will not run n8n.
If that distinction is new to you, what a VPS actually is and how a VPS differs from shared hosting are the two things worth reading before you pick a price.
How much RAM does each of these actually need?
Three tiers, and the jump between them is what drives the price. This is where the "one $5 box runs everything" argument either works or falls apart, so it is worth being concrete. Grouping the same apps by the smallest amount of memory I would put them on:
| Comfortable RAM | Apps |
|---|---|
| 512 MB | Beszel, Uptime Kuma, Fider, Memos, Excalidraw, Vaultwarden |
| 2 GB | n8n, Umami, Stirling PDF, Instatic, NexoPOS |
| 4 GB | Postiz, LiteLLM, AutoMem, AFFiNE |
Beszel is the outlier at the bottom: its agent runs in under 10 MB, which is why it sits on the cheapest tier with room to spare. The whole 512 MB group is genuinely small, and you could run all six on one cheap box and still have memory left. The moment n8n, Postiz or AFFiNE join the party you are into 4 GB and up, and the arithmetic changes.
When does self-hosting stop being cheaper?
When your time costs more than the difference. This is the part the "just self-host it" replies leave out, and it is not a small part.
A $6/mo VPS running six of these apps is $72/yr against maybe $2,000/yr of SaaS. That is a genuinely huge saving, and it is also a machine that you now own the operations for: TLS renewal, OS patches, app upgrades, database backups you have actually tested restoring, and being the person who gets paged when the disk fills at 2am. None of that is hard. All of it is recurring.
Price your own hour and multiply. Two hours a month of maintenance at $50/hr is $1,200/yr, which eats most of the gap on the small apps and none of the gap on the expensive ones. That asymmetry is the actual finding here:
- Replacing a cheap SaaS with a self-hosted app rarely pays for itself in time. Umami against Plausible saves $24/yr. If setting it up and keeping it patched costs you an afternoon a year, you lost.
- Replacing per-seat or per-meter SaaS almost always pays for itself. Canny, Portkey, Buffer and HubSpot all charge on an axis that grows with your success. Self-hosted apps charge on an axis that grows with your RAM, and RAM is cheap.
That is also the real choice behind managed versus self-managed hosting: you are not choosing between paying and not paying, you are choosing which of the two things you would rather spend, and how predictable you want the bill to be. If you want the year-level view rather than the monthly one, what hosting costs per year lays that out.
One box, or one box per app?
Both work, and they fail differently.
One box, many apps is cheapest and is what most self-hosters do. You get one thing to patch and one bill. You also get one blast radius: a runaway n8n workflow can starve everything else on the machine, and a bad upgrade takes all of it down at once. Nothing is isolated unless you make it isolated.
One pod per app costs more, because you are paying per app rather than per machine. My table above is priced this way, and it is the honest reason the self-hosted column is $107/mo instead of $6/mo. What you buy for that is that each app has its own RAM ceiling, its own restart, and its own URL, and one of them dying does not touch the others.
If you are running four small apps for yourself, one box is obviously right. If you are running something a customer depends on next to something you are experimenting with, splitting them stops being a luxury. Either way, picking the VPS itself is the same decision.
The short version
Self-hosting is not free, and the people who say it is are usually not counting their own hours. What it is, is predictable. A pod that costs $7/mo costs $7/mo whether you send 100 webhooks or 100,000, whether you connect 3 social channels or 30, and whether your feedback board has 20 users or 2,000. That is the whole trade. You give up a support contract and a polished onboarding, and you get a bill that does not have a growth clause in it.
If I had to compress 11 rows into one rule: self-host the things that charge you per seat, per channel, per operation or per tracked user. Pay for the things that charge a flat $9.
What is the one SaaS bill you looked at and thought "I could run this myself for a tenth of that"? And did you actually do it, or is it still on the someday list? I am genuinely curious how many of these get set up and then quietly abandoned.
Top comments (0)