Managed vs self-hosted: the honest decision tree (it isn't 'self-host everything')

July 15, 2026

Read enough “should I use managed services or self-host?” threads and you notice they all converge on the same, quietly wise answer. Not “self-host everything.” Not “just pay for managed.” It’s some version of: “I host everything myself — except email and DNS.” Or, more honestly still: “I like owning my stack, but I like sleeping more.”

That carve-out is the entire framework. Let me make it explicit.

The three axes people actually decide on

Strip away the ideology and every real decision comes down to three trade-offs (all three showed up, unprompted, in the same thread):

  1. Mental load vs control. Managed wins when you want reliability with minimal friction and don’t want to babysit updates, patching, and uptime. Self-host wins when the thing is core to your workflow and you need full flexibility — and only if you’ll commit the time to maintain it.
  2. Cost at scale. Managed is cheap when you’re small or the usage is modest. Past a threshold — and it’s usually a traffic threshold — the per-usage meter overtakes the cost of just running the box yourself.
  3. Data sovereignty (and learning). Sometimes the deciding factor isn’t money at all; it’s where the data is allowed to live, or wanting to actually understand your own stack.

The heuristic: move the meter, keep the chore

Here’s how I turn those axes into an actual answer for clients:

Move what scales with your success; rent what's cheap and painful to run.

Put simply: leave the meter, keep the chore. Move the costs that punish growth; pay someone to run the boring, hard, cheap stuff.

This is the “when not to move” position, in tree form

If someone tells you to move all of it, or none of it, they’re selling something — a migration, or a managed-everything bill. The honest answer was never ideological. Self-hosting isn’t a virtue; it’s a tool for killing the per-usage meter where it hurts, while keeping the convenience where the meter is cheap. Flat-rate managed services — managed Postgres, zero-egress object storage — are frequently the right call: you’re leaving the meter, not the datacenter, and not the reliability.

The output of the tree is never “self-host everything.” It’s a short list: move these three (they scale with traffic), keep those two managed (they’re cheap and a pain), and here’s the five-year math for each. That list is exactly what a bill teardown produces.


Not sure which parts of your stack are worth moving and which are cheaper left alone? That’s the whole question I answer — and sometimes the answer is “leave it.” Every message comes straight to me — I read and reply to each one myself, usually within a day, and what readers send shapes what I build next. It’s just me for now, so that’s genuinely true; it won’t be forever. Send me your setup or a recent bill and I’ll give you the move-these / keep-those split — free, within a business day.

Don't miss new posts

I publish honest, sourced breakdowns of cloud-exit economics — egress, storage, monitoring, reliability — and the occasional announcement. Leave your email and I'll let you know when something new goes up.

Double opt-in — you'll get one email to confirm. No spam, unsubscribe anytime. Read by me, never shared.