the quick math before the story: two senior engineers building DIY tenant provisioning for 10 weeks is roughly $144,000 in salary. ongoing maintenance adds another $37,000-$75,000 a year. thats the bill most teams never actually calculate, and its why DIY infra quietly bleeds money.
ok real talk for a second.
i was on a call last week with a founder who spent 4 months building custom tenant provisioning. four. months. his team of 3 engineers doing nothing but infra.
you know what his customers got in those 4 months? nothing. zero new features. the product literally stood still while they were knee deep in kubernetes yaml and terraform configs.
and heres the kicker. he told me "we thought itd take 3 weeks."
how much does DIY tenant provisioning actually cost?
lets do some quick napkin math cuz i think most people dont actually sit down and calculate this. (the full breakdown of whats inside that build is its own post.)
average senior engineer salary: lets say $150k/year. thats roughly $72/hour.
building tenant provisioning from scratch: 8-12 weeks minimum (and thats being generous).
so two engineers for 10 weeks = $144,000 in salary alone. not counting the opportunity cost of what they couldve built instead.
and that number? thats just to get version 1 working. not good. working.
then you got the maintenance. every week, 10-20 hours of keeping the lights on. ssl certs expiring at 2am. containers dying. volumes filling up. dns breaking cuz someone changed a config.
thats another $37,000 - $75,000 per year just in maintenance labor.
meanwhile ShipClaw costs less than one engineers monthly salary. for the whole platform. with isolation baked in.
the math doesnt work for DIY. it just doesnt.
what happens when you stop building infra and start shipping
heres what i keep coming back to. the teams that stopped building their own infrastructure are the ones posting revenue numbers.
Savio hit $19,613 MRR with 437 active subscriptions. He didnt get there by hand-rolling container orchestration. He got there by shipping features every week while the platform handled the boring stuff.
Marc Lou went from $0 to $4,779 MRR in 6 days. 95 active subscriptions. Six days. While other teams were still debugging their provisioning scripts, he was already collecting revenue.
these arent hypotheticals. these are real dashboards from real people who made a simple decision: stop spending engineering time on infrastructure that doesnt differentiate your product.
"but we need control"
i hear this one a lot. "we need full control over our infrastructure."
ok but like... do you tho?
most teams saying this have 12 customers. twelve. they dont need kubernetes. they need to ship features and close deals.
you know what customers actually care about? whether your AI works. whether its fast. whether their data is safe. nobody ever bought a SaaS because the infra was artisanal.
the teams that are winning right now are the ones that outsourced the boring infra stuff months ago. theyre shipping features every week while their competitors debug nginx configs.
what i keep seeing over and over
talked to maybe 40-50 teams in the last few months. the pattern is always the same:
- team decides to build multi-tenant AI product
- team starts with shared runtime (fast and easy)
- first enterprise prospect asks about isolation
- team panics and starts building custom provisioning
- 3 months disappear
- team finally gets something working but its fragile
- team spends forever maintaining it instead of shipping
its like clockwork. every single time. ive written up exactly what breaks and when if you want the gory version.
the teams that skip steps 4-7? theyre the ones closing enterprise deals while everyone else is still wrestling with devops.
the stuff that breaks at 3am
real things ive seen go wrong with DIY provisioning:
- ssl cert expired on a saturday night. 200 tenants went down. took 6 hours to fix because the engineer who set it up was on vacation
- one tenant filled their volume to 100%. monitoring didnt catch it. customer lost 3 days of data
- dns propagation issue after a migration. half the tenants couldnt reach their runtime for 4 hours
- container memory leak. slow performance for weeks before anyone noticed. three customers churned
- env variable typo in production. wrong API key injected into wrong tenant. data went to wrong place
each one of these is a real story from a real team. each one is the kind of thing that doesnt happen when the platform handles it for you.
the compounding problem
heres what gets me. its not just the direct cost.
every month you spend on infra is a month you dont spend on:
- features that would close that enterprise deal
- onboarding improvements that would reduce churn
- integrations that would open new markets
- the actual AI stuff that your customers pay you for
and it compounds. the gap between teams that outsourced infra and teams that didnt gets wider every single month.
6 months from now, the team using ShipClaw has shipped 20 features. the team doing DIY has shipped 8 and is still patching their provisioning system.
thats not a small difference. thats a completely different trajectory.
"ill just do it next quarter"
biggest lie in startups. next quarter never comes. theres always a reason to delay.
and every quarter you delay is another quarter of:
- engineering time burned on infra
- enterprise deals lost because you cant demonstrate isolation
- incidents that erode customer trust
- features that dont get built
the teams switching to ShipClaw arent doing it because they want to. theyre doing it because they ran the numbers and realized the DIY approach was killing their growth.
look im not saying its for everyone
if you got a dedicated platform team with nothing else to do, build it yourself. if infrastructure IS your product, build it yourself. if youre at a scale where platform costs are genuinely higher than build costs, build it yourself.
but if youre a team of 2-10 engineers trying to ship an AI product? spending months on tenant provisioning is probably the worst thing you could do with your time.
the visual builder takes about 15 minutes to set up your first tenant. fifteen minutes vs four months. i dont know how to make the math any clearer than that.
go try it. worst case you lose 15 minutes. best case you get 3 months of your life back. If you want to see the exact deploy flow before you try it, the 5-step deploy wizard is documented here.