The hidden cost of building
April 10, 2026
Everyone quotes the engineer salary. Nobody quotes the rest. The real cost of shipping a modern software product is the sum of a dozen small bills, most of which you do not see until they have already compounded.
This is the breakdown I wish I had before I shipped three products in twelve months.
The API bill, and how it creeps
You start with $100 a month. A year later you are at $4,000 and nobody can tell you why. Here is the actual shape of how that happens.
| Stage | Monthly spend | What drove it | What you assumed |
|---|---|---|---|
| Prototype | $40–$120 | You testing | "Should stay under $200" |
| First users | $300–$800 | 50 real users | "This scales linearly" |
| Growth | $1,200–$3,500 | Retry loops, streaming, agents | "We will optimize later" |
| Scale | $4,000–$15,000 | Long-context chains, vector costs, background jobs | "We need finops" |
The creep is not usage. The creep is **context length**, **agentic retries**, and **forgotten cron jobs** that re-embed your whole database every night because someone wrote "just in case."
The three silent drivers
2. **Agent recursion.** One user action fans out into 5–12 model calls. A 2x cost-per-call savings becomes an 8x total-cost savings.
3. **Unthrottled retries.** A model API returns a 500, your library retries 3x, your app retries 2x, your client retries 2x. One user click now charges you 12 times.
The codebase gravity problem
Three products sounds efficient. Until you are in month seven and every bug fix is a three-codebase migration.
Three products, one founder — what actually happens
**Process Steps:**
Build product A → Build product B (copy-paste A's auth) → Build product C (diverges) → Fix a bug in auth → Port to B → Miss C → C breaks in production
**Time Investment:**
fine → fine → ship → 2 hours → 3 hours → forgot → outage Sunday morning
**Total Duration:** you are now a maintenance engineer with three jobs
**Key Challenge:** shared code gravity is the thing you pay for the *second* you have more than one codebase.
The fork in the road
| Approach | Upside | Cost |
|---|---|---|
| Monorepo | Shared code, one migration | Heavier tooling, every product ships when the slowest ships |
| Separate repos | Ship independently | Every shared concern forks eventually |
| Shared npm package | Clean boundaries | You are now a package maintainer on top of a founder |
There is no correct answer. There is a choice, and you pay either way. The mistake is thinking there is a free path.
The Apple tax (and the Google tax, and the Stripe tax, and...)
Nobody tells you this upfront. Every consumer app is a stack of taxes.
| Line item | Annual cost | What it buys |
|---|---|---|
| Apple Developer | $99 | Access to the App Store (required) |
| Apple 15–30% cut | 15% (Small Business Program) / 30% | Billing, distribution, review |
| Google Play | $25 one-time + 15–30% | Same as Apple on Android |
| Stripe | 2.9% + $0.30 | Card processing (web) |
| Domain + SSL + DNS | $30–$200 | Basic existence on the internet |
| Postmark / Resend | $15–$100 | Transactional email that actually lands |
| Push notifications | $0–$99 | If you use a service layer |
| Supabase / Auth0 | $0–$599 | Auth + database + storage |
| Object storage | $50–$800 | Images, videos, files |
| Observability (Sentry, etc.) | $26–$479 | Knowing your app is broken before users tell you |
| AI APIs | $200–$15,000 | See previous section |
A small, shipping iOS + web product runs **$600–$2,000/month** in pure infrastructure before anyone is paid. Scale brings that to $5,000–$30,000.
Apple specifically
The review process is the one founders underestimate the most.
| Submission | Median time | Median rejection rate |
|---|---|---|
| First submission (new app) | 24–72 hours | ~40% |
| Update (established app) | 4–36 hours | ~10–15% |
| TestFlight (beta) | 24 hours | ~5% |
Forty percent. Your first submission has a near-coin-flip chance of coming back with a rejection. Reasons from my own ledger: a screenshot did not match the app exactly, a purchase flow used web checkout instead of IAP, a privacy string was too vague, a tracking framework was flagged. Each rejection is a 2–5 day round trip. Plan 2–3 submissions for any first release.
Verification, not interviews
The hidden cost nobody writes about: the hour you spend on every partnership, every integration, every enterprise conversation, proving you are a real company.
| Verification | Time | Frequency |
|---|---|---|
| Apple Developer identity | 1 hour + 2 days wait | Annual |
| Stripe KYC | 30 min + 1–5 days | Per account |
| Google OAuth scopes (restricted) | 4–12 hours of prep | Per scope, one-time-ish |
| Business verification (banks, payroll) | 2–4 hours | Per vendor |
| Partnership due diligence | 2–6 hours | Per partner |
| Enterprise security reviews | 10–40 hours | Per deal |
A solo founder loses **10–30 hours a month** to verification work that builds nothing and ships nothing. It is the single most under-discussed expense in early-stage building.
Infrastructure layers silently
You start with a dev server. A year later:
**Process Steps:**
Dev server → Vercel → Vercel + Supabase → + CDN → + queue → + cron worker → + separate staging → + observability stack → + secrets manager → + CI/CD → + domain management
**Time Investment:**
day 1 → week 1 → month 1 → month 3 → month 5 → month 7 → month 8 → month 9 → month 10 → month 11 → month 12
**Total Duration:** you now administer ~11 pieces of infrastructure
**Key Challenge:** each one was added for a real reason. None of them are removable. Collectively they are ~$400–$1,500/month and consume a day a week to maintain.
What this adds up to
For a solo or two-person consumer AI product in 2026, here is the actual monthly cost of "shipping" before founder salary, before marketing, before a single feature:
| Category | Low end | Realistic | High end |
|---|---|---|---|
| AI APIs | $200 | $1,200 | $8,000+ |
| Infrastructure | $150 | $500 | $1,800 |
| Platform fees (Apple/Google/Stripe) | $100 | $400 | $2,500+ |
| Tooling (email, auth, storage, observability) | $100 | $350 | $900 |
| Verification / compliance overhead | 10 hrs | 20 hrs | 40 hrs |
| **Monthly total** | **~$550 + 10 hrs** | **~$2,450 + 20 hrs** | **~$13,200 + 40 hrs** |
The punchline: a solo shipper with three products live is running a **~$30,000/year infrastructure-and-platform-tax bill** before their first engineer is paid, plus a part-time job worth of invisible verification and maintenance work.
The point
None of this is complaint. All of it is why consumer AI products are harder than the "one GPU and a weekend" narrative suggests. If you are thinking about building — or funding someone who is — these are the line items to ask about. The product quality is usually not the bottleneck. The maintenance cost of having shipped is.
