Vibe coding means describing what you want and letting an AI write the code. The question we hear next is always the same: what does it cost, and what do I need before real customers show up? We worked it out for a concrete example, an accounting platform, in two phases. All prices are list prices we found in October 2026, in the currency the vendor shows. Confirm each one before you pay, because they change often. Anything we could not price is marked as such.

Disclosure: CloudAcropolis sells a managed AI chat agent, a managed social scheduler and cloud servers, and those prices come from our own pages. Every other price comes from the vendor or a third-party source listed at the end.

Phase 0: the concept (about four months)

The goal is to prove the idea works with test data. Nobody depends on it yet.

What you need

  • A laptop running an AI coding agent, with a top-tier subscription from a major AI service provider at $100 a month. We assume the $20 tiers are too small for daily building. That is a practitioner's estimate, not a measured limit.
  • A second major AI service at $20 a month, used as a reviewer of the code the first model wrote.
  • A free code hosting plan for the code and automated tests. Private repositories are unlimited, with 2,000 automation minutes a month.
  • A paid plan from a database-as-a-service (DBaaS) provider at $25 a month for the database and user login, with daily backups kept for 7 days. The Free plan is fine to play with, but it has a 500 MB database, no automatic backups, and pauses after a week of inactivity.
  • One Linux server for the app and the workflow automation tool. A 4 vCPU / 8 GB server costs about €16-21 a month at budget providers. CloudAcropolis lists 4 vCPU / 8 GB with dedicated CPU at €29 a month.
  • LLM usage for the app itself: we budget about €30 a month, roughly €1 a day. That is an estimate, not a quote.
  • A domain at about €14 a year and hosting plus email at about €20 a year.

Free tools cover error tracking, email (300 emails a day), social scheduling (three channels) and uptime checks.

The numbers

  • Initial: about €34 (domain, hosting and email).
  • Monthly: €191-196.
  • Four months in total: about €800-820.

If you also send outbound email, add about $50 a month for verified leads, which brings phase 0 to about €1,000. That figure is derived from a September 2026 price index, not from a vendor page. Whether you may email those people depends on each person's country, so check the rules before you send.

Where a managed service saves you the skills

Some tools are free but expect a Linux administrator. Self-installing an open-source AI chat agent needs command-line skill, and in our experience its web chat does not work out of the box. A popular dashboard tool needs a separate metrics collector and configuration files before it shows anything. A simple open-source uptime monitor is easier: it checks URLs, ports, ping, DNS records and containers. Its documentation does not mention watching disk or memory, and we found no no-code tool that does.

If you do not want to run these yourself, managed options exist. Our managed AI chat agent is €49 a month for up to four agents, and our managed social scheduler is €29 a month. The managed plans list multi-factor login, EU hosting and daily backups. Setup is quoted separately, and the pages do not say whether VAT is included.

Phase 1: proper infrastructure (about six months)

One server is a single point of failure. In our experience shared-CPU plans also struggle under real load, so production needs dedicated CPU. Phase 1 separates the pieces and adds security.

The architecture

  • Edge: A web application firewall and DDoS shield in front, free with 5 rules or $25 a month with the full industry-standard ruleset, and a load balancer from $6.49 a month.
  • Two application servers behind the load balancer, each with dedicated CPU. One example at a budget provider is 2 vCPU / 8 GB at €12.49 a month. Every CloudAcropolis plan includes dedicated vCPU. We have not sized either for real load.
  • A separate workflow automation server. Production guidance is 4 or more cores and 8 GB of RAM.
  • A managed database: The paid database plan plus point-in-time recovery at $100 a month, for $125 in total.
  • Monitoring on its own server, so it can still alert when the first one is down.
  • Antivirus on every server. An open-source scanner is free but needs 3 GiB of RAM at minimum and 4 GiB preferred. Paid options start around $78 a year. Plan one extra core and 4 GB of RAM per server.
  • Off-site backups and a payment gateway of your choice.

The numbers

  • Pen test, one time: $4,999-$8,999 for a small web app and $4,999-$10,000 for a small API. Repeat it yearly.
  • Monthly: about €444-469 for the priced items (the high end adds the paid firewall plan).
  • Six months in total: about €7,700-21,800 including the pen test.

What the budget leaves out

  • VAT.
  • Payment fees. At one provider, an EU card costs 1.5% plus €0.25 and SEPA direct debit €0.35. Chargebacks cost €15-20 each.
  • Servers we could not price: the workflow automation server, monitoring and off-site storage, plus the extra core and RAM for antivirus.
  • Government filings and connectors, which differ in every country.
  • A human code review.

Four takeaways

  1. In phase 0 the AI subscriptions cost more than the servers.
  2. Free tiers are for playing. Build on a paid database plan.
  3. In phase 1 the pen test is the largest single cost.
  4. Do not run your monitoring on the server it watches.

Sources

Prices: Claude and ChatGPT plans (morphllm.com), Supabase (supabase.com/pricing), Hetzner plans and load balancer (makerstack.co), Cloudflare plans (eastondev.com), pen tests (blazeinfosec.com), ClamAV requirements (docs.clamav.net), Linux antivirus (security.org), payment fees (kolonell.com), verified lead prices (cleanlist.ai), managed OpenClaw and Postiz (cloudacropolis.com). Tool behaviour: Uptime Kuma (github.com/louislam/uptime-kuma), Grafana (grafana.com/docs). Domain and hosting prices are the author's own figures.