I Run 5 AI Apps Solo — Total Cost: $171/Month

I'm a solo developer in South Korea running five apps: a Korean language learning app, a wellness app rooted in East Asian philosophy, an AI-powered legal document reviewer, an AI vulnerability scanner, and a math learning tool for struggling students.

Not all of them are fully launched yet. One is in App Store review, one is in open beta, one is still in development. Two are live in production. This is the reality of being a solo developer — everything is at a different stage, all the time.

But here's what surprises people: the total monthly cost to run and actively develop all five is $171. Not $1,710. Not $17,100. One hundred seventy-one dollars. Let me break down every dollar.

The Stack

I use the same stack across all five apps. Not because it's the best possible choice for each one, but because when you're one person, you cannot afford to context-switch between different databases, hosting providers, and deployment pipelines.

• Mobile: React Native + Expo
• Web: Next.js
• Backend: Supabase (Postgres + Auth + Edge Functions + Storage)
• AI: Claude API (Anthropic)
• Hosting: Vercel (web) + Fly.io (one service)
• Payments: RevenueCat (mobile) + PortOne/TossPayments (Korea)
• DNS/CDN: Cloudflare
• CI/CD: EAS Build + Vercel Git Deploy
• Monitoring: Slack webhooks + pg_cron
• Social Media: Buffer

No Kubernetes. No AWS. No multi-service Docker Compose. Just managed services with free tiers, connected by Edge Functions.

Part 1: Infrastructure — What It Costs to Keep the Apps Running

These are the costs I'd pay even if I stopped developing entirely and just kept the apps alive.

Supabase — $25/month base

I have one organization on the Pro plan at $25/month. Five projects live under it: papurion-master, dalnara, k-saju-global (the Ohaeng backend — long story about the rename), claused, and suumio. Each generates about $1.65/month in Micro Compute costs, but Supabase gives compute credits that currently zero this out.

My actual invoices tell the real story: $48.84 in May, $32.20 in April, $25.00 in March. The base is $25 but compute costs fluctuate. This month's projection is $64.87 as usage grows. For this article, I'll use the current cost of $25 and note that it's trending upward.

Why Pro at all? The free tier pauses your database after 7 days of inactivity. If your app has a slow week — and new apps always have slow weeks — your users get errors. The $25/month is partly paying for "please don't turn off my database."

Claude API — $8.89/month

This is my actual bill from the Anthropic console right now. I use a mix of models: Claude Sonnet 4.6 handles most of the workload, with Claude Opus 4 for complex tasks and Claude Haiku 4.5 for high-volume, low-stakes operations like social media draft generation.

People are always shocked this is under $10. Three reasons:

First, I cache aggressively. If 50 users ask similar questions about the same Korean grammar point, I don't call Claude 50 times. I call it once, store the structured response in Supabase, and serve it from there.

Second, my AI calls are mostly batch operations in Edge Functions — generating content, analyzing documents, triaging user reports — not real-time chat. Batch is cheaper because you can optimize prompt length and control output size.

Third, I learned a painful lesson about CJK tokens. Korean, Japanese, and Chinese text tokenizes at roughly 2-3x the rate of English. Early on I was burning tokens on verbose Korean responses. Now I set max_tokens to 800 minimum (below that, Korean responses get cut off mid-sentence) but explicitly instruct the model to be concise.

Apple Developer Account — $8.25/month

$99/year, amortized. Non-negotiable for iOS apps. Google Play was a one-time $25, already paid.

Everything else — $0/month

This is not an exaggeration. Here's the breakdown:

Vercel: Free Hobby tier. All web frontends and landing pages. 100GB bandwidth, serverless functions, preview deployments on every git push. Haven't hit the limits.

Cloudflare: Free tier across eight domains — claused.co.kr, claused.kr, dalnara.app, k-saju.app, ohaeng.app, papurion.com, pie-nest.com, and writemyday.app (a journaling app I started building and killed when the unit economics didn't work out — the domain lingers as a reminder that not every idea deserves to ship). DNS, CDN, SSL, analytics, and a Workers-based CORS proxy for all of them. Total cost: zero.

Fly.io: Free tier. My vulnerability scanner runs here because it needs long-running processes that don't fit serverless.

Buffer: Free tier for social media automation. My pipeline: pg_cron triggers a Supabase Edge Function → Claude Haiku generates a post → Buffer publishes to Instagram.

RevenueCat: Free up to $2,500 in monthly tracked revenue. My apps are still in early stages, so this costs nothing.

GitHub: Free. All private repos. Unlimited.

Slack: Free. Monitoring webhooks for errors, user reports, daily metrics.

Infrastructure total: ~$42/month

Supabase $25 + Claude API $8.89 + Apple $8.25 = $42.14. That's just the "keep the lights on" number. Realistically $50-65 as Supabase compute costs grow with usage. But this isn't what I actually pay each month — keep reading.

Part 2: Development Tools — What It Costs Me to Build

These are costs that exist because I'm actively developing, not because apps are running.

Claude Max — $110/month

This is my biggest single expense. Claude Max at the 100-tier gives me extended context, priority access, and enough usage for heavy development sessions. I use it for everything: debugging code, architecting features, writing documentation, reviewing my own pull requests.

Is it infrastructure? No — the apps would run fine without it. Is it essential? Absolutely. A solo developer without an AI coding assistant in 2026 is like a carpenter without a power drill. You can work without it, but you're choosing to be slower for no reason.

Expo (EAS) — $19/month

Expo Application Services handles mobile builds for iOS and Android. The paid tier gives me enough build credits for regular development cycles without worrying about running out mid-sprint.

Development tools total: $129/month

The Full Picture: $171/month

That's $171 to run and actively develop five apps. Split per app, that's $34 each.

What This Number Doesn't Tell You

I want to be transparent about what's not in that $171.

My time. I work on these apps full-time. If I valued my hours at $50, the real monthly cost would be in the thousands. Low infrastructure cost only works because I'm trading time for money — handling ops manually, debugging issues myself, doing support directly.

Domain registrations. About $50-80/year across eight domains. Not huge, but not zero.

The apps aren't all generating revenue yet. Two are in production, one is in App Store review, one is in open beta, one is in development. This cost structure works at zero-to-low traffic. If any app gets real traction, Supabase compute costs will climb, Claude API usage will spike, and I'll need to upgrade Vercel to Pro ($20/month). These are good problems to have.

The Tradeoffs

No redundancy. All five apps share infrastructure in Seoul region. One Supabase outage takes everything down. A real company would have failover databases and multi-region deployments. I have Slack alerts and a prayer.

No staging. I deploy to production and test on production. I use Row Level Security and preview deployments, but the database is always prod. This is not best practice. It's solo-developer practice.

Shared resources. Five projects on one Supabase organization means they share connection limits. A traffic spike in one app could theoretically affect the others. Hasn't happened yet because none of them have enough traffic for it to matter.

The Automation That Makes Solo Possible

The low cost isn't the interesting part. The interesting part is how one person runs five apps without burning out.

Social media: Fully automated. pg_cron → Claude Haiku → Buffer → Instagram. I review posts occasionally. I don't write them.

User reports: When a user submits feedback, an Edge Function triages it with Claude Haiku — spam? bug? feature request? urgent? — routes it to the right Slack channel with a priority tag, and auto-responds. I intervene only when needed.

Database maintenance: pg_cron runs nightly: session cleanup, daily metric snapshots, materialized view refreshes. Things that would otherwise require me to remember to run scripts at midnight.

What I can't automate: App Store review. Apple and Google's submission process is manual and unpredictable. I spend more time on App Store rejections than on almost anything else.

Should You Do This?

If you're a developer with an idea, the barrier is genuinely near zero. Supabase free, Vercel free, Expo free, Cloudflare free. You can go from idea to App Store submission without spending a dollar on infrastructure.

But I want to push back on the "it's so cheap!" narrative, because I'm part of the problem by writing this article. The infrastructure is cheap. Building something people want is not cheap — it costs months of your life. Getting them to find it costs energy. Convincing them to pay costs more energy still.

$171/month for five apps, including my AI coding assistant and build pipeline. I can't imagine doing this any cheaper without cutting something that makes me slower.

If you're running multiple production apps for less than this, I genuinely want to hear from you. Not to compete — to learn. Hit me up and show me your bill. I'll buy you a coffee. Virtually, of course. I'm trying to keep costs down.

🧪 Beta Testers Wanted (20 spots)

I'm looking for 20 Android beta testers for two of my apps:

• Ohaeng — Five Elements wellness app based on East Asian philosophy
• Dalnara — Korean language learning through K-Drama, roleplay, and culture

Tap the link, opt in with your Google account, and you'll get access via Google Play. Testing runs 14+ days. Feedback welcome in the comments or on GitHub.

I'm Donghyun, founder of Pie Nest Inc. I build AI-powered apps for language learning (dalnara.app), wellness (ohaeng.app), legal document review (claused.kr), and security scanning (papurion.com).