Best infra setup for solo SaaS in 2025?

I've been on Railway for 18 months. Here's my honest take:

Good: - DX is genuinely fantastic. Zero friction deploys - The dashboard is beautiful and actually useful - Support has been responsive every time I needed them

Bad: - Cold starts are real. First request after idle can take 4-6 seconds. Ended up using a uptime monitor hitting my API every 5 minutes to keep things warm. - Pricing can surprise you. The "pay for what you use" sounds great until you get a $300 bill because you had a runaway cron job for a weekend.

For a solo with predictable traffic: Railway is great. For anything with bursty traffic: consider Fly.

Fly.io gang here. Switched from Render 8 months ago and haven't looked back.

The cold start issue Tobias mentioned is real on Railway. On Fly, I've never had it — their autostart/autoscale handles it properly.

The config IS more complex, but here's the thing: you write it once. My fly.toml is 30 lines and hasn't changed since I wrote it. The complexity is front-loaded.

For your use case (Next.js + Node + Postgres): Fly is overkill unless you need edge. If you're US-focused with occasional EU traffic, Render is perfectly fine and you'll save yourself a lot of YAML debugging.

Unpopular opinion: just use Vercel for the Next.js frontend and Railway for the API.

Don't overthink the "one platform" aesthetic. Each tool has strengths. Vercel's edge network for Next.js is genuinely hard to beat, and Railway's Postgres offering is solid.

Separating them also means you can scale each independently. If your API gets hammered but not your frontend (or vice versa), you only pay for what needs more resources.

My current setup: - Vercel: Next.js frontend (free tier is generous) - Railway: Node API + Postgres ($20-40/mo) - Turso for edge SQLite (experimental, mostly for fun)

Total: ~$30-50/mo for a real production app with real users.

Quick note on the cold start thing — this is largely a Postgres problem, not a hosting problem.

If you pick Railway and you want to avoid cold starts: 1. Use a hobby Postgres instance (it never sleeps) 2. Add a simple cron job on your Railway plan that pings your API every 5 min 3. Or upgrade to their performance instances which don't have the same idle behavior

I was on Railway for a year with zero cold start issues because I set it up properly from day 1. The horror stories are usually from people who didn't know about the idle timeout behavior.

Hey! DevRel at Railway here — not here to spam, but since Railway keeps getting mentioned:

The cold start issue is real and we've been working on it. Our new performance instances ($20+/mo) don't have idle timeout behavior, so cold starts aren't an issue.

For the pricing concern: we have a calculator and transparent pricing page. Most small apps run $5-15/mo. The horror stories are usually from scaling events or misconfigured services.

Happy to answer any specific questions about Railway setup.