Managed Cloud

The "I will pay money to not run servers" path. The right answer for a lot of teams; the wrong answer if your bill becomes the bottleneck.

Tiers of managed

Hyperscale (AWS / GCP / Azure)

• AWS — most surface area, deepest service catalog, highest learning curve, hardest cost-control.
• GCP — cleanest networking, best Kubernetes (GKE), better data tooling.
• Azure — boring; great if you're already in Microsoft's stack; AD integration unmatched.
• Use EKS / GKE / AKS for k8s, ECS / Cloud Run / Container Apps for containers, Lambda / Cloud Functions / Functions for serverless.

App platforms (much less ops)

• ★ Fly.io — global VMs, native Postgres, fly.toml config; great for "run my app in 6 regions."
• ★ Railway — clean UI, instant Postgres / Redis, pay-as-you-go.
• ★ Render — Heroku-shaped; static sites, services, managed Postgres / Redis.
• Koyeb — global serverless containers with great DX.
• Vercel / Netlify / Cloudflare Pages — for frontend + serverless functions; see web-dev deployment.
• Heroku — still around; less recommended in 2026 due to pricing.
• DigitalOcean App Platform — Heroku-shaped; cheaper.

Database-only managed

• Neon — serverless Postgres, branching, free tier.
• Supabase — Postgres + auth + realtime + storage.
• Crunchy Bridge — gold-standard managed Postgres.
• PlanetScale (Vitess) — managed MySQL-compatible.
• MongoDB Atlas, Redis Cloud, Upstash for adjacent.

Hybrid (managed control plane, bring your own compute)

• Encore — TS / Go SDK; deploys to your AWS / GCP.
• Nuxt UI Pro / SST + AWS — TS-flavored IaC for serverless apps.
• Pulumi / Terraform Cloud — managed IaC state; you still run the infra.

What you give up

• Cost control. AWS bills surprise you; managed Postgres is 5–10× the price of self-hosted.
• Lock-in. Lambda / RDS / DynamoDB-specific APIs cost a lot to migrate off.
• Latency to data. Cross-AZ traffic, cold starts, NAT gateway costs.

What you gain

• Time. No 3 AM "the box is down" pages. (Mostly.)
• Compliance. SOC 2 / HIPAA / FedRAMP attestation comes free with the platform.
• Burst capacity. Autoscale to thousands of instances and back.
• Geographic reach. Fly / Vercel / Cloudflare put your code near users.

Common patterns

• Vercel + Neon + Cloudflare R2 + Resend — boring, fast, cheap-at-small-scale, expensive-at-large.
• Fly + Fly Postgres + R2 + Tigris — globally distributed without AWS pricing.
• AWS ECS Fargate + RDS + S3 + CloudFront — boring AWS, no k8s.
• Cloud Run + Cloud SQL + GCS + Cloudflare — GCP version; very ergonomic.
• Cloudflare Workers + D1 + R2 + KV — the "no servers, no AWS, pay-pennies" stack.

When the bill demands a rethink

• Compute > $5k/mo — consider Fly / Hetzner.
• NAT gateway > $200/mo — you've been bitten by AWS data transfer.
• RDS > $1500/mo — Neon or self-hosted Postgres on Hetzner.
• CloudFront / S3 egress big — move static assets to Cloudflare R2 (zero egress).
• Lambda > $1k/mo — containers on Fargate or Fly may be cheaper.

Pick this if…

• You'd rather write features than run boxes.
• You're at a stage where engineer time is more expensive than infra.
• Compliance / multi-region / autoscale matters out of the gate.
• The bill isn't the dominant constraint yet.

When the bill becomes the dominant constraint, self-host fleet or self-host PaaS are real options.