Cloud and DevOps are where a lot of startup money quietly evaporates. Founders read about the stacks used by Stripe and Notion, hire someone to set up the same thing, and three months later they're paying ₹40,000 a month for infrastructure that serves 200 daily users. Here's what actually matters at each stage — and what's safe to ignore until much later.
Stage 1: 0–10,000 Users (Pre-PMF)
You need: a single managed platform that handles deploys, scaling, and SSL automatically. Vercel for frontend and Next.js apps. Render or Railway for backend services that need long-running processes. Supabase or Neon for Postgres. Resend for transactional email. Cloudflare in front of everything.
You don't need: Kubernetes, Terraform, a dedicated CI/CD pipeline beyond what your platform gives you, multiple cloud regions, a DevOps engineer.
Total monthly cost: ₹2,000–₹15,000 depending on usage. Time to set up: 2–3 days.
Stage 2: 10,000–500,000 Users (Growth)
You need: proper CI/CD with automated tests, staging environments that mirror production, observability (Sentry for errors, Vercel Analytics or Plausible for traffic, basic uptime monitoring), automated database backups with point-in-time recovery, secrets management.
You probably need: a small DevOps person (part-time or fractional) once your engineering team passes 6–8 people. Infrastructure-as-code starts to pay off here — Terraform or Pulumi for the cloud bits, GitHub Actions for the rest.
You don't need: a service mesh, multi-region deploys, a SOC 2 audit (unless you're selling to enterprises that demand it), self-hosted databases.
Total monthly cost: ₹40,000–₹3 lakhs depending on traffic and data volume.
Stage 3: 500,000+ Users or Enterprise Customers
Now the rules change. You probably need: AWS or GCP for the workloads where managed platforms become expensive, Kubernetes if your workload genuinely needs container orchestration (most don't), proper IAM and least-privilege access, audit logging, multi-region failover for tier-1 services, SOC 2 / ISO 27001 if enterprise sales need it. A real DevOps or platform team becomes essential — at this scale, broken infrastructure costs you customers, not just engineering hours.
The Mistakes We See Most Often
- Choosing AWS on day one because "that's what real companies use." Vercel + Supabase will get you to ₹20 crore in ARR. Switch when you actually hit the limits.
- Hiring a full-time DevOps engineer when you have three backend services. They'll over-engineer everything to justify their existence.
- Setting up Kubernetes for an app that gets 500 requests per minute. You're paying for complexity you don't need.
- Skipping observability until something breaks. Sentry costs ₹3,000/month and pays for itself the first incident.
- Building custom CI/CD pipelines when GitHub Actions or Vercel deploys do the same thing for free.
What "Good Enough DevOps" Looks Like
A startup with solid DevOps in 2026 has: every deploy goes through CI with tests, every commit to main hits a preview environment, production deploys are one-click or automatic, secrets aren't in git, errors page someone within minutes, backups exist and have been tested at least once, and you can describe how to recover from a complete database loss in under 30 minutes. That's it. That's the bar.
How Starduxe Sets This Up for Clients
Most of our projects launch on Vercel + Supabase + Cloudflare with monitoring via Sentry and Plausible. CI runs on GitHub Actions. Database migrations are versioned. Secrets live in Vercel's environment manager. Total setup time on a new project: under a day. Total monthly cost for a typical SaaS at launch: under ₹10,000. We move clients to AWS or GCP only when we hit a specific limit that justifies the complexity — never before.
Infrastructure is a tool, not a trophy. The best DevOps decision is usually the boring one that lets you ship faster.