AWS gives you the building blocks for Next.js: Amplify Hosting, ECS/Fargate, Lambda, S3, CloudFront and RDS. The right choice depends on whether you want a managed Next.js host or a custom container architecture.
Key takeaways
- Best for: teams with existing AWS accounts, compliance requirements or a need to colocate with RDS, S3 and other services.
- Not ideal for: small teams without cloud operations experience who only need a marketing site.
- Pin the Node version and keep production secrets out of the repository.
- Test auth, forms, images and redirects on a preview URL before promoting.
- Choose the region close to your database and users, and move long jobs to a worker.
What AWS expects from a Next.js app
A Next.js deployment has four parts: a build, a runtime, environment-specific configuration and persistent services. The build turns your routes into server and client output. The runtime serves dynamic requests and static assets. Environment values connect the app to its database and third-party APIs. Persistent services — database, object storage, queues — must live outside the app container.
| Layer | What to check before deploying |
|---|---|
| Build | Node version, install command, monorepo root, build output |
| Runtime | Serverless, container or static; timeout and memory limits |
| Configuration | Production and preview environment variables; secrets |
| Data | Managed database, connection pooling, backups and migrations |
| Files | Object storage for uploads; no reliance on the local filesystem |
| Observability | Logs, errors, uptime checks and a spending limit |
Step-by-step deployment
- Choose the runtime. Use Amplify Hosting for a managed Next.js build, or a container on ECS/Fargate if you need full control.
- Set up the network. Place the app and database in the same VPC and region. Use private subnets for the database and a load balancer for the app.
- Configure the build. Pin the Node version, install command and build output. Store environment values in SSM Parameter Store or Secrets Manager.
- Add storage and CDN. Use S3 for uploads and CloudFront for static assets and image delivery.
- Test the deploy. Check auth callbacks, database connectivity, file uploads and background jobs in a staging environment.
- Add observability. Enable CloudWatch logs, alarms, tracing and a cost budget before launch.
Environment variables and secrets
- Use IAM roles instead of long-lived access keys where possible.
- Store database credentials in Secrets Manager or SSM, not in task definitions.
- Separate staging and production accounts or at least separate VPCs and databases.
What usually breaks in production
| Symptom | Likely cause | Fix |
|---|---|---|
| Build passes locally, fails on the host | Different Node version or missing lockfile | Pin the runtime and commit the lockfile |
| Auth callback loops | Wrong callback URL or cookie domain | Set the production URL and secure cookie settings |
| Images 404 or unoptimised | Image loader or static export mismatch | Configure the loader or use a CDN |
| Database connection errors | Wrong region, pooling or TLS settings | Use the internal URL, a pooler and the required TLS mode |
| Uploads disappear | Ephemeral filesystem | Move uploads to object storage |
| Long request times out | Work is running in the request path | Move it to a queue or background worker |
Limits and cost on AWS
Costs depend on the chosen services. A container + RDS + load balancer is rarely free; a static export on S3 + CloudFront can be very cheap. Set a budget alarm.
Do not skip the preview
A preview deployment is not a production test if it uses a different database and a different auth callback. Test the real user journey with production-like data before you promote.
When to choose something else
Choose Vercel for a managed Next.js experience, or a simpler container host for a small team when the trade-offs above do not fit. The right platform is the one your team can operate safely at 2am, not the one with the best demo.
Pre-launch checklist
- The production domain, HTTPS and canonical URL are correct.
- Auth callbacks, cookies and email links use the production origin.
- Database migrations have run, and a restore has been tested.
- Every environment variable exists in production, not just preview.
- A test order, form or sign-up completes end to end.
- Robots, sitemap, analytics and consent behave as expected.
- The error page, 404 and a hard refresh on a deep route all work.
Rollback and recovery
| Failure | Immediate action | What to verify |
|---|---|---|
| Bad application release | Promote the previous build | The database change is backward compatible |
| Broken migration | Stop writes and restore from a verified backup | Data loss window and reconciliation |
| Auth outage | Roll back the auth configuration | Sessions and callback URLs |
| Payment errors | Disable the failing method | Pending orders and webhook retries |
| Traffic spike | Scale or enable a queue | Database connections and rate limits |
| Region incident | Fail over or restore elsewhere | DNS, data replication and RPO |
Observability after launch
| Signal | What to watch | Why |
|---|---|---|
| Errors | New error groups and affected routes | Catch regressions before support does |
| Latency | p95 and p99 by route | Find slow queries and cold starts |
| Database | Connections, slow queries and storage | The most common bottleneck |
| Jobs | Queue depth and failed retries | Silent backlog |
| Vitals | LCP, INP and CLS by device | Real-user experience |
| Cost | Spend and egress | Avoid a surprise invoice |
Frequently asked questions
Not necessarily. Check the runtime features you use: Server Actions, dynamic routes, image optimisation, middleware, WebSockets, background jobs and native dependencies. Small teams without cloud operations experience who only need a marketing site.
Use a monorepo when the app shares code or types with other packages and the platform supports a root directory. Use a separate repository when the deployment is simpler and the app has no shared build pipeline. The important thing is to pin the root, install command and lockfile.
Run migrations as a separate deployment step with a lock, not on every app boot. Back up first, test against a staging database and have a rollback plan. If the platform has no release phase, run the migration from CI before promoting the app.
Error rate, p95 response time, database connections, background-job failures, Core Web Vitals and the main conversion event. Add an uptime check from outside the platform and a billing alert.
Sources and further reading
- Next.js deployment guide (nextjs.org)
- Next.js self-hosting guide (nextjs.org)
- AWS Amplify Hosting documentation (docs.aws.amazon.com)
Related reading
Need a deployment review?
Send the stack, the hosting plan and the launch date. I will flag the runtime limits, data risks and missing checks before you go live.
How to cite this guide: Nur Hossain Repon, “Deploy a Next.js app to AWS: step-by-step guide”, Nhrepon Portfolio, reviewed October 2026, https://nhrepon.com/posts/deploy-a-next-js-app-to-aws-step-by-step-guide.





Comments
No comments yet.