Cloud Cost Optimisation

February 16, 2026

How to reduce cloud costs as a startup: AWS credits, Vercel vs Render vs Railway pricing, avoiding auto-scaling traps, and basic FinOps practices.

AWS Startup Credits

AWS Activate provides $1,000 to $100,000 in AWS credits depending on your application tier. Founders who apply through an accredited accelerator — YC, Techstars, and most major national accelerators qualify for the higher tiers — can access $25,000 to $100,000 in credits, which can cover a year or more of infrastructure costs for an early-stage company. Apply immediately after incorporation; the credits expire, and there's no benefit to waiting. GCP and Azure have equivalent programs — Google for Startups Workspace and Microsoft for Startups — offering similar credit amounts, and there's nothing preventing you from applying to all three.

Credits change the economics of architectural decisions. During the credit period, running a slightly over-engineered infrastructure to learn what works costs nothing. After the credits expire, that same architecture might cost $2,000 per month. The mistake founders make is designing infrastructure during the credit period without building any awareness of what each service costs at market rate. Set up AWS Cost Explorer during the credit period, tag every resource with its purpose, and know your cost per component before the credits run out. The transition from free credits to paid infrastructure is the moment most founders discover they're paying $800 per month for a NAT gateway that's easier to replace.

Vercel vs Render vs Railway Pricing

Vercel Pro costs $20 per month for teams plus bandwidth overages charged at $0.15 per GB beyond the included allocation. For most startups, this is affordable — until a piece of content goes viral. A single Hacker News front page appearance can generate 50,000–200,000 pageviews in 24 hours, and if your pages are large, the bandwidth bill can easily reach $500 in a day. Vercel's Edge Network is fast and its preview deployment system is genuinely excellent for development velocity, but set a spend alert before deploying anything that could attract sudden traffic spikes.

Render starts at $7 per month for a static site and $25 per month for a web service, with one important catch: free-tier web services sleep after 15 minutes of inactivity, causing a 30-second cold start on the next request. This is acceptable for development environments but unacceptable for production. Railway includes $5 per month in usage credits and charges usage-based pricing beyond that, with PostgreSQL and Redis included in the same platform. Railway doesn't have a sleep mode, making it preferable to Render's free tier for production. For most seed-stage startups, Railway provides the best balance of simplicity and cost — a full application stack (app server, PostgreSQL, Redis) runs $15–$40 per month with zero infrastructure configuration.

Auto-Scaling Cost Traps

Auto-scaling sounds like a feature that protects you from both underprovisioning and overpaying. In practice, several configurations can generate unexpectedly large bills. The most dangerous is a misconfigured serverless function: an infinite loop or a function that calls itself recursively in AWS Lambda or Vercel Edge Functions can run millions of invocations before the billing alert triggers. Lambda charges $0.0000166667 per GB-second — individually negligible, but a runaway function running 100 GB-seconds per second costs $1,500 per day. The mitigation is a concurrency limit on every Lambda function and a billing alert set to $100.

AWS Cost Explorer and Budget Alerts are the simplest FinOps tools available, and most startups never configure them. Create a monthly budget in AWS Billing for $100, with an email alert when spending reaches 80% of the budget. This single step prevents the scenario that every AWS-using founder has heard about: waking up to a $3,000 bill from a configuration error. Enable Cost Explorer from day one so you have 90 days of billing history when you need to diagnose a cost spike. Tag every resource with environment (production, staging, development) and service (api, worker, database) — untagged resources are unattributable costs that become impossible to optimise.

FinOps Basics

FinOps — financial operations for cloud infrastructure — is the practice of making cost-aware infrastructure decisions. For a startup, it boils down to three habits: know your cost per user, know your cost per feature, and set alerts before problems occur rather than responding after. Cost per user is your total infrastructure bill divided by monthly active users; if this number is growing faster than your revenue per user, your unit economics are deteriorating and the problem is infrastructure architecture, not sales.

Reserved instances and savings plans are the largest levers for reducing AWS costs once you have predictable baseline usage. A one-year reserved EC2 instance costs 40% less than on-demand pricing for the same instance type; a three-year reservation saves 60%. The risk is commitment: if your architecture changes and you no longer need the reserved instance, you're still paying for it. Buy reservations only for resources that have been stable for three or more months. Spot instances are even cheaper — 70–90% off on-demand — but they can be interrupted with two minutes' notice, making them suitable only for batch processing, training jobs, and other fault-tolerant workloads.

Frequently Asked Questions

How do I apply for AWS Activate startup credits? Go to aws.amazon.com/activate and apply through the Founders tier (no accelerator required, $1,000 in credits) or the Portfolio tier (requires an accelerator or VC firm that is an AWS Activate partner). The application takes 15 minutes and typically processes within 2 business days. YC, Techstars, and most major accelerators are Activate partners; check your accelerator's benefits package before applying independently.

What is a NAT gateway and why does it cost so much? A NAT (Network Address Translation) gateway allows resources in a private subnet to connect to the internet without being publicly accessible. AWS charges $0.045 per hour ($32.40 per month) plus $0.045 per GB of data processed. For startups that put RDS databases or Lambda functions in private subnets for security, the NAT gateway bill can surprise them. Consider VPC Endpoints for AWS services (S3, DynamoDB) to avoid NAT gateway data transfer charges for those specific services.

Is Vercel worth it compared to self-hosting on a VPS? For most Next.js applications at early stage, yes. Vercel's preview deployments, edge network, and zero-config deployment pipeline save significant engineering time compared to configuring Nginx, SSL, CI/CD, and monitoring on a VPS. The break-even point — where self-hosting becomes cheaper — is around $500–$1,000 per month in Vercel bills, which corresponds to significant scale. Before that point, the engineering time cost of self-hosting exceeds the hosting cost savings.

What are the cheapest ways to host a Next.js app in production? Railway is the lowest-friction paid option at $15–$40 per month for a full stack. Coolify is an open-source Heroku alternative that you self-host on a $6/month VPS from Hetzner or DigitalOcean — the total cost is $6–$10 per month at the cost of initial setup time. Fly.io offers a free tier for small applications and competitive pricing for production workloads. All three options are significantly cheaper than Vercel at medium scale.

How do I identify which AWS service is generating unexpected costs? Open AWS Cost Explorer, group by service, and set the time range to the period when costs increased. Identify the service responsible, then group by usage type to find the specific charge. Common culprits: data transfer costs from resources in different availability zones, NAT gateway data processing, CloudWatch log storage accumulating for months, and S3 request costs from frequent small-object reads. Once identified, the fix is usually a configuration change rather than a service replacement.

Related Turkish Products