AWS re:Invent 2025 - Architecting for hypergrowth: Scaling to 200 million users w/ Skyscanner-ARC209

Scaling to 200 Million Users: Lessons from Skyscanner's AWS Journey

Building an Initial Architecture

• Start with a simple 3-tier web application architecture - front-end, back-end, and data store
• Use a spectrum of compute options on AWS, from serverless (Lambda) to more managed containers (ECS, EKS, Fargate)
• Choose a middle-ground like EKS to balance control and operational overhead
• Use a purpose-built API service like API Gateway, ALB, or AppSync to expose business logic to the front-end

Scaling the Data Tier

• Begin with a relational database like Amazon Aurora, then consider NoSQL options like DynamoDB for specific use cases
• Leverage multi-region Aurora DSQL for global reach and high availability
• Implement caching with services like Elasticache to speed up reads

Scaling the Back-end

• Leverage Kubernetes autoscaling features like horizontal and vertical pod scaling
• Use cluster autoscaling with tools like Cluster Autoscaler and Karpenter to dynamically provision worker nodes
• Adopt a "cellular" Kubernetes architecture with bounded failure domains

Skyscanner's 10-Year Scaling Journey

• Started with a hybrid on-premises and AWS architecture for burst capacity
• Moved to containerized microservices on ECS, then Kubernetes (EKS)
• Evolved to a "cellular" EKS architecture with bounded failure domains
• Leveraged AWS services like CloudFront, Route53, NLBs, and caching to scale their flights pricing service
• Emphasized cost control, observability, and cultural/organizational scaling tactics

Key Lessons Learned

• Be opinionated - choose a small set of standardized, hardened technologies
• Manage blast radius and failure propagation through architectural patterns
• Speak the business language of cost and value, make it everyone's responsibility
• Stay pragmatic - less-than-perfect architectures can still serve massive scale
• Invest in observability, control plane simplicity, and operational readiness

Technical Details and Metrics

• Skyscanner serves 160M+ monthly users, handles 100B+ flight prices per day
• Runs 300+ Java microservices, 24 production Kubernetes clusters, 37,000+ cores
• Processes 400,000 service-to-service requests per second, manages hundreds of terabytes of caching
• Emits 55 billion data events per day into a 25PB data lake

Business Impact

• Skyscanner's scalable, cloud-native architecture enabled them to grow from a small startup to a global travel marketplace serving hundreds of millions of users
• Their ability to rapidly iterate, measure, and improve their architecture allowed them to keep pace with explosive growth in user demand and data volumes
• The lessons they learned around managing blast radius, cost control, and organizational scaling tactics are applicable to any business facing hypergrowth challenges

Examples and Use Cases

• Skyscanner's "cellular" Kubernetes architecture with bounded failure domains
• Use of AWS services like CloudFront, Route53, NLBs, and caching to scale their flights pricing service
• Adoption of open-source tools like Karpenter for efficient Kubernetes autoscaling