AWS API Gateway Is Built for AWS. KrakenD Runs Anywhere.

AWS API Gateway is a managed service, not a portable gateway. Its authentication integrates with IAM. Its compute integrates with Lambda. Its edge integrates with CloudFront. Inside AWS, that integration is seamless. Outside AWS, it does not exist. KrakenD runs as a Docker image on any infrastructure: AWS, GCP, Azure, on-premise, air-gapped. The same config file, the same CI/CD pipeline, the same performance characteristics regardless of where it runs.

It depends on what you are managing. AWS API Gateway removes infrastructure management but adds AWS-specific configuration, IAM policy management, Lambda dependency management, and a cost model that requires active monitoring. KrakenD adds a Docker image to your existing deployment pipeline. For teams already running Kubernetes workloads, adding KrakenD is one more deployment. The migration assessment call will tell you which category you are in.

Three things worth being clear about. First, native AWS integrations: Lambda triggers, IAM authorization, CloudWatch native metrics, and CloudFront edge integration require rewiring. Second, the managed service model: KrakenD runs on your infrastructure, you own the deployment and operations. Third, the AWS free tier: one million API calls per month free for 12 months. What you gain: a gateway that runs anywhere, costs flat per tier regardless of traffic, and has no AWS account dependency.

Yes. KrakenD routes to Lambda Function URLs or any HTTP endpoint. Lambda backends continue working during and after migration. The difference is that KrakenD is not dependent on Lambda as a deployment model, only as a backend option.

Yes. KrakenD runs as a standard Docker container and can be deployed inside a VPC on EKS, ECS, or EC2. It has no requirement for internet connectivity. For air-gapped or private subnet deployments, the same Docker image and config apply.

AWS API Gateway pricing is public: $3.50 per million REST API calls, $1.00 per million HTTP API calls. KrakenD pricing is flat per plan tier and does not scale with traffic. The migration assessment call will calculate the crossover point for your specific call volume and instance requirements, using your actual numbers rather than estimates.

Several. Data transfer out is billed separately at standard AWS Data Transfer rates. Caching on REST APIs is optional but billed hourly by cache size, a fixed cost that runs whether or not you use it heavily. WebSocket APIs add a per-million connection-minute charge on top of the per-message fee. Execution logs and detailed CloudWatch metrics are billed separately for ingestion and storage. Private APIs behind a VPC endpoint add an hourly PrivateLink charge plus per-GB data processing. None of these show up in the headline per-million-request rate, and all of them push the real bill higher than the request count alone suggests.

AWS's AI Gateway capabilities are integrated with Amazon Bedrock and optimized for AWS AI services. For teams running LLMs exclusively on Bedrock, that integration works. For teams using OpenAI, Anthropic, Azure OpenAI, Gemini, or on-premise inference alongside Bedrock, AWS AI Gateway does not govern that traffic. KrakenD's AI Gateway is native config: LLM routing, prompt guarding, semantic caching, token budgets, MCP server. Any provider, any infrastructure, the same config governs OpenAI, Bedrock, and your on-premise inference hardware from the same file.