Confluent Cloud's architecture is composed of a control plane and data plane. The data plane includes a network architecture that abstracts away individual brokers, allowing for dynamic scaling without requiring configuration changes in clients. A load balancer is used to proxy traffic to a randomly chosen, healthy broker, and an additional layer 4 load balancer provides abstraction from the brokers altogether. The control plane consists of a mothership and satellites, which communicate via Kafka. The mothership receives API requests and makes scheduling decisions, while satellites manage the lifecycle of physical clusters and report their status to the control plane. A physical stateful cluster (PSC) declaratively describes the desired state of a physical cluster, and the Confluent Operator materializes this into Kubernetes-native objects. Confluent Cloud has benefited Kafka by providing a cloud-first organization, allowing for rapid testing and deployment of new features, and enabling improvements to Kafka itself, such as improved fault tolerance and scalability. The architecture has also helped to learn valuable lessons about API design, third-party software dependencies, and cost modeling, which can be applied to other projects.