What scale will we need to build for? 500 million transactions per day.

To support 500 million transactions per day, you need a horizontally scalable, event‑driven architecture. First, partition the workload by a deterministic key (e.g., user ID or account number) and shard the data across multiple database nodes. Use a distributed database such as Cassandra or a sharded PostgreSQL cluster with consistent hashing. For the transaction layer, expose stateless microservices behind a load balancer (e.g., NGINX or AWS ALB) and enable autoscaling based on CPU or request latency. Employ a message queue (Kafka, Pulsar) to decouple ingestion from processing, allowing back‑pressure handling and replayability. Cache frequently accessed data in a distributed cache (Redis Cluster) to reduce read load. Implement circuit breakers and retries to maintain resilience. Monitoring and observability are critical: use Prometheus/Grafana for metrics, distributed tracing (Jaeger), and alerting. Finally, design for eventual consistency where possible; use idempotent operations to avoid duplicate processing. Complexity per transaction remains O(1) for read/write operations, while the system’s overall throughput scales linearly with the number of shards and nodes.