How UPI Handles Millions of Simultaneous Payments Without Crashing

India's Unified Payments Interface faces extreme traffic spikes on salary days, when simultaneous transactions can surge from roughly 100,000 to over one million requests per second. Rather than relying on a single powerful server, UPI-style systems distribute traffic across many machines using load balancers and horizontal scaling, so that no single point of failure can bring down the entire network. However, scaling the application layer alone is insufficient, as concurrent requests accessing the same bank account can trigger race conditions that allow more money to be spent than actually exists. Solving this requires additional techniques such as database-level concurrency controls, idempotency, and distributed consistency mechanisms. Together, these challenges make large-scale payment systems one of the most complex problems in distributed systems design.