Uber rewrote its ledger systems five times in ten years—and at least one of those rewrites was entirely preventable. The company migrated its payment platform to DynamoDB in 2017, betting on the database's global scale and throughput. What everyone involved missed: DynamoDB charges for every read and write, making it prohibitively expensive for a system that processes millions of transactions daily.

Each Uber trip generated multiple ledger entries, and with the company handling roughly 15 million trips per day, costs spiraled. By 2020, Uber had accumulated 1.2 petabytes of data in DynamoDB. The write operations alone cost approximately $250,000 annually, scaling to around $2.25 million by year three with growth factored in. Storage added another $300,000 per month. The total bill reached roughly $8 million before Uber pivoted.

The real problem wasn't DynamoDB itself—it's a solid choice for payment systems that can sacrifice global consistency for availability. But ledgers require strict serializability across all data. Uber essentially paid millions for a database that couldn't do what their ledger needed. They migrated to an internal solution called LSG built on DocStore, but this could have been avoided with basic napkin math upfront. History remembers the DynamoDB design as a success; the $8 million price tag tells a different story.