Why did one day of AI cost more than a month of servers?

A SaaS application experienced a massive spike in LLM API costs on a single day—costing more than an entire month of server fees—due to a retry loop triggered by a database schema mismatch. The batch job successfully called multiple LLMs and received billable results, but then failed when trying to write to a non-existent database column, causing the system to automatically retry the entire process 21 times, each time incurring full LLM charges despite the underlying problem never being fixed.