When does going serverless actually cost more than just using regular servers?

Serverless functions, like AWS Lambda or Azure Functions, are great for specific things. They handle the infrastructure for you, scale on their own, and you only pay for what you use. But I've seen situations where the serverless bill got surprisingly high, sometimes even more than a well-set-up server for the same job. Why does this happen? Are there certain apps or traffic patterns where paying per execution with serverless actually costs more?