A developer must provide a custom machine-learning library (currently 15 GB and growing) to all AWS Lambda functions in an application. Which approach satisfies this requirement so every Lambda function can access the library?

The correct approach is to use Amazon EFS. Lambda layers have a size limit of 250 MB (unzipped), which is significantly smaller than the 15 GB library. Storing the library in Amazon S3 and downloading it at runtime would introduce significant latency and incur data transfer costs for every invocation. While Lambda container images support larger sizes (up to 10 GB), the question states the library is "15 GB and growing," exceeding the container image limit. Amazon EFS allows Lambda functions to access shared file systems up to multiple petabytes, making it the ideal solution for large, growing libraries that need to be shared across multiple functions.