instead of uploading large files through fcrepo, upload them straight to s3. then create an external file in fcrepo, keeping the data consistent and making sure hyrax uses the right file all the time. This allows the files to be streamed to S3 and to still be represented in fcrepo in some way. They can also be direct downloaded from S3 and will work in serverless-iiif (but you don't really want to serve something that big from serverless-iiif)
instead of uploading large files through fcrepo, upload them straight to s3. then create an external file in fcrepo, keeping the data consistent and making sure hyrax uses the right file all the time. This allows the files to be streamed to S3 and to still be represented in fcrepo in some way. They can also be direct downloaded from S3 and will work in serverless-iiif (but you don't really want to serve something that big from serverless-iiif)