Looks like the way to do it is to write to loom format via loomR, then read that into anndata to be written as an .h5ad file.

Single cell folks need to a pick a file format/structure and stick with it.

EDIT (02/05/2021): The recent zellkonverter package from Bioc provides an easier way to do this.

In R, you can convert Seurat object to h5ad and vice versa using sceasy.

The best way now is to manually create your h5ad from seurat objects, such as put all normalized value as adata.X , raw counts to adata.raw, all gene annotations to adata.var and all cell annotations to adata.obs, etc

Or use scanpy from the very beginning.

You can also try to use the srt_to_adata function of SCP package (https://github.com/zhanghao-njmu/SCP)

library(SCP)
data("pancreas1k")
adata <- srt_to_adata(pancreas1k)
adata$write_h5ad("pancreas1k.h5ad")