In the meantime, maybe you should look into filtering some of these data to reduce the footprint of this analysis. @gregcaporaso recommends considering keeping only OTUs that appear in more than one sample in his Analyzing paired end reads tutorial. As well, @Nicholas_Bokulich recommends quality filtering prior to clustering (--p-min-quality 20 or --p-min-quality 30 when running qiime quality-filter q-score), and then removing singletons with an abundance filter (and/or chimera filtering):
https://www.nature.com/articles/nmeth.2276
To be clear, the error you are observing is actually from MAFFT, the underlying alignment tool being used by q2-alignment, so you would see this same error if you ran the same data directly with MAFFT. Maybe you can work around this large amount of memory that MAFFT is trying to use by performing some of the cleanup steps I suggest above. Please let us know how it goes! 