Are multiple samples showing this or just some? It is difficult to divine a specific reason but it is possible that some of the library fragments do not have a functional adapter on second end and thus failed to prime/generate sequencing signal.

As long as you have enough data you should simply ignore these errant reads (and their R1 counterparts) and move on with the analysis with the rest of the data. Most aligners do not allow you to mix single- and paired-end reads while aligning (bbmap does) and discarding those 3% reads is not going to make a big difference.

NextSeq2000 is a dual-color chemistry Illumina machine

G is being called when there's no fluorescence signal with the dual color chemistries. It appears that the second primer might not have bound.

source: https://www.reddit.com/r/bioinformatics/comments/16n116z/r2_novaseq_full_of_gs/