3 billion nucleotides, but each nucleotide can be one of 4 bases, meaning that it's 6 billion bits of info, or (6e9 / 8) = 750 MB of data.
But if all of the sequence was used for data, then the sperm wouldn't be a sperm. If we keep the 20,000 coding genes making up ~ 2% of the genome, that still leaves us with (750 * 0.98) 735 MB.
But an organism is more than its gene templates, it also has functional regions where things bind and block things and join other things, and we're not entirely sure what percentage of the non-coding regions this is. I'm gonna go with 80%, and that leaves us with (750 * 0.18) = 135MB
Since the sperm cell is haploid data, it has 23 chromosomes instead of the 46 (23 pairs), so it has half the data redundancy of normal DNA. We might also need to add our own error correcting codes which will reduce some of the space. I'm gonna pull a factor of 3.6 out of my ass, and thus (135 / 3.6) - voila - 37.5 MB