What a genome really costs to analyse in the cloud
We broke down compute, storage and egress for 10,000 genomes to see where the money goes.

Cloud pricing pages make it hard to answer a simple question: how much does turning one genome's reads into a VCF actually cost? We ran 10,000 genomes to find out.
Compute is not the biggest line
Alignment and calling dominate runtime, but with spot capacity and GPU acceleration they are no longer the largest cost for most labs. Over a year, storage of raw reads and intermediate files often costs more.
Storage adds up quietly
A 30x genome produces around 100 GB of FASTQ and BAM. Keeping everything in hot storage for a year costs more than the analysis itself.
- Move raw reads to archive tiers after delivery
- Keep CRAM rather than BAM where your workflows allow
- Delete intermediates you can regenerate
Egress is the surprise
Moving data out of the cloud is where budgets break. Run analysis in the same region as your storage, and send results, not reads, to downstream tools.
What we changed
We now default to compressed formats, tiered storage and region-pinned compute. For a typical lab that cut the all-in cost per genome by just over 40%.


