Secondary analysis of whole‑genome sequencing (WGS) data, comprising read alignment, variant calling, and annotation, generates approximately 1 TB per human genome and requires nearly 1,000 CPU‑hours, creating a major computational bottleneck for large‑scale projects such as Indonesia’s goal of sequencing 100,000 genomes by 2025. NVIDIA Parabricks v4.4.0 was deployed on a Lintasarta Cloudeka Kubernetes platform powered by H100 GPUs to accelerate a standard DNA‑seq pipeline. FASTQ files were aligned to GRCh38 using GPU‑accelerated BWA‑MEM, followed by GPU‑enabled Base Quality Score Recalibration and GATK4 HaplotypeCaller executed within Kubernetes jobs. Alignment and sorting completed in 2.5 minutes (peak 3.35 × 10⁹ bases/GPU/min), BQSR in 1.5 minutes, and variant calling in 7 minutes; totaling ~11 minutes per genome versus ~8 hours on 32‑core CPU servers (∼26× speed‑up), enabling processing of 130 genomes per GPU per day and a 98% runtime reduction. Cost analysis indicates an 82% hourly cost saving and a 97% reduction in energy expenditure, demonstrating that a GPU‑as‑a‑service model can dramatically accelerate and reduce the cost of large‑scale secondary genome analysis to support national biobank initiatives and democratize access in resource‑constrained environments.

Keywords: Secondary genome analysis; Cloud GPU; Kubernetes; Bioinformatics; Genomics.