Scaling your cloud platform without losing cost control

Bruno, Rokt's compute platform lead, shares how a fast-growing Kubernetes platform faced a hard truth: cloud costs were rising faster than comfort. Instead of treating this as an infrastructure problem, the team redefined the platform as a product with measurable business outcomes.

Rokt measures overall cloud platform efficiency as cloud cost/revenue, connecting engineering decisions to company performance. This is supported by unit economics such as cost per vCPU/hour, spot adoption, and resource utilization, giving teams actionable levers.

Platform efficiency became a shared responsibility through three pillars:
• Infrastructure-level optimization: hardware selection, ARM and spot adoption, cloud provider discounts, and saving plans
• FinOps in developer workflows: cost attribution and showback, usage trends, workload classification, and resource utilization analysis
• Platform modernization: Karpenter adoption, automated vertical and horizontal workload scaling, and data processing job analysis and optimization

Key takeaways:
• Turning platform efficiency into a business metric leaders care about
• Designing unit economics that drive engineering behavior
• Sharing cost ownership without slowing developers
• Making cost optimization a built-in platform capability