FinOps isn't a finance team problem — it's an engineering problem.
And on Kubernetes, it's one of the hardest engineering problems
to solve well. Over-provisioned node groups, always-on batch jobs,
statically sized data clusters, and idle Spark executors silently
drain budgets while engineering teams remain focused on features.

At Zscaler's Data Fabric team, we run a large-scale Kubernetes
platform on Amazon EKS spanning Apache Spark, Apache Flink,
ClickHouse, Kafka (Strimzi), and RisingWave — across multiple
cells and environments. Over the course of a focused FinOps
initiative, we reduced our cloud infrastructure costs by over
40% — without rewriting applications, without reducing capacity,
and without impacting SLAs.

This talk tells the complete story: the audit that revealed where
the waste was, the architecture changes we made, the tools we
used, and the numbers before and after.