How Nubank Leveraged Kubernetes for Cloud‑Native Transformation and Cost Efficiency

Business Challenges

Nubank, founded in Brazil in 2013, avoided legacy infrastructure but faced scaling issues as its user base grew to 23 million and the engineering team expanded from 30 to over 500 members.

Deployments required rotating entire stacks or cloning infrastructure, becoming slower and painful, with additional pain points such as load‑balancing and adding new security‑group rules in AWS.

Solution

Early adoption of Docker led the team to evaluate orchestrators; after testing Mesos, Docker Swarm, and others, they selected Kubernetes for its rich abstractions, community support, and Google’s Borg expertise.

The migration began by enabling developers to run cloud‑native applications on Kubernetes, initially considering Minikube for testing and staging, then moving staging workloads to a dedicated cluster.

They built a custom Kubernetes distribution using kops‑like reverse engineering, later switching to kubeadm for cluster creation, and conducted staged rollouts with about 20 customers before migrating additional partitions.

The platform incorporates monitoring with Prometheus, Thanos, Grafana, and logging with Fluentd.

“With Kubernetes and canary deployments, rolling back a change became much easier; deployments are faster, and teams deploy more frequently with greater confidence.” — Nubank Engineering Director Renan Capaverde

To aid engineers, Nubank held company‑wide training and created the NuCLI command‑line tool with over 500 automation shortcuts, simplifying log extraction and other tasks.

Initially, as a Clojure‑centric service, they faced CPU and JVM tuning challenges; after a year and a half, they refined horizontal scaling and budgeting, estimating a ~30% improvement in cost efficiency.

Further upgrades include moving to Kubernetes 1.16 and a new kernel version that better integrates JVM container support.

Kubernetes’s abstractions such as readiness and liveness probes, blue/green deployments, and the ability to run 400+ micro‑services without waiting for EC2 instances dramatically reduced deployment time from 90 minutes to 15 minutes.

“During the migration from staging to production we received abundant feedback, allowing us to improve the infrastructure, tools, and pre‑production processes.” — Nubank Software Engineer Yago Nobre

Kubernetes also facilitated expansion into new countries, simplifying the creation of new AWS accounts and accelerating infrastructure and application iteration.

Over three years, the Kubernetes, Kafka, monitoring, and related platform team grew from 5 to 50 specialists, accumulating valuable lessons to share with other organizations.

Capaverde cautions that Kubernetes is not a cure‑all; it must be understood, controlled, and tamed, but when managed properly it enables long‑term scalability and effective abstraction.

Enterprise Benefits

Developer experience improved dramatically; production deployment time dropped from 90 minutes to 15 minutes, engineers now deploy roughly 700 times per week, and overall cost efficiency increased by about 30%.