<!-- Source: https://www.nitronedge.com/case-studies/rbc-borealis-ai-red-hat-openshift-nvidia/ -->
# RBC case study: a private AI cloud on OpenShift and NVIDIA

How Royal Bank of Canada and Borealis AI built a private AI platform on NVIDIA DGX systems orchestrated by Red Hat OpenShift.

## Key facts
- **Organisation:** Royal Bank of Canada (RBC) and Borealis AI
- **Country:** Canada
- **Sector:** Banking
- **Technology:** Red Hat OpenShift, NVIDIA
- **Published by:** Red Hat, 4 August 2020
- **Reported result:** A private GPU platform for research and production AI, orchestrated by OpenShift on NVIDIA DGX systems

Royal Bank of Canada (RBC) is Canada's largest bank. Its research institute, Borealis AI, founded in 2016, needed a way to move AI models from research into production at bank scale. Red Hat described the resulting private AI platform in 2020. It is an older story, but it shows a pattern many banks are only now adopting.

## The situation

AI and machine learning work needs both CPU and GPU capacity, and RBC's teams had been experimenting with several open source tools, including Apache Spark, Kubernetes and Slurm. The bank wanted one platform that could run research and production workloads together, inside its own data centres, with the security and compliance a bank requires.

## What they did

- Built the Borealis AI computing platform on NVIDIA DGX systems running Red Hat Enterprise Linux.
- Used Red Hat OpenShift to orchestrate the containers running on those GPU systems.
- Used the NVIDIA GPU Operator for OpenShift, so GPUs can be scheduled like any other resource, and NVIDIA's NGC catalogue for GPU-optimised software.
- Worked with Red Hat training and consulting teams alongside RBC's IT and development teams.

## Results reported

Red Hat's article describes the architecture and its purpose but does not publish performance, cost or timeline figures, so we do not quote any. The value of the story is the design: one controlled platform that takes models from research to production.

## What we take from it

**Private AI is not new, and it works.** Banks that need to keep data and models in-house have run GPU platforms on Kubernetes for years. The tooling has matured since 2020, which makes the path easier today.

**Treat GPUs as shared, scheduled capacity.** Making GPUs schedulable through the platform, rather than assigning machines to teams, is what lets research and production share expensive hardware.

**Research to production is the hard part.** A single platform for both removes the rebuild step where many AI projects stall.

## How this applies in our markets

Canadian banks preparing for OSFI's Guideline E-23 on model risk need clear control over where models run and how they are validated. A private, well-controlled platform helps. The same approach suits Gulf and Indian banks with strict data residency rules who want to run large models on their own hardware.

## FAQ

### Did NitronEdge work on this project?
No. This summarises Red Hat's published article about RBC and Borealis AI, with our own view and a link to the original.

### Why include a 2020 story?
It is one of the clearest published examples of a bank running a private GPU platform on Kubernetes, a design many banks are adopting now for AI agents.

## Sources
- [Red Hat blog: Powering digital transformation at Royal Bank of Canada with Red Hat platforms](https://www.redhat.com/en/blog/powering-digital-transformation-royal-bank-canada-red-hat-platforms)
