Managing infrastructure for billions of daily transactions across 200+ critical systems is no small feat. For global banks like Santander, the complexity of expanding into new financial services created a major bottleneck: provisioning new environments could take up to 90 days. This intensive operational effort stifled innovation and agility. The solution wasn't just another tool, but a fundamental shift in approach through platform engineering. This post explores Santander's 'Catalyst' platform, a transformative initiative developed with AWS, which turned months of waiting into hours—or even minutes—of automated deployment. You can read the full case study in the original AWS Architecture Blog post.
The Catalyst Architecture: A Control Plane for the Cloud
Catalyst's core is a control plane cluster built on Amazon EKS (Elastic Kubernetes Service), acting as the central brain for orchestration. The magic happens through three key components:
- Universal Provisioning with Crossplane: Instead of managing each cloud service's unique API, Crossplane acts as a declarative resource provisioner. Developers define what they need (e.g., 'a database with these specs'), and Crossplane handles the 'how' across AWS and other providers, ensuring consistency.
- GitOps-Driven Deployment with ArgoCD: All infrastructure and application stacks are defined as code in Git repositories. ArgoCD continuously synchronizes the live state with the desired state in Git, enabling automated, auditable, and rollback-safe deployments.
- Centralized Governance with Open Policy Agent (OPA): A policies catalog uses OPA to enforce security, compliance, and architectural standards automatically before any resource is provisioned. No more manual checklists.
This architecture abstracted the underlying cloud complexity, providing developers with a unified, self-service portal while giving platform teams centralized control and visibility.
Beyond Efficiency: Catalyzing Strategic Workloads
The true test of a platform is its ability to support complex, business-critical workloads. Catalyst excelled here:
| Workload | Key Components | Business Impact |
|---|---|---|
| Generative AI Agents Stack | Amazon Bedrock, S3, KMS, custom IAM | Reduced implementation time from 105 days to 24 hours, eliminating dozens of provisioning tickets. |
| Modern Data Platform | Databricks integration, data lakes, automated ETL | Reduced ~3,000 monthly tickets for data experimentation environment setups. |
| Cloud Process Orchestration | AWS Step Functions, retry patterns, centralized monitoring | Modernized legacy workflows, improving reliability and observability. |
Limitations and Considerations
While the results are impressive, this approach has its nuances:
- Significant Upfront Investment: Building a mature internal platform requires dedicated, skilled teams (platform engineering, DevOps) and time. It's a strategic, long-term investment, not a quick fix.
- Vendor Lock-in & Complexity: Heavy reliance on specific tools (Crossplane, ArgoCD, OPA) creates a deep knowledge dependency. The control plane itself becomes a critical system to maintain and secure.
- Cultural Shift is Mandatory: The technology fails if developers don't adopt the self-service model. It requires fostering a culture of trust, ownership, and collaboration between platform and product teams.
Conclusion and Your Next Steps
Santander's Catalyst demonstrates that platform engineering is a powerful lever for digital transformation, moving from being a cost center to an innovation enabler. The gains in speed, standardization, and security are transformative.
What's the next step for you?
- Assess Your Friction Points: Are your developers spending more time managing infrastructure than building features? Is compliance a manual, slow process?
- Start with a Product Mindset: Treat your internal platform as a product. Identify your key 'customers' (development teams) and their most painful workflows.
- Iterate on a Foundation: You don't need to build everything at once. Start by standardizing one common stack (e.g., a containerized web app) using GitOps and policy-as-code on a managed Kubernetes service.
Platform engineering is reshaping how enterprises use the cloud. As seen in other infrastructure innovations like the Microsoft Maia 200 AI Inference Accelerator, the focus is on delivering specialized, efficient capabilities at scale. Similarly, understanding the full stack of modern development, from hardware to application, is key. For instance, just as AI upscaling transforms visual fidelity in gaming as discussed in our look at NVIDIA's DLSS technology, a robust platform transforms development velocity and operational reliability in the enterprise.