Behind the scenes of cloud: how I think about multi-cloud network architecture

July 2026 · 7 min read cloud architecture

Working in a mission-critical environment, I've learned that the starting point of any architecture is never the tool — it's having a minimal vision of where I want to get to. Before drawing any topology, I think about how it will positively impact the environment we're operating in. Multi-cloud is no exception.

Thinking about multi-cloud network architecture means, in practice, intentionally connecting multiple cloud providers — and, most of the time, the local datacenter too — to optimize workloads. And "intentionally" is the right word: every decision carries a trade-off between latency across clouds, data egress cost, end-to-end security complexity, and standardized management and governance.

Good multi-cloud isn't about using every cloud. It's about using the right one at the right time.

The pillars behind this architecture

Over time, I've learned to organize this complexity into five fronts. They aren't independent — a decision in one almost always pulls a thread in the other four — but they help structure the thinking when designing a project.

if: connectivity

1. Cloud-to-cloud connectivity and traffic

  • Avoid relying on the public internet for sensitive traffic — prioritize private, virtual connections (Direct Connect, ExpressRoute, Interconnect).
  • Use PoPs or neutral datacenters as convergence points between clouds, cutting hops and latency.
if: egress

2. Data egress costs

  • Map the data flow and keep frequently-communicating services within the same region or provider.
  • Place interdependent databases and applications geographically close, to avoid unexpected spikes on the bill.
if: security

3. Security and identity

  • Adopt a Zero Trust model — "never trust, always verify" — with continuous validation and MFA on every access.
  • Unify firewalls, IAM and compliance with cloud-native platforms (CNAPP), instead of managing each environment in isolation.
if: observability

4. Visibility and monitoring

  • Standardize logs and metrics. Every cloud has its own dashboard (CloudWatch, Azure Monitor, GCP Operations) — centralizing everything in one tool avoids diagnostic blind spots.
if: governance

5. Governance and automation

  • Automate provisioning with cloud-agnostic tools like Terraform, ensuring consistent configuration and cutting manual error.
  • Standardize on Kubernetes and APIs to keep portability and avoid vendor lock-in.

Before I design, I ask

To help structure the projects I work on, I always start by writing down the prerequisites. Three questions guide nearly everything that follows in the design:

  1. 01Which cloud providers are already in use — AWS, Azure, GCP, OCI — or which ones will be part of the project?
  2. 02What's the main goal behind adopting multi-cloud: cost reduction, high availability, or redundancy?
  3. 03Is there a real need for integration with the physical environment, on-premises?

With those three answers in hand, I start research, put together the first sketches, and build decision matrices to visualize the main scenarios — and the main problems — that might come up during implementation. It's an iterative process: the first sketch almost never survives intact to the final version, and that's fine.

The sketch never survives first contact with reality. And that's fine — that's exactly what it's for.

To go deeper

A few topics I'd recommend researching further, beyond what fits in one post:

I hope this post helped, at least a little, in understanding the process and steps that have always worked for me and my team. Next week, I'll follow this same line of thinking toward automation — my first steps with Infrastructure as Code.

Next post: my first steps with Infrastructure as Code.

See all posts