Cloud Networking Fundamentals for Network Engineers: AWS, Azure and GCP Explained
Conxiea
28 August 2026
→ See How Conxiea's AI InfraOps Platform Automates Network and Cloud Operations
The good news for network engineers moving into the cloud: the fundamentals aren't new. Subnets, routing, gateways, and access control all exist in AWS, Azure, and GCP in some form. The bad news: the implementation is different enough from on-prem networking that assuming direct transfer will cause real mistakes, misconfigured route tables, security groups that don't do what you expect, and connectivity that looks right on paper but doesn't work.
This guide maps the on-prem concepts you already know onto their cloud equivalents across the three major providers, and flags the places where the differences actually matter.
This post is part of our complete roadmap for network engineers moving into network automation and cloud automation. If you're mapping out the full path, start there.
Person reviewing cloud infrastructure and network status on a tablet in a data centre corridor
The Core Concept: Virtual Networks
Every major cloud provider gives you an isolated virtual network to build inside: a VPC (Virtual Private Cloud) in AWS and GCP, a VNet (Virtual Network) in Azure. Conceptually, it's the same thing you already understand as a layer 3 network boundary, a private address space that you define, subnet, and control routing and access within.
The difference starts with scope. An on-prem network is bounded by physical infrastructure you can walk up to. A VPC or VNet is a logical construct defined entirely in software, which means it can be created, destroyed, and reconfigured in seconds rather than requiring physical changes. That's the entire premise cloud automation is built on, but it also means the discipline of defining things correctly in code matters more, because there's no physical layer forcing consistency.
Subnets and Addressing
Subnetting inside a VPC or VNet works the way you'd expect, CIDR blocks, address ranges, and the usual planning around growth and segmentation. The key difference is how subnets map to availability: cloud providers organise infrastructure into regions and availability zones, and best practice is to spread subnets across multiple availability zones for resilience, something with a closer on-prem analogy to distributing infrastructure across separate racks or sites rather than a single subnet design decision.
Each provider also reserves a handful of addresses per subnet for internal use, AWS reserves five, for example, which trips up engineers doing capacity planning based on pure CIDR math without accounting for it.
Routing: Similar Concept, Different Mechanics
Route tables exist in all three clouds and function similarly to what you already know: they determine where traffic goes based on destination. The mechanics differ though. In AWS, route tables are explicitly associated with subnets, and there's no dynamic routing protocol running by default, routes are static entries you define. Azure and GCP work similarly, with their own naming and association models.
This is a meaningful shift for engineers used to BGP and OSPF doing the routing work. Cloud networking within a single VPC or VNet is largely static routing by design. Dynamic routing reappears when you get into hybrid connectivity and multi-VPC architectures, which is where BGP knowledge becomes directly useful again.
Gateways and Internet Access
Getting traffic in and out of a virtual network requires an explicit gateway resource, there's no default path to the internet the way a physical network might have a default gateway configured on a router. AWS uses an Internet Gateway for outbound/inbound internet access and a NAT Gateway for private subnets that need outbound-only access. Azure and GCP have equivalent constructs with different names.
The pattern to internalise: public subnets route to an internet gateway, private subnets route to a NAT gateway or equivalent for outbound-only access, and nothing gets a path to the internet unless a route table explicitly sends it there. It's a more deliberate, explicit model than many on-prem networks, where a default route often exists almost as an afterthought.
Hybrid Connectivity: Where On-Prem Knowledge Comes Back
This is where cloud networking gets genuinely interesting for network engineers, and where existing knowledge becomes directly applicable again. Connecting a cloud environment to on-prem infrastructure, or connecting multiple VPCs and VNets together, brings BGP, VPN, and dedicated circuit concepts straight back into play.
| Provider | Dedicated connectivity | Site-to-site VPN | Multi-VPC/VNet |
|---|---|---|---|
| AWS | Direct Connect | Site-to-Site VPN | Transit Gateway |
| Azure | ExpressRoute | Site-to-Site VPN | Virtual WAN |
| GCP | Cloud Interconnect | Cloud VPN | Network Connectivity Center |
Transit Gateway, Virtual WAN, and Network Connectivity Center all solve the same underlying problem: routing traffic between many VPCs or VNets without a full mesh of point-to-point connections between all of them, conceptually similar to a hub-and-spoke WAN design many network engineers have already built on-prem. BGP is commonly used across these dedicated connections and VPN tunnels, which is exactly the protocol knowledge a traditional network engineer already brings.
Security Groups and Network ACLs
Access control in the cloud splits into two layers that don't map cleanly onto a single on-prem equivalent. Security groups (AWS, and similar constructs in Azure and GCP) are stateful, attached to individual resources rather than a network boundary, and only need rules in one direction since return traffic is automatically allowed. Network ACLs sit at the subnet level, are stateless, and require explicit rules in both directions.
The stateful, resource-attached model of security groups is the part that catches out engineers used to firewall rules applied at a network boundary. It's a more granular, closer-to-the-workload model, which is powerful but requires a different mental model for auditing what's actually allowed to talk to what.
Multi-Cloud: Same Concepts, Different Vocabulary
Once you're solid on one provider, the other two are largely a vocabulary exercise rather than a new set of concepts. A VPC is a VNet is a VPC. Route tables, gateways, and security boundaries exist everywhere, named differently and configured through different interfaces and APIs, but solving the same underlying problems.
The practical advice: get genuinely fluent in one provider first, ideally whichever one your target employer or current organisation actually runs, before spreading across all three. Depth in one beats shallow familiarity with three when it comes to actually being useful in an interview or on the job.
Where This Leads: Automating What You've Just Learned
Understanding cloud networking concepts is stage three of the roadmap. Stage four is defining and deploying that infrastructure as code instead of clicking through a console, which is where Terraform comes in. Our guide to Terraform for network engineers picks up directly from here.
Want to see cloud networking and automation running together in a real production environment?
At Conxiea, we design and automate hybrid and multi-cloud network infrastructure for clients as part of our AI InfraOps platform.
No obligations. Whether you're a business planning a cloud migration or an engineer building these skills, we're happy to talk it through.
Related Reading
- Network Automation and Cloud Automation: The Complete Career Roadmap for Network Engineers
- Terraform for Network Engineers: Automating Cloud and Network Infrastructure with IaC
- Python for Network Engineers: The Practical Skills You Need for Network Automation
- NetDevOps Explained: How Network Engineers Become Network Automation Engineers
- How to Automate Your Network: A Practical Guide for IT Teams