Lab Environments Part 6: Cloud-Hosted Labs, and Picking One

Part 5 covered the container-native end of this landscape. This closing part covers the last category, renting the compute rather than owning it, then pulls all five categories from this series into one comparison and a plain framework for picking.

Cloud-hosted labs.

Two distinct things fall under this heading. The first is renting general-purpose cloud compute sized to run any of the previous five categories as a nested workload: a cloud VM running Proxmox or ESXi nested, or EVE-NG or CML hosted in the cloud, which is exactly the service a company like CloudMyLab sells, hosted EVE-NG and CML instances rather than hosted compute you build on yourself. The second is vendor-hosted sandboxes built for a specific purpose, Cisco DevNet’s sandboxes being the clearest example, temporary access to a pre-built environment for testing an API or a feature without standing up anything at all.

The tradeoffs are straightforward. No hardware to buy, size, or maintain, and pay-as-you-go or hourly billing suits a short, occasional burst of study far better than an always-on home lab that sits idle most of the week. Against that: nested virtualization inside a cloud VM carries a real performance cost worth checking before committing to it for anything CPU-heavy, and some vendor image licenses explicitly restrict cloud reselling or redistribution, which is worth reading before assuming a vendor image that works fine on a home Proxmox host is equally fine rehosted on a cloud instance.

The five categories, side by side.

Hypervisor foundations, Proxmox, ESXi, VMware Workstation/Fusion, give you the compute layer everything else in this list can run as a guest on top of. Free and open (Proxmox), free with a messy recent licensing history (ESXi), or free since 2025 with an account requirement (Workstation/Fusion).

Network emulation platforms, EVE-NG, GNS3, pnetlab, give you real vendor images wired into a GUI topology. EVE-NG’s free tier tightened sharply in 2026; GNS3 stayed free and open throughout; pnetlab trades legal ambiguity for the most permissive node limits of the three.

Vendor-official all-in-one, Cisco Modeling Labs, Packet Tracer, trade single-vendor coverage for a solved image problem. CML now has a genuinely permanent free tier; Packet Tracer isn’t trying to be an emulator at all, and that’s a feature for teaching and exam prep, not a shortcoming.

Container-native, containerlab, netlab, trade the GUI canvas for topology-as-code that fits a CI pipeline, at the cost of not every vendor image existing as a container the way it exists as a VM appliance.

Cloud-hosted trades ownership for elasticity, at the cost of nested-virtualization performance and licensing terms that need checking per vendor image.

A plain decision framework.

Needs real vendor CLI behavior for an exam or a client config: Cisco Modeling Labs if it’s Cisco-specific, EVE-NG Professional if it needs to cover other vendors too, pnetlab if the 2026 free-tier tightening matters more than the unresolved legal question around the fork.

Teaching or exam prep where zero licensing friction matters more than exact CLI behavior: Packet Tracer.

Testing your own automation or tooling against representative device behavior, not the device itself: containerlab or netlab.

Foundational compute for any of the above, and hardware is available: Proxmox by default, ESXi specifically for VMware-admin skills.

Short, occasional bursts of study without owning hardware at all: a cloud-hosted instance or a vendor sandbox.

Where this series leaves off.

This was deliberately an overview, not a deep dive into any single tool, and it doesn’t solve the one problem named back in Part 1: none of these six posts hands over a single vendor image. That’s still on whoever’s building the lab, a support contract, a CCO account, or an NFR agreement, no platform choice changes that.

The next full deep dive on this site picks up Proxmox specifically, building on the same pve host, vmbr-bridge pattern, and per-lab network isolation already used for the FortiManager lab, the Sn1per re-verification work, and the BloodHound graph testing, real infrastructure this site already runs, rather than a platform chosen for how it looks in a comparison table.