[Back to blog](https://zsoftly.ca/blog)
 [Engineering](https://zsoftly.ca/blog/category/engineering)

# What is Metal-as-a-Service (MaaS)?

Metal-as-a-Service turns a rack of physical servers into an API-driven pool. Here is when bare-metal provisioning matters and how ZCP delivers it.

ZSoftly Team

May 16, 2026

2 min read

Share

Metal-as-a-Service (MaaS) is the discipline of provisioning physical servers the same way cloud providers provision virtual machines: through an API, in minutes, with the OS installed and the network configured. No hypervisor in the path. No noisy neighbors. Bare hardware.

Two reference implementations define the category:

- [Canonical MAAS](https://maas.io/), open-source, uses PXE, IPMI, and Redfish to find and image machines on the network. Canonical positions MAAS as the “Day-0” provisioning layer underneath OpenStack and Kubernetes deployments.
- [OpenStack Ironic](https://wiki.openstack.org/wiki/Ironic), originally a driver inside OpenStack Nova, now a standalone project managing the full lifecycle of physical nodes: enrollment, provisioning, maintenance, and decommissioning.

Both expose bare-metal hosts through a cloud-style API. The hypervisor layer is removed.

## When MaaS makes sense

Most workloads run fine on virtual machines. MaaS earns its place in four specific situations:

1. High-performance and HPC workloads. Virtualization adds latency and CPU overhead. For tightly coupled MPI jobs, low-latency trading, or large in-memory databases, bare metal removes the variance.
2. GPUs. Training and inference clusters benefit from direct PCIe access to accelerators. Sharing a GPU host across tenants introduces driver, isolation, and scheduling complexity. Bare metal sidesteps this.
3. License-bound software. Some commercial software (Oracle, certain SAP modules, legacy ISV products) is priced per physical socket or per core on the underlying host. Running on a virtualized public cloud either breaks the license or makes it prohibitively expensive.
4. Dedicated tenancy and compliance. When a regulator, customer contract, or threat model requires physical isolation, MaaS gives you a host no other tenant touches. This also removes a class of side-channel attacks possible in shared-hypervisor environments.

A 2024 [IEEE-published analysis](https://ieeexplore.ieee.org/document/10823401) of cloud service models notes bare-metal options are an increasingly common addition to IaaS catalogs because of these workload shapes.

## How ZCP delivers Bare Metal today

ZCP offers dedicated bare metal across data center facilities in Canada, the United States, the United Kingdom, Europe, and Australia. We operate and manage servers in over 28 locations globally, including carrier-neutral colocation in Montreal with IX peering and BGP. GPU-optimized configurations are available for bare metal and private cloud deployments. GPU is not available on ZCP public cloud infrastructure at this time. Pricing is custom, sized to your workload and target location. Contact us and our team will scope, configure, and provision your server.

What you get:

- Dedicated physical server with no shared tenancy
- Hardware configured to your requirements: CPU, RAM, GPU, NVMe
- OS provisioning and hardening included
- Ongoing monitoring, capacity management, and hardware replacement

This is not click-to-provision through the portal. You contact us, we build it and hand it over. Self-serve bare metal provisioning is on the roadmap.

For teams wanting the full managed stack on hardware they control (CloudStack, Ceph, Kubernetes), the [Private Cloud Build-Out](https://zsoftly.ca/services/private-cloud) is the right fit. It starts at $4,500 setup plus $1,200/mo operations.

[Contact us to discuss your bare metal requirements.](https://zsoftly.ca/contact)

## Related articles

Engineering

November 4, 2026

### [How Zenphry Built a Reusable Office Connection for Its Remote Team](https://zsoftly.ca/blog/remote-teams-office-network)

How Zenphry built an approved remote-access foundation with a Headscale VPN control plane on ZCP, a Dallas exit node, Tailscale clients, and managed routers.

Engineering

November 4, 2026

### [Why I Love Orca for ZSoftly’s Development Workstations](https://zsoftly.ca/blog/shared-development-workstation-orca-zcp)

How ZSoftly is designing dedicated developer workstations on ZCP, with Orca across interactive work and a separate CI/CD path for execution.

Engineering

October 12, 2026

### [A Backup Is Not a Disaster Recovery Plan](https://zsoftly.ca/blog/backup-is-not-disaster-recovery)

A successful backup job does not prove you can recover. Separation, verification, restore testing, and operational ownership create real disaster recovery readiness.
