When to choose Bare Metal Compute for AI or licensing-restricted workloads

Reason 1: AI and HPC performance

Worldstream describes Bare Metal Compute as physical servers made available directly, without a virtualisation layer: the workload runs on dedicated hardware, giving maximum performance, predictability and control. For most applications, the overhead a hypervisor adds is small enough to ignore. For a specific set of workloads it isn't:

• High-performance computing (HPC), where every CPU cycle and every millisecond of I/O latency compounds across long-running jobs.
• AI and ML workloads, particularly training and inference jobs that need direct, predictable access to compute and memory bandwidth.
• Big data analytics and large databases, where consistent, high-throughput I/O matters more than the flexibility a hypervisor offers.
• Latency-sensitive applications, where even small, variable delays introduced by a virtualisation layer are unacceptable.

On Bare Metal Compute, all CPU, memory and storage resources are fully available to your workload, with no other tenant sharing the machine and no hypervisor scheduling cycles on your behalf. If you do want a virtualisation layer on top, you choose and configure it yourself rather than inheriting Worldstream's.

Reason 2: software licensing that restricts virtualisation

The second reason is less about raw performance and more about what your software vendor will actually let you do. Some enterprise database and legacy software licences are written around physical hardware, not virtual machines. Two patterns come up most:

• Per-physical-core licensing. The licence fee is calculated from the number of physical CPU cores available to the software, not virtual cores. On a shared, virtualised host, working out how many physical cores your VM is entitled to touch can get complicated, and vendors size their licence audits around that ambiguity. On a bare-metal server, the physical core count is exactly what's in the machine: nothing to reconcile.
• Licences that prohibit virtualisation entirely. Some enterprise database platforms and older, legacy line-of-business software are licensed on the explicit condition that they run on physical hardware, full stop. Running them on a VM, even one with dedicated resources, can breach the licence agreement regardless of performance.

What "per-physical-core licensing" means in practice: instead of paying per virtual CPU assigned to your instance, the vendor counts the physical CPU cores in the server the software runs on and charges against that number. It's common with some enterprise database platforms. The practical effect is that your licensing cost, and your compliance position, depends on the physical core count of the exact machine, not on what a hypervisor presents to your VM. Running that software on genuinely dedicated hardware removes the ambiguity: there's one server, with a known number of physical cores, and nothing else contending for them.

If neither reason applies to you

If your workload virtualises fine and your software licensing doesn't care about physical cores, you likely don't need Bare Metal Compute. A Flexible VPS or Dedicated Server is usually the simpler fit. See Bare Metal Compute vs. Dedicated Servers for how the two physical-hardware products compare.