Products & Tools

vCluster Platform v5.0 brings self-service provisioning to bare-metal AI hardware

vCluster Platform v5.0, released November 19 in limited availability, lets AI cloud customers provision bare-metal GPU servers and launch Kubernetes, Slurm or inference clusters without provider staff handling each request.

VCluster Adds Self-Service Private Clouds for Dedicated AI Hardware
VCluster Adds Self-Service Private Clouds for Dedicated AI Hardwareblickpixel / pixabay
By Marcus Bennett5 min read

Why it matters

  • vCluster Labs released Platform v5.0 in limited availability on November 19.
  • Customers can self-provision bare-metal GPU servers and launch Kubernetes, Slurm or inference clusters from approved templates.
  • The vMetal component maintains a single inventory of physical servers with tenant boundaries and quotas.
  • The vendor recommends staying on long-term support releases Platform v4.12 and vCluster v0.37 for current deployments.

vCluster Labs released vCluster Platform v5.0 in limited availability on November 19, giving AI cloud providers a way to offer self-service environments on dedicated physical GPU servers. Customers can provision machines from their allocated capacity, select an operating system image and launch Kubernetes, Slurm or inference clusters without asking the provider's operations team to handle each request.

The release targets a practical bottleneck in the AI infrastructure market. A customer may rent an entire block of GPU servers yet still depend on the provider's staff to prepare machines for every new project. Platform v5.0 moves those provisioning decisions to the customer while leaving hardware ownership and capacity allocation with the provider.

That division of labor matters commercially. Providers selling bare-metal capacity (physical servers rather than virtual instances that share a host with unrelated customers) have long faced a trade-off: dedicated hardware offers predictable performance and exclusive accelerator access, but each new workload has traditionally required manual setup of the operating system, access configuration and cluster preparation. vCluster's answer is to keep the hardware dedicated while making the provisioning workflow resemble a public cloud's.

What does Platform v5.0 actually change?

The release brings physical servers into a single inventory maintained by vMetal, the provisioning component beneath the platform. Operators work through a defined sequence:

  • An operator assigns capacity to customers or leaves it available on demand.
  • The operator decides which machine types, operating system images and cluster templates each customer can use.
  • The platform adds an explicit customer boundary, called a tenant, above individual projects.
  • Quotas limit how much capacity a customer can claim.
  • Customers cannot see inventory they have not been granted.

For a customer's platform team, the tenant model carries a concrete benefit: several internal groups can draw from the same contract without each needing a new allocation from the provider. A research group could set up a scheduled Slurm environment (a workload manager commonly used for high-performance computing jobs) while an application team runs Kubernetes for an inference deployment, all under one agreement.

The provider retains control over what is possible. It decides which configurations it supports, and the customer chooses among the approved options. The company says providers can use its console directly or build a branded portal on the same underlying API. That distinction matters to operators selling infrastructure under their own brand: the customer-facing interface can differ without requiring a separate inventory and provisioning system for each service tier.

Why does self-service on bare metal matter now?

The AI cloud market has converged on similar hardware. If providers sell comparable GPU servers, the differentiator shifts to operational delivery. The immediate test, as the release frames it, is operational: how quickly a provider can deliver working infrastructure after allocating hardware, and how often customers need assistance afterward.

The business case rests on staff time. A smaller number of steps handled by a provider's operations team could make each customer easier to onboard, while buyers who prefer a fully managed cluster can purchase a different level of service built on the same inventory. A provider may use one pool of machines for simple server rentals and for more managed products simultaneously.

The release does not establish how much either side would save. The economics depend on how heavily the hardware runs and what level of support customers require after provisioning.

Idle hardware presents a second economic question. Reserved capacity gives buyers predictability, but those machines remain unavailable to other customers even when idle. An on-demand pool increases flexibility, though only when the provider can reliably meet demand spikes. Platform v5.0 lets providers offer both models. It does not remove the operator's need to set reservation terms or plan for shortages.

What does limited availability mean for operators?

Version 5.0 is not generally available. The vendor's documentation recommends that operators retain the existing long-term support releases, Platform v4.12 and vCluster v0.37, for current production deployments.

Operators evaluating the new version face two verification tasks before adoption:

  • Assess compatibility with their existing provisioning and networking tools.
  • Verify that customers can move from a machine allocation to a usable cluster without creating new support bottlenecks.

That second task is the core of the product's promise. Self-service provisioning only delivers value if it reduces, rather than redirects, the support burden. The vendor itself frames the open question: whether self-service reduces support bottlenecks remains something operators must measure.

Who bears the remaining work?

The platform shifts decisions, not responsibility, to defined boundaries. The provider still owns the hardware, still allocates capacity and still determines which machine types, images and cluster templates exist. The customer gains choice within those boundaries — how much control depends entirely on what the allocation includes and which templates the provider approves.

This structure positions vCluster between two existing models. Pure bare-metal rental leaves all software setup to the buyer. Fully managed AI cloud services absorb setup into the price. Platform v5.0 attempts a middle path: dedicated machines, per-customer networking and self-service provisioning, with managed tiers available on the same inventory for buyers who want them.

If the operational measures improve without making dedicated capacity harder to manage, self-service could become a meaningful distinction between AI cloud providers selling otherwise similar hardware. Until general availability, operators will be watching two numbers: deployment speed after allocation, and assistance requests per customer afterward.

Source: unite.ai

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

791 articles

Related articles

  1. NVIDIA Open-Sources Run:ai Under Apache 2.0
  2. Broadcom expands VMware Cloud Foundation with AI and security
  3. AMD Ships Zen 6 EPYC Server Chips Built for AI Agent Workloads
  4. AMD launches 256-core EPYC 9006 CPUs to fight Intel and Nvidia
  5. Nexalus and RENDA Launch Nova8, a Liquid-Cooled AI Server Built for Retrofit

« Previous articleNext article »