Product Highlight — VM Pools: Large-Scale Sovereign Federation

Every AI initiative runs on GPUs, and GPUs have become the scarcest resource in enterprise computing. On-demand GPU capacity at the major cloud providers is often unavailable when a project needs it, and a training run that cannot claim a GPU does not start. In federated projects, progress depends on how many times a team completes the loop of train, inference, review, adjust, and train again, so every stalled start costs a turn of that loop.

Our customers felt that shortage early, and we moved quickly. In June 2026 we released VM Pooling in the Rhino Federated Computing Platform (FCP), which lets a Rhino Client reserve compute capacity up front so high-demand workloads get guaranteed availability and utilization you can plan around. Since then we have expanded VM Pooling to give customers more control over where their compute runs and what it costs.

• One Code Object, multiple pools: A Code Object can now target multiple pre-allocated VM pools instead of just a single one. A federated node can scale as a workload demands across pools with different compute specs, and sites can keep separate regional pools so one Code Object serves federated training and inference across regions while data stays sovereign.
• Compute auditability: Run details now show which VM pools a run requested alongside the resources it actually claimed, so usage stays visible, both when you are debugging a slow start and when you are reading a compute bill.

Why this matters for our customers

While compute scheduling reads like an infrastructure detail, it impacts how the network as a whole operates and manages the research calendar.

• Critical initiatives keep moving: Reserved GPU capacity means a run starts when the team is ready, not when a cloud region frees up hardware. 
• Scale without moving data: Region-specific pools let compute grow wherever the data sits, so a collaboration spanning US and EU sites adds capacity without data leaving its jurisdiction.
• Privacy built into every run: Every run executes on a VM that is destroyed when the run finishes, wiping all working data. An organization running many overlapping projects on shared pools never carries one project's data into another project's run.
• Predictable service and cost across regions: Reserved pools give each region a known baseline cost and consistent performance, backed by real utilization data for forward planning.

Those benefits play out for all of our customers across industries and use cases. In clinical development, an analysis that gates a go/no-go decision runs on the study's schedule. Model training on medical imaging across sites is inherently GPU-bound and bursty; project-level pools now absorb the bursts instead of rationing them. In payer and population health analytics, recurring cross-organization runs feed reporting cycles where one missed window delays entire cycles.

Full detail for this release is in the Rhino FCP release notes. If you're sizing VM pools for a new collaboration and want a second opinion on the configuration, your Rhino representative can help.