GPU orchestration for the mid-market rack.
CALSTIQ delivers bare-metal efficiency for fleets from a quarter-rack to a few racks of H100s — the operators the hyperscaler platforms quietly ignore. No abstraction tax, no platform team required.
Infrastructure-first capabilities.
Every layer designed for small-to-mid scale clusters where every cycle, watt, and dollar is accounted for.
Real-time fleet observability
Per-core utilization, thermal envelope, power draw, and NVLink topology streamed at sub-frame intervals.
Fractional GPU slicing
Schedule multiple tenants on a single H100 with MIG-aware isolation, fair-share queueing, and zero hypervisor tax.
Direct-to-metal RDMA
Optimized drivers for InfiniBand and RoCE v2 tuned for small-cluster topologies — no fabric controller required.
Tenant-aware billing
Metered usage by slice, project, and namespace. Export to your billing stack or run the built-in invoice engine.
PXE-to-production in minutes
Bare-metal provisioning, image management, and driver pinning. No Kubernetes operator engineering required.
Air-gap and sovereign ready
Runs fully offline. SSO, SCIM, audit logs, and signed control-plane updates suitable for regulated workloads.
Built for operators the hyperscaler platforms overlook.
CALSTIQ is the orchestration layer for teams with real hardware and finite footprints — not 100-node clusters and a six-person platform team.
Regional clouds & neoclouds
Stand up a billable GPU product on a quarter-rack of H100s without rebuilding a hyperscaler control plane.
Enterprise on-prem AI
Give internal ML teams self-service access to the GPUs sitting in your datacenter — with quotas, projects, and chargeback.
AI labs & research groups
Share scarce H100/A100 capacity across teams with fair-share scheduling, reservations, and live preemption.
Colocation operators
Offer managed GPU-as-a-Service to colo tenants — multi-tenant isolation, metering, and a customer-facing portal included.
Right-sized for your rack.
Hyperscaler orchestrators like Rafay are engineered for fleets that don't look like yours. CALSTIQ is the inverse.
| Metric | CALSTIQ | Hyperscaler orchestrators |
|---|---|---|
| Ideal Fleet Size | 0.25 – 10 racks | 100+ nodes, multi-region |
| Control-plane Overhead | ~1.2% native | 8 – 15% abstraction tax |
| Setup Time | Hours, PXE bootstrap | Weeks of platform engineering |
| Cost Structure | Fixed per rack, per year | Consumption-weighted, tiered |
| Air-gap Deployment | First-class | Limited or unsupported |
| Tenant Billing | Built-in metering | Bring-your-own integration |
Ready to operationalize the hardware you already own?
We'll walk through your fleet, your workloads, and what a CALSTIQ deployment looks like — usually inside a week.