Ivertiq Node
Cost-competitive on-prem for pilots, departmental workloads, and moderate concurrency.
Example configurations: GB10-class and RTX 6000/RTX PRO 6000-class platforms (selected during scoping).
Deployment range
One Full-Stack AI Harness. A range of deployments.
Same vertically optimized stack — sized to economics, concurrency, and operating model — from highly cost-competitive on-prem nodes, to high-throughput on-prem stations for concurrent sessions, to highly scalable multi-tenant siloed private cloud.
One Harness · deployment range
Key diagram
Product tiers
How to choose across the range
Same Harness in every tier. Product names describe capacity and operating model; hardware classes are configurations under those tiers.
| Deployment Appliance | Hosted Model Engine | Target Deployment Description |
|---|---|---|
| Ivertiq Node (GB10, Dual GB10) | Ivertiq Swift (or quantized Prime) |
Edge / single-workstation appliance for plant AP offices & SMBs that require low concurrency. |
| Ivertiq Station (RTX PRO 5000, RTX PRO 6000) | Ivertiq Prime (and multi-stream Swift) |
High-throughput on-premise rack/tower for mid-size enterprises with multiple concurrent users. |
| Ivertiq Private Cloud | Ivertiq Ultra (Full fleet orchestration) |
Dedicated datacenter cluster / private VPC for multi-subsidiary enterprise fleets. |
These are points on a spectrum, not a closed menu — additional on-prem configurations can sit between Node and Station as workloads evolve.
Sovereignty in deployment
Adapt privately. Serve on-prem.
High-throughput post-training happens on private cloud where GPU is abundant. Production serving is the appliance in the customer’s room — GB10-class unified memory for agent traces, or RTX 6000 / RTX PRO 6000-class workstations where the IT standard is a PCIe GPU. Two workstation boards are two memory domains, not one pooled fabric; the Harness is designed for that, rather than assuming NVLink.
Ivertiq Full-Stack Platform
The deployment is the physical embodiment of the stack
Customers buy a working vertical AI system — localized models, governed agents, and workflow apps — not a GPU server plus a model download. Product tiers describe capacity and operating model; hardware classes are configurations under those tiers.
Vertical Apps
Industry workflows, agents, and solutions
AI Harness Control
Orchestration, agents, policies, guardrails, HITL & audit
Ivertiq Core Foundation
Local & vertical LMs · knowledge engine · inference optimization
What changes across the range
Capacity
Concurrency & context
From departmental pilots to heavier orchestration and more concurrent users — then to siloed cloud scale.
Platform class
On-prem configurations
Node and Station are offered on multiple platform classes — including GB10-class and RTX 6000/RTX PRO 6000-class — sized for cost, concurrency, and workload during design-partner scoping.
Ops model
Owned vs managed
Customer-owned on-prem appliances or multi-tenant private cloud with isolation and auditability.
Ivertiq Station
High-throughput on-prem for heavier agent orchestration, larger context, and more concurrent sessions.
Example configurations: higher-capacity GB10-class and RTX 6000/RTX PRO 6000-class platforms (selected during scoping).
Ivertiq Private Cloud
Highly scalable multi-tenant siloed private cloud — SaaS-like ops with tenant isolation, controlled serving, and auditability, without public LLM API dependency for core corporate data.
These are points on a spectrum, not a closed menu. The same Harness can be sized across additional on-prem configurations as workloads and hardware platforms evolve.