On-Prem AI Hardware

Start at the workload you need. Expand as adoption grows.

PHOENIX hardware is sized around the models, number of users, agent concurrency, coding workloads, RAG volume and plant integrations required. The configurations below are practical deployment classes rather than fixed appliance specifications.

AI Starter Node

Pilot / Department Deployment

1 enterprise-class GPU, typically 48–96 GB GPU memory
128–256 GB ECC system memory
4–8 TB fast NVMe local storage
10 GbE network connectivity
Runs LabSource, PHOENIX CIP, local RAG and selected local models
Best for initial engineering teams, pilots and focused plant applications

Best fit: entry deployment, a small engineering group, evaluation, private RAG, coding and initial agents.

Production AI Server

Recommended Plant Deployment

2–4 enterprise GPUs, typically 48–96 GB GPU memory each
256–512 GB ECC system memory
8–16 TB high-speed NVMe storage
10/25 GbE networking
Redundant power and rack-server configuration
Supports concurrent users, multiple models, agents, analytics and coding workloads

Best fit: production LabSource at a factory, mine, mill or power facility with multiple engineering and operations users.

AI Factory Server

Large Site / Multi-Team Deployment

4–8 high-memory enterprise GPUs
512 GB–2 TB ECC system memory
20–60 TB NVMe / high-performance local data tier
25/100 GbE networking options
High concurrency for multiple local models and specialized agents
Supports larger RAG libraries, coding, analytics and plant-wide AI workflows

Best fit: large industrial sites, central engineering groups and organizations building a private AI factory.

Hybrid / GPU Expansion

Scale Without Replacing the Core

Keep PHOENIX CIP, Vault, plant context and sensitive RAG data on-prem
Add another GPU node or rack as workload grows
Optionally use approved external GPU services for temporary high-compute workloads
Separate model-compute expansion from the plant security boundary
Preserve the local LabSource environment as the controlled system of record

Best fit: variable workloads, temporary model experiments, growth beyond the initial server or multi-site expansion.

Deployment Scope

Final hardware is selected after workload sizing. Customer-provided rack space, electrical power, UPS, network connectivity and suitable cooling are normally required. Facility electrical work, rack modifications and dedicated HVAC/cooling are separately scoped where needed.

Software Included in the Architecture

The server is the engine room — not the complete solution.

LabSource

Local LLMs, RAG, engineering agents, coding, analytics, model evaluation and private AI workflows.

PHOENIX CIP + Vault

Critical Infrastructure Protection, credentials, role controls, network governance, agent/tool permissions and audit.

Plant Connections

Approved access to DCS, PLC, historians, CMMS, ERP, databases, engineering files, Git, APIs and other internal sources.

Infrastructure That Scales With Adoption

The server is the local engine for a growing capacity-intelligence environment.

The hardware provides local compute for multiple LLMs, RAG, coding, reporting and agents — avoiding ongoing external per-token inference charges when local models are used. The business value comes from combining that compute with PHOENIX asset evidence, maintenance integration, production context, domain expertise and governed access to internal information.

Hardware Options