Private inference appliances

AI workstations for firms that keep files inside.

Compact North-class towers with professional NVIDIA GPUs, current Intel and AMD CPUs, ECC options, and enough local VRAM for serious private legal, accounting, tax, and advisory workloads.

192GB VRAM per Pro build
20% margin included
45L North-class target
Configurator

Four base variants. Real knobs.

Start with Intel or AMD, then choose RTX 6000 Ada or RTX PRO 6000. Everything else tunes the build without turning the range into a maze.

Chassis
Fractal North Chalk White with mesh side panel
Chalk White / Mesh 360
Finish
Side panel
GPU count
Machines
System memory
NVMe scratch
Commissioning
Networking
Display output
Machine spec

The ports are part of the appliance.

Each build ships as a firm-ready workstation target, not a loose parts list. Exact rear I/O varies by board, but these are the standard minimums carried into the estimate.

Network 2x 10GbE RJ45 NIC

Dedicated add-in NIC, onboard 2.5GbE fallback, Wi-Fi 7 / Bluetooth where the selected board supports it.

Video 8x DisplayPort 2.1

HDMI is supplied through active DisplayPort-to-HDMI 2.1 adapters when selected.

USB / front I/O USB-C + 2x USB-A

North front panel plus rear USB-C/USB-A from the selected workstation motherboard.

Power / cooling 1600W class PSU

Platinum-rated power target, front-to-back airflow, blower GPU spacing, and acoustic fan curves.

Storage layout 2TB OS + 8TB scratch

Separate OS drive, configurable NVMe scratch, encrypted project volume, and room for internal backup media.

Software handoff Linux or Windows

NVIDIA drivers, CUDA, container runtime, Ollama/vLLM-ready baseline, and remote admin enabled on request.

North compact build

192GB of local VRAM, dressed for the client floor.

The signature North build targets two RTX PRO 6000-class blower cards in a compact tower with real oak, controlled airflow, and office-grade acoustics. It is built for private inference near sensitive matter data, ledgers, tax workpapers, deposition sets, and firm knowledge bases.

Quiet hardware for loud workloads.

01

Private by default

Keep privileged text, client files, and internal prompts on local hardware.

02

Professional GPU memory

Choose RTX 6000 Ada for 48GB per card or RTX PRO 6000 for 96GB per card.

03

ECC where it matters

Run validated ECC memory when long jobs and firm records deserve the safer path.

Pricing basis

Current market inputs, not fantasy MSRP.

The calculator uses a public July 2026 memory index, spot ECC listings, NVIDIA workstation GPU pricing, and a flat 20% build margin. Component markets are moving fast, so this is a sales estimate rather than a checkout price.

Build brief copied