IRON / PERSONAL COMPUTE

Local AI workstation.

Language. Vision. Voice.
Images. Video.

Explore the hardware

IN DEVELOPMENT

SPECIFICATIONS

01 / HARDWARE

Three configurations.

Planned specifications

Explore planned configurations

IRON ONE

32GB

GPU memorySingle GPU

GPU
GeForce RTX 5090
CPU
Ryzen 9 9950X
MEMORY
128 GB
STORAGE
4 TB NVMe

IRON PRO

96GB

GPU memorySingle GPU

GPU
RTX PRO 6000 Blackwell
CPU
Threadripper PRO 9955WX
MEMORY
256 GB ECC
STORAGE
8 TB NVMe

IRON ULTRA

4 × 96GB

GPU memory384 GB aggregate, not unified

GPU
4 × RTX PRO 6000 Max-Q
CPU
Threadripper PRO 9975WX
MEMORY
512 GB
STORAGE
16 TB NVMe

Configuration targets. Enclosure fit, thermals and performance have not yet been validated.

02 / SOFTWARE

IN DEVELOPMENT

Iron Stack.

Models, a scheduler and a local API.
Managed through Iron Console.

Explore planned workloads
PLANNED WORKFLOWLOCAL INFERENCE
INPUT

Prompt + documents

LANGUAGE MODEL

Qwen3.8-27B

CANDIDATE MODEL
OUTPUT

Text, code, JSON

INPUT

Image + question

MULTIMODAL MODEL

Qwen3.8-27B

CANDIDATE MODEL
OUTPUT

Visual analysis

INPUT

Text or audio

SPEECH MODELS

Qwen3-TTS / ASR

CANDIDATE MODEL
OUTPUT

Speech or transcript

INPUT

Prompt + reference

IMAGE MODEL

FLUX.2 [klein] 4B

CANDIDATE MODEL
OUTPUT

Generated or edited image

INPUT

Prompt + first frame

VIDEO MODEL

LTX-2.5

CANDIDATE MODEL
OUTPUT

Video + audio

Model support is under evaluation. Compatibility and commercial use depend on the model, runtime and license.

In development.

Pricing and release dates
have not been announced.

What is being built?

A family of local AI workstations, plus Iron Stack and Iron Console. Hardware configurations, enclosure design and software are in development.

Will it work offline?

Local inference is the intended default. Model downloads and updates need a connection. Optional external services would require an explicit choice.

Are there performance benchmarks?

Not yet. The configurations above are engineering targets. Measured performance will depend on the final hardware, model and runtime.