The Box
Local AI. No rack required.
A compact, self-contained AI computer for running models and processing large datasets locally. First production run now in preparation.
AI compute you can carry
Running serious AI locally usually means specifying and maintaining a workstation: GPU compatibility, power, cooling, drivers, cabling and everything that eventually breaks. The Box packages that infrastructure into a complete computer designed to work out of the box.
Our prototype was originally developed for remote geological fieldwork, but user testing identified a broader need: running private AI models, processing hyperspectral imagery, cleaning large archives and working with scientific datasets without sending them to external cloud services.
Built for work that shouldn't depend on the cloud
Private AI
Run local models against sensitive company data without routinely sending that information to third-party inference APIs.
Offline AI
Take serious compute into environments where connectivity is slow, intermittent or nonexistent.
Scientific Computing
Process large imagery, hyperspectral datasets, geological data and other computationally intensive scientific workloads locally.
Local Models
Run open-weight and specialist models close to the data they work with, including knowledge too private, specialised or recent to exist in general-purpose model training datasets.
Specifications
- GPU
- NVIDIA RTX PRO 4000 Blackwell SFF
- VRAM
- 24 GB GDDR7 · 432 GB/s
- GPU Compute
- 24 TFLOPS FP32
- CPU
- AMD Ryzen 7 5800X · 8 cores / 16 threads · up to 4.7 GHz
- Memory
- 32 GB DDR4*
- Storage
- 1 TB NVMe SSD*
- Power
- 250 W system design target
- Enclosure
- Pelican 1300 · 233 × 178 × 156 mm
- Cooling
- Custom liquid cooling with external active cooling
- Operating mode
- Complete standalone computer. Connect to monitor and keyboard/mouse or compatible field interface.
* Configurable. Early registrations will determine the configurations offered at launch.
Built, not rendered
The Box is working hardware. We designed the enclosure around a compact GPU/CPU system, built and tested the first prototype, and are now preparing the design for small-scale production.
One enclosure. Different workloads.
The first production run is being designed around configurable compute, memory and storage options. Tell us what you need to run — we'll use early registrations to determine the configurations offered at launch.
Join the first production run
We're currently finalising manufacturing specifications and pricing for the first batch of The Box. Join the waitlist to receive production specifications, pricing and expected availability when pre-orders open.