NVIDIA NIM Inference Microservice Appliance (Project Digits)

The NVIDIA Project Digits (NIM Inference Microservice Appliance) is a palm-sized desktop AI supercomputer built around the Grace Blackwell GB10 Superchip, combining a Blackwell GPU with a Grace CPU on a single unified memory architecture. The system delivers up to 1 petaflop of FP8 compute performance, enabling enterprises to run large language models with up to 200 billion parameters locally without requiring rack-mounted data center infrastructure. The GB10 Superchip integrates a fifth-generation NVLink interconnect, allowing two units to be paired via NVLink-C2C to double memory capacity to 1TB and extend model support up to 405-billion-parameter models, making it uniquely scalable at the desktop form factor.

Project Digits runs the NVIDIA DGX OS, a Linux-based operating system pre-configured with the full NVIDIA AI software stack including CUDA, cuDNN, TensorRT, and the complete NVIDIA NIM microservices catalog. This software-defined approach allows enterprises to deploy production-grade inferencing endpoints directly on the appliance without cloud dependency, making it well-suited for regulated industries, classified environments, and latency-sensitive edge AI workloads. The unified 128GB LPDDR5X memory pool is shared coherently across the GPU and CPU, eliminating PCIe bottlenecks and enabling efficient large-batch LLM inferencing at the desktop.

Designed for enterprises that require sovereign AI infrastructure — where data must remain on-premises or in air-gapped networks — Project Digits bridges the gap between cloud-scale AI capability and the physical and logistical constraints of branch offices, clinical environments, defense installations, and secure research facilities. Its compact form factor (roughly the size of a Mac mini) operates on standard AC power, requires no specialized cooling, and connects via standard networking interfaces, dramatically lowering the barrier to deploying frontier AI models outside of the hyperscaler cloud.

Ideal for

• Air-gapped government and defense LLM inferencing where data sovereignty and network isolation are mandatory compliance requirements
• On-premises clinical decision support using large medical language models in hospital networks where patient data cannot leave the facility
• Low-latency real-time financial analytics and risk modeling using locally hosted 70B–200B parameter models without cloud round-trip latency
• Secure enterprise copilot and RAG (Retrieval-Augmented Generation) deployments for legal, pharmaceutical, and IP-sensitive R&D environments
• Edge AI inference at manufacturing or industrial sites requiring autonomous LLM reasoning without reliable WAN connectivity
• University and research institution AI supercomputing for training and fine-tuning medium-scale models without purchasing rack infrastructure

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.