Receive HP ZGX Nano as an operational local AI platform? WZ-IT configures Open WebUI, a local model, users, network, hardening, backup, and a support path. Explore the AI Cube
HP ZGX Nano G1n AI Station is HP's compact GB10 platform for local AI. It combines 128 GB of unified memory, DGX OS, and the HP ZGX Toolkit for model testing, inference, and development. Use by business departments requires the developer foundation to be extended with identity, an interface, data handling, and operations.
HP-specific foundation
HP documents configurations from 1 TB up to 4 TB of NVMe storage. The system uses a 20-core Arm processor, integrated Blackwell GPU, 10 GbE, and ConnectX-7. Windows is not supported; DGX OS is the intended operating environment.
Since October 2026, HP has been one of the OEM partners through which NVIDIA sells a GB10 variant with 64 GB of unified memory (NVIDIA blog). This variant is available from 23 October 2026 and targets models up to about 100 billion parameters. Whether a specific ZGX Nano model has 64 GB or 128 GB depends on the SKU; this article refers to the 128 GB version.
For considerably larger models, HP ZGX Fury has been available to order since September 2026. It is based on NVIDIA GB300 with 748 GB of coherent memory and therefore belongs to a different device class than ZGX Nano.
The HP ZGX Toolkit bundles tools such as VS Code, Python, MLflow, and Ollama. These help development teams but do not replace a central user interface or operating model for employees.
From workstation to internal AI service
Production setup includes documented DGX OS, firmware, drivers, and toolkit versions; named administrators and SSH access; a runtime and model selected for the workload; Open WebUI with TLS, users, groups, and model permissions; storage planning; independent backup and restore; monitoring; and a clear separation between platform incidents and vendor hardware cases.
Choose storage for actual use
A 1 TB configuration may be sufficient for a few model artefacts and testing. Several quantisations, multimodal models, or larger knowledge collections need more headroom. Containers, logs, updates, and temporary files also consume storage.
A backup on the same SSD does not protect against device or drive failure. Configuration, chat data, and knowledge spaces belong on an external target.
When HP ZGX Nano fits
ZGX Nano is relevant where HP is the strategic workstation supplier and the ZGX Toolkit adds value for development teams. For business users, the configured chat and knowledge interface matters more than the development toolkit.
The smaller SSD configuration is best suited to a narrowly defined model set. Several models, extended testing, and larger local document collections favour more capacity. As the platform does not support Windows, custom agents, scripts, and drivers should be checked for ARM64 compatibility. Staff browser access is unaffected by that architecture.
AI Cube and ongoing operations
The AI Cube turns the hardware into a prepared platform with Open WebUI, runtime, agreed model, hardening, functional testing, initial setup, and support. Staff then use chat and personal or shared knowledge spaces through the browser.
WZ-IT can take responsibility for updates, monitoring, and support. Greater throughput or model size leads to additional AI Cubes or an AI Cube Custom architecture.
