HP launches the HP ZGX Nano G1n AI Station. It’s a small computer powered by Nvidia’s GB10 Grace Blackwell chip. The mini PC is designed to support local AI workloads.
HP embraces Nvidia’s GB10 Grace Blackwell Superchip with the launch of the HP ZGX Nano G1n AI Station. It’s a small computer that looks a bit like a NUC from the outside. The internals, however, reveal a completely different target audience.
Nvidia Chip
The Nvidia GB10 Grace Blackwell Superchip is an AI chip developed by Nvidia and tailored for PCs. The chip combines a Blackwell GPU with an ARM-based Grace CPU. Both chiplets share 128 GB of LPDDR5x memory. Nvidia bonds the CPU, GPU, and memory together via its own NVLink-C2c interconnect. The CPU has twenty cores: ten ARM Cortex-X925 and ten Cortex-A725 cores.
Think of the superchip as an Nvidia CUDA alternative to the x86-based AMD Ryzen AI Max (Strix Halo). With that chip, HP has had a system on the market since the beginning of this year.
The HP ZGX Nano G1n AI Station runs on Nvidia DGX OS and includes the Nvidia AI software stack. The large amount of shared memory lets developers run AI models with up to 200 parameters on the device. Not enough? You can link two units together to double the compute power.
HP isn’t the first to launch such a small AI computer. That was Nvidia itself with the
Local Inference
HP does add its own twist with the HP ZGX Toolkit. The toolkit provides developers with open-source frameworks and the ability to evaluate models via Ollama. HP is focusing on a streamlined way to run models locally.
This lets you offload AI workloads from a “regular” computer—such as another HP PC, but also a Mac or Linux system—to the HP ZGX Nano G1n AI Station. In that case, the computer serves as a local inference machine, accessible over the network.
HP aims to launch the HP ZGX Nano G1n AI Station this autumn. We’ll have to wait a little longer for pricing.