
What happened
Japan HP announced the HP ZGX Fury G1n AI Station, a tower workstation with the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip and 748GB of coherent memory, priced from 2849万円 and shipping October 27, 2026.
Why it matters
It lets companies run AI training, fine-tuning and inference locally instead of in the cloud, so sensitive data stays in-house; the company expects sales of several dozen units a month.
What to watch
Sales of a rack-mountable 5U unit with a three-year on-site warranty begin October 27, with orders taking roughly 1 to 1.5 months to ship.
WHO IT HITSManufacturers, research and development teams, and public-sector IT staff who handle confidential CAD/CAE or technical data can now keep AI work on-premises rather than sending it to a cloud or remote data center.
Summaries like this, every morning.
The HP ZGX Fury G1n AI Station is not a clean-sheet design: HP built it on the basic architecture of NVIDIA's DGX Station, the desktop-side AI supercomputer, and sells it under its own brand. That lineage explains the spec sheet — a GB300 pairing a Blackwell Ultra GPU with a 72-core Grace CPU, 748GB of coherent memory, and support for models up to 1 trillion parameters, including fine-tuning of models above 100 billion parameters.
What HP is using to differentiate is software and support rather than silicon. It bundles the HP ZGX Toolkit and HP Z Runtime at no charge, and says it is the only OEM maker among those building DGX Station-based products that currently holds Red Hat certification. A Japan HP product manager noted that large enterprises and public organizations often run Red Hat Enterprise Linux on core systems, so patches can be managed the same way. Support for NVIDIA OpenShell, an open-source runtime that runs AI agents in isolated sandboxes, targets privacy and security-conscious operations.
HP is positioning the machine as a complement to the cloud, not a replacement. A product manager pointed to the cloud's strength in flexibly scaling resources with data volume and usage frequency, and said customers should choose hybrid on-premises and cloud setups. The company expects many early buyers to run pilot deployments before moving to production, forecasts sales of several dozen units a month, and has placed a unit at its Tokyo office for customers to test.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
KnowledgeWork raised ¥4.8 billion in debt financing, including a credit facility, from seven lenders including…

Money Forward Home started connecting 'Money Forward ME' to Meta's personal AI agent 'Muse', so users who gran…

GENSHI AI opened applications through Dec 25, 2026 for 5 partner institutions to test RoundTrace, a chat tool…

Apollo's Torsten Slok said the S&P 493 — the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia an…
Meta beat revenue estimates with sales up 26%, but third-quarter capex of $34.9 billion topped operating cash…

Goldman Sachs strategists including Kamakshya Trivedi wrote in a Friday note that the yen is a more attractive…
