HP, Red Hat and NVIDIA Plan an Enterprise Edge AI Platform

HP, Red Hat and NVIDIA have announced a planned open, enterprise-grade AI platform intended to extend data-center AI architecture to edge deployments. The offering combines HP’s ZGX Fury AI station with Red Hat AI Factory with NVIDIA, with the goal of running production inference nearer to the people, applications, machines and data that use it. 213

What HP announced

At the center of the hardware stack is the HP ZGX Fury, powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip. HP says the system is designed to provide up to 20 PFLOPS of FP4 AI performance for local inference workloads. 14

The announcement is about more than a workstation: HP is positioning the combined hardware and software as an enterprise platform for running and orchestrating multiple AI workloads outside a centralized data center. It is designed to support deployments that are local, cloud-based or span both environments. 15

Red Hat AI Factory’s role

Red Hat AI Factory with NVIDIA supplies the hybrid-cloud software and operations layer. Red Hat describes it as a jointly supported enterprise AI offering that combines its open hybrid-cloud platform and AI engineering capabilities with NVIDIA’s AI software, models, microservices, networking and accelerated infrastructure. 12

In HP’s proposed deployment, that layer is intended to help organizations deploy and manage AI models and applications across edge sites, data centers and cloud environments. The platform is also described as open and virtualized, with workload isolation and operational control intended to let organizations run multiple AI workloads on the same system while maintaining separation between them. 813

Why move inference closer to operations?

The central design choice is local or edge inference: processing model requests nearer to where data is created or used, rather than relying solely on remote cloud infrastructure. HP cites needs around latency, privacy, data sovereignty and connectivity as reasons enterprises may want this option. 8

That makes the platform relevant where operational data is sensitive, connectivity is constrained or a response must happen close to a physical process. HP specifically highlighted several settings:

• Manufacturing, where models can run closer to machines and production data.
• Software development, including departmental or on-premises inference and AI-assisted workflows.
• Healthcare, where sensitive data may need to remain closer to its source.
• Government, including air-gapped or isolated environments where cloud connectivity may be unsuitable. 58

These are proposed use cases, not performance guarantees. Whether an organization benefits will depend on its model sizes, data-handling requirements, network design and operational tooling.

Availability and what is still planned

HP said the ZGX Fury is available to order and is certified for Red Hat Enterprise Linux. 48 The broader ZGX Fury plus Red Hat AI Factory with NVIDIA solution, however, was announced as a planned platform rather than a fully general-availability integrated service. 12

HP also said customers will be able to evaluate the combination in a sandboxed environment on HP devices running Red Hat AI Factory with NVIDIA before moving use cases to production. 2 The announcement did not provide a specific launch date, eligibility requirements or technical configuration for that evaluation environment.

The practical takeaway

HP’s announcement is an effort to package powerful on-premises NVIDIA hardware with a Red Hat-managed hybrid AI software stack. For enterprises, the appeal is not simply the stated 20-PFLOPS FP4 figure; it is the prospect of running production AI nearer to operational data while keeping a consistent deployment and management approach across edge, data-center and cloud locations. 11213

Organizations evaluating it should distinguish between what is available now—the ZGX Fury hardware and its Red Hat Enterprise Linux certification—and what remains a planned integrated platform, including the announced sandbox evaluation path. 24