HPE Unveils Next-Gen AI Factory and Supercomputing Advancements with NVIDIA

Key Takeaways

• HPE adds NVIDIA Vera CPU blades to GX240 rack
• Quantum‑X800 InfiniBand offers 800 Gb/s per port
• Vera Rubin NVL72 targets 1‑trillion‑parameter models
• XD700 packs 128 Rubin GPUs per rack
• Multi‑tenant AI factories support GPU passthrough and MIG

Summary

HPE announced a suite of next‑generation AI factory and supercomputing solutions co‑engineered with NVIDIA, featuring the first NVIDIA Vera CPU compute blades, Quantum‑X800 InfiniBand switches, and the Vera Rubin NVL72 rack‑scale system for trillion‑parameter models. The portfolio also adds the high‑density XD700 server with up to 128 Rubin GPUs per rack, Blackwell RTX PRO 6000 GPUs, and expanded software services such as multi‑tenant GPU passthrough, Red Hat integration, and NVIDIA Mission Control. Availability spans from spring 2026 for multi‑tenant AI factories to early 2027 for the new hardware. These offerings target research labs, sovereign entities, and large enterprises seeking to merge AI workloads with traditional HPC.

Pulse Analysis

The HPE‑NVIDIA collaboration arrives at a moment when demand for exascale AI infrastructure is exploding across research institutions and cloud providers. By integrating NVIDIA’s latest Vera CPUs and Blackwell GPUs into its Cray‑based supercomputing platform, HPE delivers unprecedented compute density while preserving the energy efficiency that large‑scale data centers require. The addition of Quantum‑X800 InfiniBand, with 800 Gb/s per port, ensures that bandwidth‑bound AI workloads can scale without bottlenecks, positioning the GX5000 series as a premier choice for hybrid AI‑HPC workloads.

Beyond raw performance, HPE’s new AI Factory portfolio emphasizes operational flexibility. The Vera Rubin NVL72 rack‑scale system is engineered for frontier‑scale models exceeding one trillion parameters, leveraging six‑generation NVLink and ConnectX‑9 SuperNICs for seamless scaling. Meanwhile, the XD700 server, built on the Open Compute Project framework, doubles GPU density per rack, delivering up to 128 Rubin GPUs while cutting power and cooling footprints. Coupled with liquid‑cooling innovations, these hardware advances reduce total cost of ownership and accelerate time‑to‑insight for AI‑driven research and enterprise applications.

Software and service integration complete the full‑stack proposition. Multi‑tenant capabilities, powered by NVIDIA MIG and SUSE virtualization, enable secure GPU sharing across virtual machines, while Red Hat Enterprise Linux and OpenShift support streamline deployment in enterprise environments. NVIDIA Mission Control tools such as Run:AI and Dynamo provide automated orchestration and autonomous recovery, simplifying management of massive AI clusters. Together, these capabilities give organizations—from sovereign cloud providers to biotech firms—a ready‑to‑run platform that can handle the most demanding AI and HPC tasks today and scale for future breakthroughs.