HPE Accelerates AI Workloads with Cutting-Edge Datacentre Solutions

Why It Matters

HPE’s 1.6 Tbps Ethernet solution challenges InfiniBand’s dominance, enabling faster, cheaper AI deployments and giving cloud providers a competitive edge in scaling massive GPU clusters.

Key Takeaways

• AI workloads demand ultra‑high bandwidth, low‑latency networking for clusters
• HPE introduced 1.6 Tbps per‑port switches and routers to accelerate AI
• Intent‑based Abstra and AI‑Ops simplify AI fabric deployment
• Customers using HPE power half‑million GPU clusters efficiently
• Ethernet switches match or exceed InfiniBand performance in AI

Summary

At Mobile World Congress 2026, HPE unveiled a purpose‑built networking portfolio designed to accelerate artificial‑intelligence workloads across distributed GPU clusters. The company emphasized that modern AI training and inference demand far more bandwidth, lower latency, and tighter congestion control than traditional data‑center traffic, making the network the critical fabric for performance.

HPE’s offering spans QFX switches, PTX routers, and SRX firewalls, highlighted by the industry‑first 800 Gbps devices and the newly announced 1.6 Tbps‑per‑port switches and routers. Integrated AI‑optimized software provides congestion management, load‑balancing, and other performance‑enhancing features, while the Abstra intent‑based networking suite and AI‑Ops‑driven MIS technology automate design, deployment, and troubleshooting of the AI fabric.

During the demo, HPE showcased three use cases: the end‑to‑end AI fabric within a data center, seamless inter‑data‑center and edge inference connectivity, and automated fabric management that delivers rapid visibility and root‑cause analysis. Real‑world customers include a large model‑builder operating over half a million NVIDIA GPUs, a neo‑cloud provider supporting both AMD and NVIDIA GPUs, and a Korean enterprise that replaced InfiniBand with HPE’s Ethernet switches, reporting equal or better performance.

The rollout signals a shift toward open, ultra‑high‑speed Ethernet as a viable alternative to proprietary interconnects, promising lower total‑cost‑of‑ownership and faster time‑to‑market for AI services. By simplifying network operations and delivering unprecedented bandwidth, HPE positions itself as a strategic enabler for cloud providers, enterprises, and AI innovators seeking to scale next‑generation workloads.

Original Description

As AI workloads scale across distributed GPU clusters, networking becomes critical to performance. Amit Sanel explains how HPE delivers purpose-built AI fabrics with high-bandwidth switching, intelligent congestion management, and AIOps-driven operations to optimise AI training and inference across modern datacentres.

Featuring: Amit Sanyal, Senior Director, Product Marketing for Data Center Networks, HPE

Recorded March 2026

#telecomtv #mwc26 #hpe