HPE's annual conference opens with a networking-first keynote, unveiling liquid-cooled switches, quantum-safe firewalls, and AI rack solutions as Juniper Networks integration reaches full operational status.
HPE opened its Discover 2026 conference in Las Vegas with a keynote that signaled a clear strategic pivot: networking now leads the AI infrastructure conversation. CEO Antonio Neri devoted the first half of the hour-long presentation to HPE Juniper Networking products, positioning the newly integrated portfolio as the foundation for AI data centers.
The messaging order was deliberate. While competitors at Dell Technologies World emphasized compute-driven transformation, HPE led with switches, routers, and firewalls — a networking company that also sells AI compute, as opposed to a compute company that sells networking gear.
Juniper Integration Complete
The headline announcement formalized what HPE has been building since closing the $14 billion Juniper acquisition: HPE Juniper Networking now sits inside HPE AI Data Center Solutions as a unified business unit. The integration brings Juniper's routing, switching, and security portfolio under the HPE GreenLake cloud operating model.
Liquid-Cooled Switching for AI Racks
HPE unveiled the QFX5252, a scale-up switch designed for the AMD Helios AI rack. The 1U platform runs dual Broadcom Tomahawk 6 ASICs, supports UALink over Ethernet (UALoE), and ships with SONiC plus AI-native operations. Liquid cooling comes standard — a requirement for the density Helios targets.
For scale-out fabrics, the QFX5250 delivers 102.5 Tbps in a liquid-cooled chassis. The PTX12000 router adds 800 Gbps routing with ZR/ZR+ coherent optics for inter-data-center links. HPE pointed to its earlier coverage of 800G ZR+ optics using Marvell COLORZ 800 technology for technical depth.
Quantum-Safe Firewall in 1U
The SRX4700 packs 1.4 Tbps of firewall throughput in a single rack unit with post-quantum cryptography support. HPE representatives noted the platform aligns with their CyPerf test platform, which now exceeds 2 Tbps line-rate testing capacity.
Inference-Optimized Switching
The QFX5140 Inference Switch targets a specific capacity gap: 16 Tbps in 1RU, positioned between the common 12.8T and 25.6T merchant silicon points. HPE says the architecture optimizes for inference workloads rather than training-scale throughput.
The MX301 router extends the portfolio to the inference edge with 1.6 Tbps full-duplex 400G connectivity and sixth-generation Trio silicon, managed through AIOps.
AI-Native Operations and Agentic Management
HPE Marvis, the conversational AI assistant from Mist, now integrates with Aruba Central for campus and branch networking. The broader Agentic Enterprise framework introduces agent registration, governance, and security controls — addressing the "shadow workforce" of autonomous AI agents operating in enterprise environments.
HPE also announced support for NVIDIA OpenShell and NemoClaw, signaling alignment with NVIDIA's agentic AI stack.
Private Cloud AI and Storage Foundation
HPE Private Cloud AI received inference-focused updates: multi-node inference, a unified gateway, and shared KV cache capabilities. The platform runs on HPE Alletra Storage MP X10000, which now includes real-time data enrichment and native Model Context Protocol (MCP) integration. NVIDIA CMX integration was not explicitly mentioned but remains part of the underlying architecture.
Tiered AI Factory Portfolio
HPE segmented its AI factory offerings by customer class:
- HPE Private Cloud AI — Enterprise deployments
- HPE AI Factory at Scale — Neoclouds and hyperscalers
- HPE AI Factory Sovereign — Government and defense
Confidential computing is now standard across the NVIDIA portfolio within HPE AI Factory, including upcoming Vera-based systems.
Power and Customer Validation
Siemens Energy appeared as a customer vignette, describing predictive failure analysis on HPE infrastructure — not power sourcing, as some analysts anticipated. Vultr confirmed a next-generation AI build-out partnership leveraging HPE Networking.
Strategic Contrast
The keynote's structure revealed HPE's differentiated bet. Jensen Huang appeared onstage at Dell Technologies World 2026; he was absent from HPE Discover. NVIDIA remains a critical partner — Spectrum-X, BlueField-3, and ConnectX integrate into HPE AI Grid — but HPE's narrative centers on the network as the AI bottleneck, not the GPU.
That framing matters. As AI clusters scale beyond single racks, the interconnect becomes the limiting factor. HPE's portfolio — now spanning Juniper routing, Helios liquid-cooled racks, Alletra storage, and GreenLake operations — aims to own that layer entirely.
The keynote concluded after roughly 60 minutes. HPE executives move to breakout sessions for deeper technical briefings on each product line.
Comments
Please log in or register to join the discussion