HPE Discover 2026: a bet on AI networking, Gen12 servers and Private Cloud AI
Servix technical team · Updated
HPE Discover 2026, the main annual conference of Hewlett Packard Enterprise, took place in Las Vegas from June 16 to 18. The key theme of Antonio Neri's keynote: HPE positions itself primarily as a networking company for the AI era and builds compute and storage around the network. After the completion of the Juniper integration, this area has been combined into HPE AI Data Center Solutions. We have collected what really matters for data center projects.

QFS5252 switch (Broadcom Tomahawk 6) for scale-up accelerator interconnect
Networking at the center of the strategy
The main block of announcements covered networking equipment for AI factories — both for connecting accelerators within a rack (scale-up) and between nodes (scale-out):
- QFS5252 — a scale-up switch built on two Broadcom Tomahawk 6 with UALink over Ethernet support and liquid cooling; designed to link GPUs into a single domain.
- QFX5250 — a scale-out switch with 102.5 Tbps throughput and liquid cooling for horizontal scaling of AI clusters.
- QFX5140 — a 16 Tbps inference switch in 1U on Broadcom Trident 5 for model inference infrastructure.
- PTX12000 — a router with support for 800G and ZR/ZR+ coherent optics for data center interconnect (DCI).
- MX301 — a 1.6 Tbps edge router in 1U with 400G ports for inference at the edge.
- SRX4700 — a 1.4 Tbps firewall in 1U with quantum-resistant cryptography.

QFX5250 scale-out switch, 102.5 Tbps
Servers and the AI platform
In compute, HPE showed the new generation of ProLiant and a redesigned stack for private AI:
- HPE ProLiant Gen12 DL394 with NVIDIA accelerators — a platform for training and inference.
- HPE ProLiant EL2000 — a compact edge server on Intel Xeon 6.
- HPE Private Cloud AI — an updated solution with NVIDIA integration and Alletra Storage MP X10000 storage, with multi-node inference, a unified gateway and a shared KV cache. It is offered in three tiers: Private Cloud AI (enterprises), AI Factory at Scale (hyperscale) and AI Factory Sovereign (government and defense). Systems based on NVIDIA Vera get confidential computing by default.
Data storage
HPE Alletra Storage MP X10000 — storage for AI workloads with file and object access and real-time data enrichment. The logic is simple: next to accelerators you need not only a network but also fast access to large volumes of unstructured data.
Management and ecosystem
- HPE Mist and the virtual assistant Marvis now manage HPE Networking CX switches through natural-language actions.
- HPE CloudOps extends Morpheus to legacy VMware environments.
- HPE Unleash AI — a partner program for the AI ecosystem.

PTX12000 router for data center interconnect (DCI)
What this means for customers in Uzbekistan
The practical takeaway from Discover 2026: when designing AI infrastructure, the network stops being “plumbing” and becomes the defining element — both the cost and the achievable performance of a cluster depend on it. Even if a large AI factory is not on the table, the trends apply to ordinary data centers: the growing role of 400/800G, liquid cooling arriving in dense configurations, demand for fast storage systems for unstructured data, and the move to quantum-resistant perimeter protection.
Most of the announced platforms will not be available to order right away — but the current HPE ProLiant Gen11, Aruba CX and Alletra lines already cover these tasks and remain relevant. We will help select a configuration for a specific workload and calculate the power and cooling infrastructure. Write to [email protected] or on Telegram.
Source and photos: ServeTheHome, Patrick Kennedy, 16.06.2026 — servethehome.com. Published with attribution.
Related selections and tools: HPE ProLiant Gen11.
Related reading
HPE introduced ProLiant Gen12 on Intel Xeon 6
ProLiant Gen12 on Intel Xeon 6: more cores and memory channels, new iLO, firmware protection. When a generation change is justified and when Gen11 from the catalog is the better deal.
HPE ProLiant DL145 Gen11 with NVIDIA: an edge node for AI inference
A server with an NVIDIA accelerator for on-site inference: how inference differs from training, which card is needed, how much video memory and when it pays off.
HPE ProLiant DL325 and DL345 Gen12: single-socket servers on AMD EPYC 9005
HPE ProLiant DL325 and DL345 Gen12 on AMD EPYC 9005: 12 memory channels, one socket instead of two, license savings. How they differ and what to choose now.
SFP28, QSFP28 and DAC for the server room and data center: what to choose for 25G and 100G
When to use DAC and when optics at 25G and 100G: cable length and FEC, SR or LR, MPO or LC, breakout cables. Fibreon modules and DAC cables are in stock in Tashkent.




