Vultr Picks HPE and NVIDIA for Enterprise AI Services

Alternative cloud provider Vultr has chosen HPE and NVIDIA to fuel its large-scale datacenters designed to support enterprise customers. The move highlights demand for the enterprise segment, while opening up potential possibilities for further Vultr/HPE arrangements.

“AI infrastructure remains significantly underbuilt globally, and enterprises increasingly require high-performance AI compute integrated seamlessly at the edge,” stated Vultr CEO J.J. Kardwell in a press release. “We are expanding our capabilities with a focus on supporting enterprise demand for decentralized, latency-sensitive workloads across Vultr’s extensive global network.”

Vultr CEO J.J. Kardwell. Source: HPE

Vultr’s deployments will include the NVIDIA GB300 NVL72 by HPE, a version of NVIDIA’s popular integrated rack system supported by HPE and packaged with HPE’s direct liquid cooling technology. Vultr will connect multiple rack systems with NVIDIA’s Spectrum-X Ethernet, which in turn is part of NVIDIA’s AI Computing by HPE series of options that include pairing HPE ProLiant Compute XD685 servers with NVIDIA GPUs.

HPE Gets a Visibility Boost

The news boosts HPE’s visibility as a solution provider for alternative cloud providers serving enterprise customers, an area in which Vultr is leading. It also hints at other announcements made at HPE’s Discover 2026 customer conference in Las Vegas this week—announcements that could put HPE at odds with NVIDIA.

At the show, HPE revealed a slew of networking capabilities that leverage its $14-billion purchase of Juniper Networks last summer. And in the process, it turned up a bit of a surprise.

Specifically, HPE has gotten into scale-up networking for AI, a kind of side hustle to its push into automated agentic AI and inference networking. The scale-up effort provides an Ethernet-based alternative to NVIDIA’s NVLink architecture, one based on Juniper technologies, though it may be a while before HPE’s solutions hit the mainstream.

Here’s the detail: Making good on earlier announcements, HPE has introduced the HPE Juniper Networking QFX5252 switch, a scale-up interconnect module for use with AMD’s Helios integrated rack for AI. It’s not clear whether the switch will be sold for non-Helios configurations, and HPE didn’t immediately respond to a request for comment.

Still, the QFX5252 offers some intriguing possibilities: It supports data throughput rates to 260 terabytes/second; it is based on UALink over Ethernet (UALoE), an emerging but incomplete working spec of the UALink Consortium, in which HPE has been active from the start; and it’s managed by the HPE Networking Data Center Director, which includes the Apstra intent-based networking capabilities.

Separately, Juniper Networks’ routing technology has also been deployed in an integrated AI rack called the Napier TDN72 introduced this week by Tensordyne AI. That startup claims to outperform NVIDIA across a range of criteria thanks to its uniquely designed Napier chips. And the TDN72 incorporates Juniper’s PTX10008 router ASICs, adapted to fit Tensordyne's custom AI compute trays, as part of its backplane architecture.

A Foot in the Door

HPE’s other Juniper news this week includes the addition of a new QFX5140 Switch for inference clusters and edge AI; the integration of Juniper’s Mist AI AIOps technology with the HPE Networking CX switches; and the addition of Juniper's Marvis AI “self-driving network” capabilities to HPE Aruba Central.

The successful integration of Juniper Networks’ products into HPE switches for use in networking for AI (and AI for networking) can’t come a moment too soon. Demand for inference and agentic AI is growing fast among enterprises of the kind using Vultr’s services.

“The success of agentic AI in the enterprise depends on a modern networking foundation built for autonomous workflows, where network performance, reliability, and intelligence determine the effectiveness of the entire AI architecture,” said Rami Rahim, EVP, President and General Manager, Networking, HPE, in a press release. “HPE is delivering that foundation.”

HPE also has gotten a significant foot in the door with Vultr, which could eventually lead to deployment of HPE’s scale-up interconnection technologies, despite the potential for pushback from NVIDIA. After all, it’s clear that hyperscalers are adapting NVLink alongside their own scale-up interconnects. Why not large enterprises as well?

Futuriom Take: Vultr’s adoption of HPE-supported NVIDIA racks could lead to more interesting developments, as HPE expands its networking capabilities to scale-up within the rack.