HPE has booked its first order for the AMD Helios AI Rack by HPE, a $1.2 billion commitment from Vultr to deploy the 72-GPU racks across Vultr’s cloud data centers in the United States. HPE announced the order on September 30, the same day it held its Networking investor day, and the deal is the first customer proof point for the rack-scale design HPE, AMD, and Broadcom lined up in December, in which HPE supplies the scale-up Ethernet switching and software inside each Helios rack.
Vultr, which HPE describes as the world’s largest privately held cloud infrastructure company, said in July it was taking pre-orders for reserved Helios capacity for deployments spanning 2027 through 2028. The HPE order puts HPE-built racks behind that program. “Demand for high-performance AI infrastructure continues to outpace available capacity, and our work with AMD and HPE enables us to bring Helios rack-scale compute and scale-up networking online for customers faster,” said J.J. Kardwell, CEO of Vultr. “Together, we’re delivering the memory capacity, interconnect performance, and energy efficiency customers need to run the largest and most demanding AI training and inference workloads.”
What Is in a Helios Rack
Each rack carries 72 AMD Instinct MI455X GPUs with 6th Gen AMD EPYC “Venice” CPUs, AMD Pensando Vulcano AI NICs, and the ROCm software stack. The MI455X is AMD’s first CDNA 5 accelerator, with 432GB of HBM4 per GPU, and a full rack adds up to 31TB of HBM4, 1.7PB/s of memory bandwidth, and 2.9 exaFLOPS of FP4 by AMD’s figures. The GPUs share memory over UALink over Ethernet (UALoE), the open scale-up fabric AMD runs on standard Ethernet, with 260TB/s of scale-up bandwidth inside the rack and 43TB/s of scale-out bandwidth leaving it.
HPE’s piece of the rack is largely the network: six HPE Juniper Networking QFX5252 scale-up Ethernet switch trays sit in every Helios rack and connect all 72 GPUs, and HPE says the switching and its software were purpose-built for Helios. The order also extends a relationship that predates the merger: HPE says Vultr has worked with Juniper Networks for nearly three years, and that collaboration continues now that Juniper is part of HPE.
Liquid Cooling, Services, and HPE’s Networking Pitch
The racks use HPE’s direct liquid cooling, which HPE positions as exascale-derived, and HPE Services handles deployment and liquid-cooling integration for large AI clusters. HPE sells the combination as part of its AI Data Center Solutions portfolio, an integrated stack with AMD built on open standards. “AMD Helios by HPE demonstrates HPE’s unique ability to engineer the complete AI architecture with rack-scale compute, scale-up networking, and direct liquid cooling, all delivered by our global services,” said Antonio Neri, president and CEO of HPE.
Vultr already offers AMD Instinct capacity and built a hybrid cloud reference design with AMD and NetApp earlier this year, and Microsoft said in July it would ramp Helios at scale on Azure.
Amazon