HPE Lands $1.2 Billion AMD Helios AI Rack Deal with Vultr
Hewlett Packard Enterprise (HPE) has secured its first major customer for AMD’s Helios AI Rack, marking a significant challenge to Nvidia’s dominance in the AI server market. On September 30, 2026, HPE announced that Vultr, a leading cloud infrastructure company, placed a $1.2 billion order for the AMD Helios AI Rack by HPE systems. This order represents the first commercial deployment of a rack-scale design developed collaboratively by HPE and AMD over nearly a year. The deal pits 72 AMD Instinct MI455X GPUs per rack against Nvidia’s NVLink-based systems, testing whether standard Ethernet can compete with Nvidia’s proprietary interconnect.
The Helios AI Rack features 72 AMD Instinct MI455X GPUs, paired with sixth-generation AMD EPYC “Venice” CPUs and AMD Pensando Vulcano AI network interface cards. Each GPU includes 432GB of HBM4 memory, resulting in a total rack memory of 31TB and a rack compute capacity of 2.9 exaFLOPS. The system uses AMD’s ROCm software stack and HPE’s direct liquid cooling system. HPE is betting on Ethernet as an alternative to Nvidia’s NVLink, aiming to offer more supplier options and reduced vendor lock-in.
Vultr’s CEO, J.J. Kardwell, highlighted the supply constraints in the AI infrastructure market, stating that the collaboration with AMD and HPE enables faster deployment of high-performance AI infrastructure. Antonio Neri, HPE’s president and CEO, framed the order as proof of HPE’s ability to engineer complete AI architectures. The deal was unveiled at HPE’s Networking investor day, where the company raised its financial outlook for the coming fiscal year.
While the Helios rack matches Nvidia’s GB200/GB300 NVL72 racks in GPU count, performance parity remains unproven. AMD’s published figures for the Helios rack have not been independently verified, leaving open questions about how it will perform against Nvidia’s systems in real-world applications. The market reaction to the deal underscores the competitive stakes for AMD in the AI data center space.