HPE shares rise after $1.2 billion AMD Helios order

Vultr plans to install the Helios systems at its U.S. cloud data center locations.[1] HPE and AMD will support the company in serving service-provider and enterprise demand for AI model training and inference.[1] Each rack combines 72 AMD Instinct MI455X GPUs with AMD EPYC “Venice” CPUs, AMD Pensando Vulcano AI networking cards and AMD ROCm software.[1] AMD designed Helios for high-density trillion-parameter model training and high-volume inference.[1]

HPE supplies six Juniper Networking QFX5252 scale-up Ethernet switch trays per rack.[1] Those trays connect all 72 GPUs using high-bandwidth, low-latency, standards-based Ethernet, and the rack supports open, interoperable scale-up fabrics such as UALink over Ethernet.[1] HPE also provides direct liquid cooling and deployment services.[2] The company says its expertise deploying and liquid-cooling large AI clusters helps reduce project and operational risk.[1]

HPE raised its fiscal 2027 networking revenue growth outlook to a range from the high teens to the low 20s percent, up from its earlier forecast of 14% to 17%.[2] HPE identified Helios as the one new factor since that earlier outlook.[2] Rahim estimated that Helios networking could represent more than $1 billion of opportunity over the next two years, while networking tray orders had already exceeded $200 million.[2][4] HPE did not disclose how much of Vultr’s $1.2 billion order is for networking.[2]

Neither HPE nor Vultr disclosed the number of racks or a delivery schedule.[2] AMD CEO Lisa Su said initial Helios shipments would begin later in the third quarter, increase in the fourth quarter and continue ramping through 2027.[2] HPE said its fiscal 2027 networking revenue would be more back-end loaded than usual, partly because of Helios shipment timing.[2] The order builds on nearly three years of Vultr–Juniper collaboration, and HPE is now supplying Vultr with rack-scale systems based on both AMD and Nvidia GPUs.[1][2]