HPE announced a $1.2 billion order from Vultr for AMD Helios AI Rack by HPE systems on September 30, 2026. Vultr plans to deploy the systems in its U.S. data centers. HPE called it its first order for the new Helios system.
What HPE specified for each Helios rack
HPE’s configuration pairs 72 AMD accelerators with six scale-up Ethernet switch trays per rack. The announcement also names the system’s processors, network interface cards, software, fabric support and cooling.
| Component | Announced configuration |
| GPUs | 72 AMD Instinct MI455X GPUs per rack |
| Scale-up networking | Six HPE Juniper Networking QFX5252 Ethernet switch trays per rack, connecting all 72 GPUs |
| CPUs | AMD EPYC “Venice” CPUs |
| AI network interface cards | AMD Pensando Vulcano AI NICs |
| Software | AMD ROCm |
| Rack-scale fabric | Supports open fabrics such as UALink over Ethernet (UALoE) |
| Cooling and services | Direct liquid cooling; HPE Services support for end-to-end deployment and liquid-cooling expertise |
Scale-up networking connects the GPUs within the rack. HPE says the QFX5252 switch trays connect all 72 AMD Instinct MI455X GPUs in each system.
How HPE, AMD and Vultr fit into the order
Vultr is the customer and plans to operate the systems in its U.S. data centers. HPE supplies the rack-scale systems, scale-up networking and services; AMD provides the named compute and networking components and ROCm software.
HPE linked the order to nearly three years of collaboration between Vultr and Juniper Networks, which is now part of HPE.
