Vultr commits $1.2 billion to HPE racks for its AI cloud

Privately held Vultr sells cloud-computing capacity to enterprises and service providers through virtual machines, dedicated servers and multi-node clusters. HPE said Vultr placed a $1.2 billion order for AMD Helios AI Rack systems that will support model training and inference at the cloud provider’s United States data centers.

Vultr’s product documentation shows how customers use that infrastructure. An administrator can choose virtualized instances or single-tenant bare-metal servers, combine CPU or GPU nodes into a cluster and select a scheduler such as Slurm or Kubernetes. Vultr provisions the networking, records resource use and lets the customer add or remove nodes.

The order adds the physical equipment behind that service. HPE says the integrated racks combine AMD GPUs and CPUs with software, Ethernet switching and direct liquid cooling. HPE will also provide deployment expertise for the dense computing systems.

Vultr presents computing as capacity that customers can scale when workloads change. Underneath that flexibility sits a capital-intensive operation: processors, networking and cooling must be installed before Vultr can make the capacity available. The order makes the size of that advance hardware commitment unusually visible.

This is HPE’s first order for the Helios system. HPE said its infrastructure relationship with Vultr began through Juniper Networks nearly three years ago and continued after Juniper became part of HPE. The announcement did not specify the number of racks, the deployment schedule or the payment structure.