HPE wins a $1.2 billion Vultr order to supply AMD Helios AI racks for US data centers, giving the new rack-scale system a substantial commercial deployment. The hardware is designed for large-model training and inference, with HPE providing the networking layer that connects the accelerators.

 

Hewlett Packard Enterprise disclosed the order on September 30 during its networking investor day. The agreement deepens an existing relationship with Vultr while adding AMD infrastructure alongside Nvidia systems that the cloud provider selected from HPE earlier this year.

 

The order establishes four concrete points:

  • Vultr committed $1.2 billion to the deployment
  • AMD Helios racks will enter US data centers
  • HPE switches and networking software are included
  • The systems target AI training and inference workloads

 

HPE Wins $1.2 Billion Vultr Order for Helios

Reuters reported that HPE will deliver AMD Helios AI racks equipped with accelerators, HPE networking switches and software. The systems are destined for Vultr facilities in the United States, where the privately held cloud operator sells computing capacity to enterprises and developers.

 

The size of the commitment matters because Helios is entering a market dominated by Nvidia’s rack-scale platforms. A cloud provider placing a billion-dollar order signals that AMD’s architecture has advanced beyond demonstrations and partner road maps into a deployment with a defined customer and geography.

 

HPE did not disclose the number of racks, delivery timetable, individual system price or specific data-center locations. Those omissions prevent a calculation of the deployment’s compute capacity, power requirement or revenue recognition schedule, so the $1.2 billion value remains the clearest measure of its scale.

 

Vultr had already announced support for Helios in July. The cloud provider said customers would access the rack-scale system through its console and application programming interface, positioning the hardware as part of its composable cloud rather than as a dedicated installation for a single customer.

 

AMD Helios Combines GPUs, CPUs, Networking and ROCm

AMD describes Helios as an integrated building block for frontier AI, foundation-model training and large-scale inference. The design combines Instinct MI455X accelerators, sixth-generation EPYC server processors, Pensando networking and the ROCm software stack in one rack-scale architecture.

 

That integration addresses a major problem in AI infrastructure: accelerators cannot deliver their rated performance if memory, networking, power delivery or software becomes a bottleneck. Rack-scale systems coordinate those components before deployment, reducing the engineering work required of a cloud operator assembling large clusters.

 

AMD says Helios can deliver exaflop-class AI performance and high memory capacity and bandwidth. Those are vendor claims, and the Vultr order does not provide independent benchmarks. Real-world performance will depend on workload type, cluster size, software maturity and the efficiency of communication between racks.

 

HPE’s role extends beyond reselling AMD processors. The company supplies scale-up networking within the rack, broader data-center connectivity, management software and deployment services. That combination lets HPE compete for more of the system value while giving Vultr one vendor responsible for integration and support.

 

Vultr Adds AMD Capacity Alongside Nvidia Systems

The Helios purchase is not an exclusive shift away from Nvidia. In June, Vultr selected HPE’s Nvidia GB300 NVL72 systems and Spectrum-X Ethernet for separate large-scale AI data-center deployments. HPE said those environments would support model training, inference and private-cloud workloads.

 

Running both suppliers can give Vultr more capacity options and reduce dependence on a single accelerator ecosystem. It also creates operational complexity because customers may need different software images, optimization libraries and monitoring practices for Nvidia CUDA and AMD ROCm environments.

 

AMD’s opportunity depends partly on whether cloud customers can move models and inference services between those stacks without costly rewriting. ROCm has expanded its framework and library support, but Nvidia retains a large developer ecosystem and a long deployment history in production AI clusters.

 

For customers, the important question is not only peak chip speed. Availability, cost per token, memory capacity, network performance and software support determine whether a platform is economical. Vultr can expose those tradeoffs directly by offering both architectures through the same cloud operator.

 

HPE Ties the Order to Faster Networking Growth

HPE announced the Vultr order while raising its networking outlook. The company now expects networking revenue to grow at a high-teens compound annual rate from fiscal 2026 through fiscal 2029, compared with its previous forecast of 5% to 7% annual growth over an earlier period.

 

HPE also lifted its fiscal 2027 networking growth forecast to the high teens through low 20s percentage range. Its networking business expanded after the acquisition of Juniper Networks, giving HPE a wider portfolio for connecting servers, data centers, campuses and service-provider infrastructure.

 

The order shows why networking has become central to AI system sales. Large clusters require high-bandwidth links inside each rack and across thousands of accelerators. Congestion or slow collective communication can leave expensive processors idle, turning the network into a direct constraint on training time and inference cost.

 

Execution remains the next test. HPE and AMD must deliver systems at scale, while Vultr must power, cool and operate them reliably. Shipment milestones, customer availability and measured performance will reveal whether the $1.2 billion commitment becomes a durable alternative to Nvidia-based AI infrastructure.

 

Related Research