The AI hardware fight is becoming bigger than a GPU comparison. Vultr has placed a $1.2 billion order for AMD Helios AI racks from HPE, buying a complete system that combines compute, networking, cooling, software and deployment services.1, 2
HPE announced a $1.2 billion order from Vultr for AMD Helios AI Rack by HPE systems. Each rack combines AMD accelerators, CPUs, networking and ROCm software with HPE scale-up switching, liquid cooling and services. It is HPE’s first order for the new Helios system. Deployment timing, margins and future utilization were not disclosed.1, 2
The Vultr order
The commercial unit is becoming the rack
HPE describes Helios as an integrated rack-scale system rather than a collection of standalone components. AMD supplies GPUs, CPUs, NICs and ROCm software. HPE adds scale-up networking, cooling, services and rack engineering.1
| Layer | Supplier / component | Why it matters |
|---|---|---|
| Accelerators | AMD Instinct MI455X | Model training and inference compute |
| CPU / NIC | AMD EPYC and Pensando | Host processing and network interface |
| Scale-up switching | HPE Juniper QFX5252 | Connects all 72 GPUs inside the rack-scale fabric |
| Cooling / services | HPE | Deployment and operation of dense systems |
Networking is part of the AI-system sale
The order supports HPE’s argument that its Juniper networking acquisition can participate directly in AI infrastructure spending. If scale-up switching becomes a standard component of large AI systems, networking revenue can grow alongside accelerator demand.1, 2
A cloud buyer is a useful commercialization test
Vultr operates AI cloud infrastructure, so the order ties Helios to a buyer that has to resell training and inference capacity to customers. That is more informative than a lab prototype, but the announced order still does not reveal utilization, customer demand or profitability after deployment.1
What the $1.2 billion does not tell us
- How quickly Vultr will deploy the racks.
- HPE’s gross margin on the order.
- How much of the system value goes to AMD versus HPE components and services.
- The utilization and economics of Vultr’s eventual customer workloads.
The order is an early signal that AMD’s challenge to Nvidia will be sold as a system, not just a processor. Winning AI infrastructure business increasingly means proving the whole rack can be delivered, connected, cooled and operated at scale.
Sources and methodology
Sources checked October 1, 2026. Dates and periods for individual figures are stated beside them.
- HPE press release via HPCwire: $1.2B Vultr AMD Helios order ↗Accessed 2026-10-01
- FierceWireless: HPE lands first AMD Helios order in $1.2B Vultr deal ↗Accessed 2026-10-01
Scope and assumptions
HPE did not disclose deployment timing or order-level margins.
The order does not establish future utilization or customer demand for Vultr’s resulting capacity.
Continue reading
AMD Is Paying $8.2B for World Labs. The Bigger Bet Is the Feedback Loop →
The Next AI Chip Bottleneck May Be the Connections Between the Chips →