HPE says Vultr placed a $1.2 billion order for AMD Helios AI racks

HPE says Vultr has placed a $1.2 billion order for AMD Helios AI Rack by HPE systems, a deal that puts AMD's latest rack-scale AI infrastructure into Vultr cloud data centers in the United States. The announcement matters because it is not another chip launch in isolation. It is an order for full racks that combine compute, networking, software and liquid-cooling services into a deployment package aimed at model training and inference.

What HPE says Vultr is buying

According to HPE, this is its first order for the AMD Helios system. The company says each rack integrates 72 AMD Instinct MI455X GPUs with AMD EPYC "Venice" CPUs, AMD Pensando Vulcano AI NICs and AMD ROCm software. HPE also says the racks use purpose-built HPE Juniper Networking QFX5252 scale-up Ethernet switch trays, with six switch trays per rack connecting the 72 GPUs over standards-based Ethernet.

That detail is important for buyers watching the AI infrastructure market. The practical issue is no longer only whether a provider can get accelerators. Large customers also need the surrounding network fabric, cooling design, service plan and software stack to behave like one system. HPE is pitching Helios by HPE as an open Ethernet-based alternative for cloud providers and enterprises that want rack-scale AI capacity without treating every layer as a separate integration project.

Why it matters for cloud AI capacity

Vultr says demand for high-performance AI infrastructure continues to exceed available capacity, and HPE says the systems will support both model training and inference workloads. If the deployment lands as described, Vultr gets another way to sell dense AMD-based AI capacity to customers that may not want to build data-center clusters themselves.

The CyberOGZ read is that the order is a useful signal for the post-GPU-scarcity phase of AI infrastructure. The headline number is large, but the more telling piece is the rack design: 72 GPUs, dedicated scale-up switching and direct liquid-cooling expertise bundled together. For teams choosing cloud AI providers, the decision point will be less about a single accelerator brand and more about whether the provider can offer predictable throughput, memory bandwidth, availability and operating cost at cluster scale.

There are still open items to watch. HPE did not give a rack count, a detailed delivery calendar or customer pricing for Vultr's eventual services. Those details will determine how quickly the order turns into usable capacity for developers and enterprises. For now, the announcement shows AMD, HPE and Vultr trying to convert demand for AI compute into a more packaged cloud infrastructure product.

Sources

• www.hpe.com

Cover image: j_cadmus, source, licensed under BY.