Vultr to Deploy AMD MI455X Chips and Helios Rack-Scale System for AI Cloud Expansion

Vultr has announced it will deploy AMD's Instinct MI455X graphics processors and support the AMD Helios rack-scale system, positioning itself among the first cloud providers to offer access to the new hardware. The move expands Vultr's artificial intelligence infrastructure as cloud providers compete to attract companies training large models and running inference workloads.

The AMD MI455X is built on the CDNA architecture and includes HBM4 memory with higher bandwidth, alongside support for lower-precision AI data types. These features matter because developers increasingly need to fit larger models into memory while reducing the cost of serving AI applications once deployed. Helios adds a rack-scale configuration supporting 72 GPUs for customers needing large accelerator clusters.

Vultr is now taking pre-orders for reserved capacity tied to the new systems, with deployments planned across 2027 and 2028. The company said the systems will support AI training, inference, fine-tuning, agentic AI, and scientific computing workloads. For cooling, Vultr will incorporate direct liquid cooling options in response to the heat and power demands of large-scale AI deployments.

The announcement reflects the intensifying competition among cloud providers to secure early access to next-generation chips from major semiconductor groups. As AI workloads shift from experimentation to production, cloud platforms are racing to lock in supply of the latest accelerators. For Vultr, which positions itself as a global cloud infrastructure provider, early access to AMD's newest accelerator offers a competitive differentiator.

Vultr CEO J.J. Kardwell noted that customer demand is evolving from AI experimentation into production use, pushing buyers toward systems that can handle repeated inference tasks, larger models, and workloads spread across multiple servers and racks. He said the AMD Instinct MI455X and Helios architecture give customers the flexibility, control, and price-to-performance needed to scale AI initiatives.

The deployment also underscores the growing importance of memory and cooling in AI infrastructure. As model sizes continue to grow, effective heat management and memory bandwidth have become core technical challenges for cloud providers, going beyond simple compute scaling.