AMD to Supply Helios AI Systems to Microsoft in Second Half
AMD will begin supplying its Helios AI systems to Microsoft in the second half of this year, the company announced on July 20.
Microsoft will deploy Helios in its Azure cloud platform to support customer AI models and inference workloads. The rollout expands the partnership between the two companies following Microsoft's adoption of AMD's MI300X graphics processing units (GPUs) in 2023.
Microsoft will also introduce two new virtual machine (VM) series, Azure HDv2 and Azure HXv2. The companies jointly designed the HDv2 VM platform, which is optimized for agentic AI workloads, while the HXv2 series targets electronic design automation (EDA) applications.
Beyond Microsoft, AMD has secured Meta, OpenAI, and Oracle as Helios customers.
Meta announced in February that it plans to deploy AMD GPUs across 6 gigawatts (GW) of AI infrastructure and is expected to utilize approximately 1GW of Helios capacity this year. OpenAI and Oracle also plan to deploy Helios beginning this year.
Helios is AMD's rack-scale AI system that integrates GPUs, central processing units (CPUs), networking equipment and cooling infrastructure into a single platform. AMD first unveiled plans earlier this year to launch the system in the second half of 2026.
Each Helios rack is equipped with 72 AMD Instinct MI455X GPUs and 18 sixth-generation EPYC "Venice" CPUs. Built on an open rack architecture, the platform is compatible with third-party GPUs and CPUs.
The system also incorporates AMD's Pensando data processing unit (DPU). AMD strengthened its networking portfolio through the acquisition of Pensando Systems in 2022.
The MI455X is AMD's first GPU to feature HBM4 (High Bandwidth Memory 4). In March, AMD signed a strategic memorandum of understanding with Samsung Electronics to use Samsung's HBM4 memory in the MI455X.
Samsung's HBM4 delivers data transfer speeds of up to 13 Gbps and memory bandwidth of 3.3 TB/s.