Summary
AMD says Meta has installed initial Helios rack-scale systems and is testing them ahead of a broader deployment. The companies also described work spanning AMD CPUs, GPUs, networking and AI software.
AMD says Meta has installed its initial Helios rack-scale systems and is testing them ahead of a broader deployment. The announcement also describes an expanded collaboration across processors, accelerators, networking and software, with AMD technology already used in several parts of Meta’s computing infrastructure.
The distinction between testing and deployment matters: the initial Helios systems are installed, but AMD describes them as being evaluated on the way to wider use, rather than as already broadly deployed at Meta.
Helios systems are at the testing stage
Rack-scale systems combine computing equipment into a coordinated installation, rather than supplying only an individual processor or accelerator. AMD’s announcement positions Helios as part of a broader infrastructure effort involving compute, networking, systems and software. It does not provide technical specifications for the Meta installation.
The companies say their collaboration has developed over several years. AMD reports that Meta has deployed millions of EPYC processors across multiple generations and is a lead customer for 6th Gen EPYC CPUs. AMD EPYC processors support the dedicated computing environments behind Meta’s Muse personal AI agent, according to the announcement.
AMD also says Instinct GPUs are already in production at Meta for AI training, ranking and recommendation systems. Meta was an early definition customer for the MI450 architecture and is expected to deploy a custom Instinct GPU optimised for its workloads. The announcement links this work to future AMD platforms, including MI500-series GPUs and next-generation rack-scale systems.
Collaboration extends to models and software
The partnership also covers the software needed to run AI workloads on AMD hardware. AMD says it has worked with Meta to advance ROCm through PyTorch, and that it provided Day-0 support for Meta’s Muse Spark Open models on Instinct GPUs and Muse Glimmer on Ryzen AI PCs and Radeon GPUs. Day-0 support refers to software support available from a model’s initial release.
For large-scale AI training, the companies have been optimising recommendation models on AMD Instinct MI350X GPUs and contributing improvements to PyTorch, Triton and ROCm. These projects illustrate why AI infrastructure depends on more than accelerator chips alone: the hardware, networking and software must work together to run and train models at scale.
AMD says the companies are aligning product roadmaps and engineering decisions around Meta’s current and future workloads, with an ambition to build and deploy infrastructure at gigawatt scale. That describes the direction of the collaboration, while the nearer-term milestone in this announcement is the installation and testing of the initial Helios systems. AMD also says it will invest $1 billion in Meta’s equity offering.