
AMD and Meta expand AI collaboration with Helios systems
AMD says initial Helios rack-scale systems have been installed and are being tested at Meta. The collaboration spans EPYC processors, Instinct accelerators, networking and open-software development.
Initial Helios installations are being tested

AMD’s 27 October announcement says initial Helios solutions have been installed at Meta and are being tested ahead of broader deployment. That does not mean Meta’s entire infrastructure has moved to the new systems. The companies emphasise jointly developing hardware and software for current and future AI workloads.
Helios describes rack-scale system design beyond an individual accelerator card. Coordinating CPUs, accelerators, networking and software makes data movement between components an important consideration for large models. The announcement concerns preparation for wider use of that integrated infrastructure.
EPYC and Instinct already support services
AMD says EPYC processors support the computing environments behind Meta’s Muse personal AI agent. It reports that Meta has deployed millions of EPYC CPUs across generations, is a lead customer for sixth-generation EPYC and is collaborating on future generations including Verano.
Instinct accelerators are used for training, ranking and recommendation workloads. Meta is described as an early definition customer for MI450 and will use a custom Instinct solution optimised for its workloads. MI500-series GPUs and future rack-scale systems form part of the forward-looking roadmap.
Open software and model support
ROCm, PyTorch and Triton feature in the software collaboration. The companies highlight optimisation for large-scale training and recommendation models on AMD accelerators. Day-zero support includes Muse Spark Open on Instinct and Muse Glimmer on Ryzen AI PCs and Radeon GPUs.
AMD also announced that it will invest $1 billion in Meta’s equity offering. This is an announced investment plan rather than a completed transaction reported here. The technical partnership extends beyond supplying one product, with the companies seeking to shape system and software decisions around future model requirements.
Sources and further information: AMD official announcement.



