AMD and Meta Deepen Partnership With EPYC CPUs, Helios Racks and Instinct GPUs

AMD and Meta announced further deepening of their strategic partnership, in which Meta will leverage AMD's Helios rack-scale solutions to power the latest AI models such as the Muse AI Agent, according to the report. AMD confirmed that initial AMD Helios rack-scale systems are being deployed at Meta.

AMD chief executive Lisa Su said Meta is a strategic partner for AMD and that the companies have built the partnership through years of technical collaboration. She said that with Helios now running at Meta, the partnership is being taken to a new scale, and that the companies are building and deploying infrastructure at gigawatt scale.

The report says Meta has deployed millions of EPYC processors and is a lead customer for 6th Gen EPYC CPUs, with AMD Instinct GPUs in production for training, ranking and recommendation. It adds that Meta is described as an early definition customer for the Instinct MI450 series and will deploy a custom Instinct GPU optimized for its workloads, and that the partnership will expand into future AMD platforms such as Instinct MI500 and next-gen rack-scale systems.

Meta is said to deploy a heterogeneous fleet of NVIDIA GPUs, AMD GPUs and proprietary MTIA accelerators to reduce costs and mitigate vendor lock-in, while the report notes that a long history of NVIDIA-specific optimizations hinders performance portability to other hardware. A cited technical paper using AMD Instinct MI350X versus NVIDIA B200 for recommendation model training as a case study reports up to 1.5x improvement in end-to-end training throughput compared with a directly ported baseline, with contributions upstreamed to the open-source PyTorch, Triton and ROCm ecosystems.

According to the report, combining EPYC CPUs including Venice, Helios rack-scale systems and Instinct GPUs with software co-optimization and open ecosystems like ROCm and PyTorch is enabling Meta's most ambitious models while reducing reliance on any single vendor.