AMD deploys Helios systems at Meta and invests 1 billion dollars

AMD says its first Helios rack-scale AI systems have been installed and are being tested at Meta, with a wider deployment planned. The companies are expanding a long-running supplier relationship into joint work across CPUs, GPUs, networking, complete rack systems and software.

Meta already uses millions of AMD EPYC processors and is a lead customer for the sixth-generation EPYC line. AMD Instinct accelerators are also in production for training, ranking and recommendation workloads. The companies now plan a custom Instinct GPU optimized for Meta, while their engineering work will influence future MI500 accelerators and next-generation rack-scale systems.

The partnership also reaches Meta’s models and agents. AMD says EPYC powers the dedicated compute environments behind the Muse personal AI agent, while its hardware provides immediate support for Muse Spark Open and Muse Glimmer. Engineers from both companies are contributing training optimizations to PyTorch, Triton and ROCm, which may also benefit developers outside Meta.

AMD will invest 1 billion dollars in Meta’s equity offering. The strategic signal is larger than that investment: Meta is giving AMD a role across the full AI stack instead of relying on it only as an alternative accelerator supplier. Helios is still in initial testing, so the broader rollout and its economics remain to be demonstrated.