AMD and Meta Announce 6-Gigawatt Partnership to Power Next-Generation AI Infrastructure

AMD and Meta have announced a massive 6-gigawatt strategic partnership to build out Meta next-generation AI infrastructure. This multi-year agreement centers on the deployment of custom AMD Instinct GPUs across several hardware generations. The collaboration aims to accelerate the development of sophisticated AI models that will power experiences for billions of users.

Custom Silicon for Hyperscale Needs

The first phase of the deployment will feature a custom GPU based on the AMD Instinct MI450 architecture. These chips are specifically optimized for Meta unique workloads, featuring 432 GB of HBM4 memory to handle massive model weights. This specialized silicon will be housed in the AMD Helios rack-scale architecture, which was developed through the Open Compute Project for maximum efficiency. Beyond AI-specific tasks, these GPUs provide the raw throughput needed for a variety of high-performance computing needs. While their primary role is driving inference for Meta Llama models, the architecture is designed for the extreme density required by modern hyperscale data centers. This ensures that the infrastructure remains flexible as Meta computational needs evolve outside of standard AI training.

Integrated Systems and Software

Meta will also be a lead customer for the 6th Gen AMD EPYC processors, codenamed Venice. Built on a 2nm process, these CPUs will work alongside the MI450 GPUs to provide the orchestration and performance-per-watt necessary for gigawatt-scale operations. The integration of high-performance CPUs and GPUs is a critical pillar for Meta goal of creating personal superintelligence. To align their long-term interests, AMD has issued Meta performance-based warrants for up to 160 million shares of common stock. These warrants vest in tranches as Meta achieves specific deployment milestones, starting with the first gigawatt of shipments scheduled for the second half of 2026. This structure ensures both companies are deeply committed to the technical success of the rollout. AMD Chair and CEO Dr. Lisa Su stated that the collaboration places AMD at the center of the global AI buildout. Meta CEO Mark Zuckerberg noted that the partnership is a vital step in diversifying their compute resources for the future. Together, the companies are aligning their roadmaps across silicon and software to enable AI innovation at a global scale.