Meta Platforms Inc. announced on April 20, 2026, a significant expansion of its strategic partnership with Broadcom Inc. to co-develop multiple future generations of its custom silicon, the Meta Training and Inference Accelerator (MTIA). This collaboration is designed to accelerate Meta's internal artificial intelligence infrastructure capabilities and optimize its hardware for specific software workloads. The expanded agreement includes an initial deployment of AI hardware exceeding one gigawatt (GW) in power capacity, with plans for a broader multi-gigawatt rollout across Meta’s global data center network over the coming years.
The MTIA project is a cornerstone of Meta’s strategy to reduce its reliance on third-party merchant silicon and tailor its hardware for the company's specific social media, advertising, and metaverse applications. Broadcom, serving as the primary ASIC (Application-Specific Integrated Circuit) partner, provides the underlying physical design, high-speed interconnect technology, and manufacturing coordination. This new phase of the partnership focuses on the third and fourth generations of MTIA chips, which are engineered to handle increasingly complex recommendation algorithms and large-scale generative AI models.
Meta’s Vice President of Infrastructure, Santosh Janardhan, stated that the partnership with Broadcom is essential for scaling the company's AI compute efficiency and maintaining a competitive edge in infrastructure. According to official statements, the move to a multi-gigawatt deployment signifies one of the largest custom silicon commitments in the technology industry to date. Broadcom CEO Hock Tan noted that the collaboration leverages Broadcom’s expertise in co-packaging optics and high-bandwidth memory interfaces to meet Meta’s specific power, thermal, and performance requirements.
The initial one-gigawatt deployment is already underway, focusing on Meta’s latest generation of data centers. These facilities are being specifically retrofitted or built to support the high-density power requirements of the new MTIA chips. The multi-gigawatt roadmap is expected to span the next three to five years, integrating Broadcom’s latest 2nm and 3nm process technology nodes. This expansion follows Meta’s previous disclosure of its second-generation MTIA chip, which demonstrated significant performance improvements over its predecessor in ranking and recommendation tasks.
While specific financial terms of the expanded deal were not disclosed, the scale of the power capacity and the volume of silicon required suggest a multi-billion dollar commitment. Broadcom’s role includes providing the necessary intellectual property for PCIe Gen6 and Gen7 interfaces, as well as custom HBM3e and HBM4 integration. Meta confirmed that the MTIA chips will work alongside existing GPU clusters to provide a heterogeneous compute environment for its AI Research and product development teams.