Meta Platforms and Broadcom Inc. have expanded their partnership to develop custom AI processors for Meta's global data-center infrastructure [1, 2].
The agreement allows Meta to reduce its reliance on third-party chip designers while accelerating its internal AI roadmap. By developing bespoke silicon, the company can optimize hardware specifically for its large-scale AI workloads and compute demands [1, 3].
The updated deal extends the partnership through 2029 [3, 5]. While some reports describe the arrangement as a three-year deal [2], other sources indicate the timeline extends further into the decade [3].
Technical specifications for the project include the use of a 2-nanometer manufacturing process [3]. This advanced node is designed to increase efficiency and performance for the next generation of AI hardware.
Meta plans a rollout capacity of at least one gigawatt [2, 3] for these AI chips. The scale of the deployment reflects the massive energy and compute requirements needed to sustain the company's AI ambitions across its global network [1, 4].
Broadcom provides the essential intellectual property and design expertise required to bring these custom chips to market. This collaboration enables Meta to scale its infrastructure more rapidly than by relying on off-the-shelf components [2, 5].
“The updated deal extends the partnership through 2029.”
This partnership signals a strategic shift toward vertical integration in the AI sector. By designing custom silicon with Broadcom, Meta is attempting to decouple its growth from the supply constraints and pricing power of dominant GPU providers. The move to 2-nanometer technology and gigawatt-scale deployment suggests that Meta is preparing for a significant increase in the complexity and energy demands of its future AI models.
