Meta Platforms is intensifying its collaboration with Broadcom Inc. on custom artificial intelligence processors. This extended partnership will focus on the design and deployment of Meta's custom-designed chips for AI workloads, known as Meta Training and Inference Accelerators (MTIA).
Meta has committed to an initial deployment of 1 gigawatt of its MTIA chips, with plans to scale to multiple gigawatts. These chips will leverage Broadcom's technology, including its two-nanometer process for custom silicon in the AI industry. The initiative aims to build a substantial computing foundation for Meta's ambition of delivering 'personal superintelligence.'
The MTIA chips offer Meta an alternative to high-cost GPUs from companies like Nvidia and AMD. Similar to custom chips developed by Google and Amazon, MTIA are application-specific integrated circuits (ASICs) designed for efficiency in narrow computing tasks.
In related developments, Broadcom CEO Hock Tan, who has served on Meta's board, will transition to an advisory role. Meta also continues to pursue significant chip deals, including commitments with AMD, Nvidia, and Arm Holdings, alongside agreements for rented AI infrastructure.