Meta unveiled four new generations of its MTIA (Meta Training and Inference Accelerator) chips on March 11, 2026, the 300, 400, 450, and 500 models, designed for specific AI workloads at lower cost and improved energy efficiency. The company formalized a partnership with Broadcom on April 14, 2026, to co-develop these chips through at least 2029. Meta plans to double its computing capacity from 7 gigawatts in 2026 to 14 gigawatts by 2027. The MTIA chips are not meant to replace Nvidia and AMD GPUs but to complement them by targeting inference tasks where specialized hardware delivers better performance per watt. The key question remains whether Meta will deploy these chips for training workloads, which would represent a far more direct challenge to Nvidia’s core business.
Source: Read the original article
