Meta Accelerates Plans For Custom AI Training Chips

Meta is accelerating development of proprietary processors designed to train future artificial intelligence models, Meta CFO Susan Li confirmed Wednesday at a Morgan Stanley technology conference. 

The announcement shows that Meta intends to deepen its hardware sovereignty despite recently securing major supply agreements with Nvidia and AMD for chips and infrastructure that power the company's AI systems. 

Li emphasized that Meta's workloads are highly specialized, particularly in ranking and recommendation systems that drive content feeds and suggestions across Facebook, Instagram, and WhatsApp, making them well-suited for custom-built hardware. 

Meta has already deployed custom silicon at scale for these ranking and recommendation tasks, but the company now expects to expand chip development into more demanding areas, including AI model training, which typically requires massive computing power. Developing its own training chips could allow Meta to improve computational efficiency while reducing reliance on external suppliers and mitigating the soaring costs of third-party hardware in an increasingly volatile semiconductor supply chain.