Meta Plans In-House Chips for Training Advanced AI Models

Meta Platforms Inc. has big plans for its own custom silicon, even after securing major deals with leading chipmakers. The company intends to develop processors capable of training future AI models, according to its chief financial officer, Susan Li.

“Some of our workloads are very customized to us,” Li said Wednesday at a Morgan Stanley technology conference. “We started with ranking and recommendations workloads, where we rolled out custom silicon at scale. Over time, we hope to expand this, eventually including AI model training.”

While not a traditional cloud computing provider, Meta operates some of the largest data centers used to train and run AI models. Recently, it finalized massive agreements with Nvidia and Advanced Micro Devices to supply the chips and equipment powering its AI workloads. Alongside these deals, the company continues to develop its in-house AI processors.

Li explained that Meta uses different chip types for different tasks. “Given our current needs, we consider what the best chip is for each use case,” she said. “Custom silicon is a key part of that strategy.”