Meta Platforms is expanding its custom silicon program to develop proprietary chips for complex AI model training. CFO Susan Li confirmed the company will transition from recommendation-focused chips to hardware for demanding AI workloads.
The strategic shift aims to enhance performance and operational efficiency through hardware tailored to Meta's specific requirements. This move seeks to decrease Meta's dependency on external suppliers including Nvidia and AMD.
Meta already utilizes custom silicon for ranking and recommendation workloads. Initial testing for the company's first in-house AI training chip commenced in early 2025.