Meta has unveiled its latest iteration of custom-made chips designed specifically for artificial intelligence workloads. The new version offers improved performance compared to the previous generation, supporting Meta's ranking and recommendation ads models on Facebook and Instagram. The company views these chips as part of its growing investment in AI infrastructure, enabling the delivery of new and enhanced experiences across its apps and technologies.

Last year, Meta introduced its Meta Training and Inference Accelerator (MTIA), its first-generation AI inference accelerator tailored to Meta's AI workloads. The new MTIA iteration, as part of Meta's full-stack development program, provides a significant boost in compute and memory bandwidth while efficiently serving ranking and recommendation models for high-quality user recommendations. Additionally, Meta has revised its approach to dealing with AI-generated manipulated media, expanding its labeling of content "Made with AI" on platforms such as Facebook, Instagram, and Threads.