SilverTorch: Index as Model — A New Retrieval Paradigm for Recommendation Systems
Meta's SilverTorch unifies recommendation retrieval into one neural network with 23.7x throughput gains
Meta is replacing fragmented microservice-based retrieval pipelines with SilverTorch, a unified neural network that treats the item index as a model tensor. The system achieves 23.7x higher throughput and 20.9x better compute cost efficiency while improving recommendation quality within sub-100ms latency budgets. Accepted to SIGIR 2026, this represents a significant systems-level rethink for large-scale recommendation infrastructure, though it is more of an MLSys engineering advance than a core AI/LLM industry signal.