This blog post discusses Intelligent Delivery, an ML-driven system designed to optimize resource allocation across ad serving architectures. It addresses challenges in maximizing efficiency due to the growing complexity of ad ranking models and their infrastructure. With a focus on machine learning and AI technologies, the system achieves a ~10% improvement in ad ranking performance by intelligently directing compute resources based on demand. The post also details the architecture, capabilities of Intelligent Delivery, and ongoing efforts to improve real-time optimization and GPU utilization for large models.