Meta Introduces Intelligent Delivery for Ad Ranking Systems

On November 12, 2026, Meta Engineering published an introduction to Intelligent Delivery, an optimization scheme for its ads ranking system. The post describes a shift in how Meta approaches ads ranking as frontier systems move toward larger models, broader candidate sets, and richer delivery architectures. In that context, compute resources do not scale at the same pace as model and system complexity, so the central challenge is no longer only how to build more capable models but how to operate them efficiently across the serving stack.

To address this, Intelligent Delivery adds a machine learning driven adaptive optimization layer that sits inside the system. Rather than applying a fixed amount of compute to every request, the layer dynamically allocates available resources by evaluating which ad requests, models, and chips stand to gain the most from an additional unit of compute, and then steering that compute to those points. Meta Engineering states that, building on multiple prior rounds of system evolution, the scheme delivers an additional performance gain of roughly 10% for ads ranking.

At the time of publication, the engineering post is the only available description of the work, and no further details on rollout plans or subsequent progress have been disclosed in the available material.