Meta details Intelligent Delivery ad compute allocation system
Understand this story
Meta introduces Intelligent Delivery, its machine learning-driven ad compute resource allocation system, saying that from 2017 through the first half of 2026, the system delivered roughly an additional 10% improvement in ad ranking performance on top of model and system advances. The system predicts the marginal utility of compute and uses multi-layer mechanisms to allocate resources within the system. Meta says that in the first half of 2026, agents participated in about 20% of Intelligent Delivery optimization; in its most highly automated mechanism, agents handle about 80% of the engineering work, covering understanding system conditions, proposing hypotheses, running experiments, and converging on final changes. All final changes are still reviewed and approved by engineers before going live.
AI overview based on source reports · Updated 3 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Meta EngineeringIntelligent Delivery: Making Every Unit of Compute Serve Better Ads
Meta describes Intelligent Delivery, an ML-driven system that allocates compute across its ads serving stack toward a system-wide optimum, contributing roughly 10% incremental ranking performance on top of model and system advancements from 2017 through the first half of 2026. It combines a shared Resource Core Model (6B sparse and 20M dense parameters) that predicts each request's outcome, compute consumption, and latency impact, with a hierarchy of allocation engines spanning region, hardware SKU, and host-level decisions. Agentic AI now helps propose and test optimizations, contributing to about 20% of Intelligent Delivery optimizations in the first half of 2026 and about 80% of engineering effort on the most automated mechanism, with engineers approving every change.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 12 02:00
- Change in the past 24 hours
- -50%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.