Meta Unveils Intelligent Delivery to Allocate Ad Compute More Efficiently

At a glance

Meta has introduced Intelligent Delivery, an AI system that decides where to spend computing power in its ad recommendation pipeline. It was built because ad models are growing faster than the hardware that runs them.

Key points

1. 1

   The same extra compute can have very different effects depending on the ad request, model, or chip.

2. 2

   The system uses AI to predict, for each request, where extra compute will most improve overall performance.

3. 3

   Meta says it has added about 10% to ad ranking performance since 2017, on top of model and system upgrades.

4. 4

   It also moves resources across stages and data centers and uses AI agents to detect and respond to system changes.

Why it matters

As AI services demand more computing, getting better results from the same resources affects both cost and speed.