Inside Meta’s new Andromeda: a billion ads scored per second

Retrieval is the first stage of Meta’s ad recommendation process: it narrows a huge pool of eligible ads to a small set in milliseconds before a costlier ranking stage scores them. Meta’s engineering team says the new Andromeda now scores every eligible ad at near ranking-level quality and surfaces top candidates at over one billion ads per second, which it calls a 100x lift over the 2024 system. Since 2025, Meta says, Andromeda has delivered a 2.4% increase in conversions on Facebook.

The pressure behind the rebuild is telling. Generative AI tools have helped advertisers create hundreds of millions of new ads, so AI is now generating ads faster than older systems could evaluate them.

Meta borrowed mixture-of-experts ideas from large language models, using specialized user models as experts, and claims near ranking-level quality at roughly 300x fewer inference-time FLOPs than the 2024 model. It then fused model and system work into a single GPU kernel and moved index matching and selection onto the same GPU server, which it says cut network traffic about 100x. The 2024 Andromeda launched on NVIDIA’s GH200 Grace Hopper Superchip; in early 2026 Meta and NVIDIA began co-optimizing each pipeline component, reporting 2x to 5x speedups per component. Meta says the pipeline is not at its ceiling, with INT8 quantization and Grace Blackwell work ahead.

The broader takeaway is that model, system and hardware should be designed together when workloads are memory-bound. It sits alongside Meta’s rework of how its ad system names things, swapping one-off IDs for a shared semantic vocabulary (GenRecs tokenization).

The goal is more precise ad matching at lower cost, and Meta lists the inputs: sequences of ad engagement, cohort representations, LLM-derived semantic features and engagement across Meta’s surfaces. Every efficiency gain here is an efficiency gain in modeling people’s interests for commercial persuasion. There is a fair case for users, since relevant ads beat irrelevant ones. The feedback loop deserves attention, though: generative tools flood the system with ads, a faster retrieval engine sorts them, and the whole thing gets better at finding the version that converts. The harm, if it scales unchecked, is subtle: more optimized persuasion and less visibility into why an ad found you.

It also explains the spending. Meta is creating a new share class to fund its AI push (Meta’s new shares), and the ad engine is where that investment pays. More in our Meta coverage and generative AI coverage.