Skip to content
Brief

Meta Andromeda

Ad Retrieval and Ranking Inf

Delivers significantly more relevant and higher-quality ad experiences at global scale by combining advanced deep learning retrieval models with specialized AI hardware, improving both advertiser ROI and user experience.

Last updated May 11, 2026 by ATDb automated enrichment

Founded
2023
HQ
Menlo Park, California, United States
Parent
Connections
1

At a glance

Employees
10000+
Stock
META
1corporate family

About

Proprietary internal ad retrieval engine powering Meta's global advertising ecosystem across Facebook, Instagram, and Messenger

Meta Andromeda is Meta's next-generation personalized advertising retrieval system, designed to dramatically improve the relevance and quality of ads served across Meta's family of apps including Facebook, Instagram, and Messenger. The system represents a significant architectural leap in how Meta matches ads to users, utilizing a combination of NVIDIA Grace Hopper Superchips and Meta's own custom Meta Training and Inference Accelerator (MTIA) silicon to power deep learning retrieval models at unprecedented scale. Andromeda was developed to address the growing complexity of Meta's ads ecosystem, where billions of ad candidates must be rapidly ranked and filtered to surface the most relevant options for each individual user in real time. By deploying advanced retrieval architectures on specialized AI hardware, Meta achieved measurable improvements including a reported +6% gain in retrieval recall and +8% improvement in overall ad quality scores. The system began rolling out in late 2024 and reached global deployment in October 2025. Within the AdTech ecosystem, Andromeda is significant as a demonstration of how large-scale AI infrastructure investment directly translates into advertising performance gains. It positions Meta as a leader in AI-driven ad relevance technology, reinforcing its competitive moat against rivals like Google and TikTok. Because Andromeda is an internal Meta system rather than a commercial product, it is not sold to external advertisers or agencies — its value flows through improved ad monetization across Meta's owned-and-operated properties.

Business model

Internal Infrastructure / Proprietary Ad Tech

Target market

Internal Meta Platform (not externally sold)

What they offer

  • Andromeda Retrieval Engine

    Core deep learning-based system that retrieves and ranks the most relevant ad candidates from billions of options for each user query in real time

  • MTIA-Accelerated Inference

    Ad ranking inference powered by Meta's custom MTIA chips, enabling high-throughput low-latency model execution at global scale

  • Grace Hopper Integration

    Utilizes NVIDIA Grace Hopper Superchips for training and inference of large-scale retrieval models, enabling more complex and expressive ad matching

  • Personalized Candidate Retrieval

    Advanced two-tower and retrieval model architectures that surface personalized ad candidates tailored to individual user context and intent signals

Key features

+6% improvement in retrieval recall over prior systems+8% improvement in overall ad quality scoresReal-time retrieval across billions of ad candidatesHybrid hardware deployment on NVIDIA Grace Hopper and Meta MTIA chipsDeep learning-based two-tower retrieval architectureGlobal scale deployment across Meta's family of appsTight integration with Meta's broader ads ranking pipeline

Use cases

Personalized ad retrieval for Facebook News FeedInstagram Reels and Stories ad matchingMessenger sponsored content targetingAudience Network ad relevance optimizationReal-time ad candidate shortlisting for downstream ranking models

Customer segments

Internal Meta advertising business unitsMeta advertisers (indirect beneficiaries via improved ROAS)Facebook and Instagram users (indirect beneficiaries via ad relevance)

Tech & specs

Technology stack

NVIDIA Grace Hopper SuperchipsMeta MTIA (Meta Training and Inference Accelerator)PyTorch (Meta's primary ML framework)Two-tower deep learning retrieval modelsApproximate Nearest Neighbor (ANN) searchMeta's internal distributed systems infrastructureCustom CUDA kernels and inference optimizationLarge-scale embedding models

Security & compliance

GDPRCCPAMeta Platform Privacy PolicyISO 27001 (Meta corporate)

Deployment

Cloud

API

No

Corporate history
  • 2023Founded
Connection details

Explore further

3 views