JOSH TERRY
Creative Strategy and Design Operations for AI Infra @Scale, featuring Meta’s first-generation AI-focused silicon: MTIA.
The Problem
Address the challenges of attracting and retaining top technical talent in AI infrastructure, while managing media perception and securing positive coverage for Meta's investments and innovations in AI technology.
The Insight
AI infrastructure is no longer just for workloads. It’s fundamental to achieving both near-term product success and the long-term vision for the metaverse.
The Strategy
Create a technical deep dive that analyzes the architecture, performance, and learnings from Meta’s first-generation Training and Inference Accelerator.
The Creative Idea
Meta's AI workloads, crucial for various applications like content understanding and ads ranking, demand increasingly efficient hardware solutions. Traditional GPUs proved suboptimal for running specific recommendation workloads at Meta's scale. To address this, Infra and Engineering teams designed the Meta Training and Inference Accelerator (MTIA) ASICs, tailored for recommendation models. Integrated seamlessly with PyTorch, MTIA optimizes ranking systems while maintaining developer efficiency and the user experience.
In an explainer blog post, Meta gives a first-time view at how MTIA revolutionizes AI workloads, offering insights into its design, integration, and impact on future innovations.
In an explainer blog post, Meta gives a first-time view at how MTIA revolutionizes AI workloads, offering insights into its design, integration, and impact on future innovations.
The Work
This blog post was one of four outlining Meta’s latest investments in AI infrastructure during the Meta AI Infra @Scale initiative. Working closely with Engineering and Infrastructure teams, Technology Communications developed this scientific feature that housed detailed editorial, docu-style video, infographics, and architecture diagrams in a cohesive, data-driven narrative.
The moment also consisted of a virtual webcast event featuring six technical presentations inclusive of a fire-side chat with Meta executives, a content-first social campaign to bring awareness and boost engagement, and an embargoed media strategy to pre-brief a mix of business, tech and trade journalists.
The moment also consisted of a virtual webcast event featuring six technical presentations inclusive of a fire-side chat with Meta executives, a content-first social campaign to bring awareness and boost engagement, and an embargoed media strategy to pre-brief a mix of business, tech and trade journalists.
The Results
@Scale blog posts experienced a substantial increase in engagement, with nearly 30,000 views within the first 24 hours. Of those, 16,802 were for the deep dive into Meta's first-generation Training and Inference Accelerator (MTIA). The average engagement time was three minutes and 54 seconds — marking a 30% boost in average time spent reading compared to top-performing posts from the previous year.
Produced at Meta
Collaborative Partners
Karla Marccuri, Creative Lead
Jason Young, Design Lead
Patricia Geagea, Art Director
Jaime Castaneda, Front-End Developer
Julia Vargas Mondragon, Video Producer
Enrica Cavalli, Videographer + Editor
Alyssa Newcomb, Writer
Oliver Libaw, Editor
Tom Parnell, Comms Lead
Collaborative Partners
Karla Marccuri, Creative Lead
Jason Young, Design Lead
Patricia Geagea, Art Director
Jaime Castaneda, Front-End Developer
Julia Vargas Mondragon, Video Producer
Enrica Cavalli, Videographer + Editor
Alyssa Newcomb, Writer
Oliver Libaw, Editor
Tom Parnell, Comms Lead
Next Project
ART DIRECTION + DESIGNA Case for Keeping it Random