META PARTNERS WITH ARM TO POWER AI RECOMMENDATIONS ON FACEBOOK AND INSTAGRAM

Meta Platforms is deepening its investment in custom computing with a landmark agreement that brings Arm Holdings into the core of its data center strategy.

- 

The partnership will integrate Arm's Neoverse architecture across Meta’s recommendation and personalization engines—the Al systems that determine what billions of users see on Facebook, Instagram, and Threads each day.

By reengineering its infrastructure around Arm chips, Meta aims to cut energy costs, improve inference performance, and reduce reliance on traditional x86 processors from Intel and AMD. The collaboration also expands Arm's presence in hyperscale cloud environments, marking one of the architecture’s most significant wins in large-scale Al computing.

META’S NEXT DATA

INFRASTRUCTURE EVOLUTION

Meta will deploy Arm-based compute clusters in upcoming data centers, including a new $1.5 billion facility in Temple, Texas, that joins its global network of nearly 30 sites. The company expects to achieve substantial gains in performance per watt as it migrates Al inference workloads—responsible for ranking feeds, recommending content, and running ad-optimization models—to the Arm platform.

Internally, Meta’s engineers are rebuilding core components of its Al software stack for Arm, including PyTorch, FBGEMM, and various internal frameworks used for training and serving recommendation models. The two companies are co-developing compiler optimizations and runtime libraries, much of which will be contributed back to open source.

That contribution aligns with Meta’s long-term strategy: reducing dependence on proprietary ecosystems while fostering innovation in open Al infrastructure. Similar to its previous efforts with PyTorch, the company views open-sourcing as a way to attract developers and partners while ensuring flexibility across future chip generations.

WHY ARM'S EFFICIENCY MATTERS