Meta and Arm have formed a multi-year strategic partnership focused on optimising AI software and infrastructure for enhanced power efficiency across diverse computing platforms. Meta plans to migrate its AI ranking and recommendation systems, which serve billions of users on platforms like Facebook and Instagram, to Arm's Neoverse platform. This shift aims to leverage Arm's architecture to achieve higher performance with reduced power consumption compared to traditional x86 systems.
The collaboration includes optimising Meta's core AI software, such as PyTorch and its Executorch runtime, for Arm architectures. These optimisations, incorporating Arm's KleidiAI technology, will be contributed to open-source projects. This makes them accessible to the wider developer community. The partnership seeks to improve AI application performance and streamline model deployment across edge devices and cloud infrastructure.
By optimising AI frameworks and large language models, the collaboration strives to deliver efficient and intelligent experiences to users globally. Meta anticipates spending up to $72 billion on AI infrastructure, making power consumption a critical factor. The move to Arm-based platforms is expected to yield significant savings through improved energy efficiency.
