Arm and Meta announced a multi-year strategic partnership in October 2025 focused on optimizing artificial intelligence software. Meta plans to use processors based on the Arm architecture, particularly those integrated into Nvidia platforms for data centers. Santosh Janardhan, Meta’s Head of Infrastructure, emphasized that this collaboration will help extend innovation to more than three billion people who use Meta’s apps and technologies. Rene Haas, CEO of Arm, added that the partnership will deliver smarter, more energy-efficient intelligence from small devices to large systems.
This initiative builds on a long-standing collaboration on hardware and software. Meta is deploying a large number of Arm Neoverse cores as part of Nvidia GB200 and GB300 NVL72 systems, which combine Nvidia Blackwell graphics processors with Grace processors based on Neoverse-V2. The goal is to achieve higher performance and lower energy consumption compared to x86-based systems.
AI Software Optimization
The collaboration focuses on improving Meta’s core artificial intelligence software technologies. For example, the PyTorch framework (a machine learning framework) is now optimized for Arm, including the ExecuTorch runtime, which uses Arm KleidiAI for maximum performance per watt. This means greater efficiency for both Meta and the global open-source community.
Other optimizations include the Facebook General Matrix Multiplication (FBGEMM) library and the vLLM engine for inference in data centers. These changes leverage Arm vector extensions and performance libraries, delivering measurable gains in inference efficiency and throughput. All these improvements are contributed back to the open-source community so that developers around the world can benefit from them.
Meta is using these optimizations for its ranking and recommendation systems, which power discovery and personalization in apps such as Facebook and Instagram. Arm Neoverse makes it possible to achieve parity in performance per watt, underscoring the scalability of the Arm architecture in hyperscale environments.
Expanding AI
The partnership strengthens collaboration on software optimizations across the entire ecosystem, from the cloud to edge devices. ExecuTorch is now enhanced with Arm KleidiAI, increasing efficiency across billions of devices. Together, they make model deployment easier and improve the performance of artificial intelligence applications from small devices to the cloud.
These open-source projects are key to Meta’s artificial intelligence strategy, enabling development and deployment for applications ranging from recommendations to conversational intelligence. Both companies plan to continue expanding optimizations in the future, helping millions of developers build efficient artificial intelligence on Arm everywhere.
This alliance covers everything from megawatt-scale data centers to foundational artificial intelligence software. Together, Arm and Meta are ushering in the next era of intelligent, energy-efficient experiences for billions of people around the world. The partnership does not focus on developing custom Arm processors for Meta, but on optimizing existing platforms for better performance and lower energy costs.
Sources: theinformation.com and newsroom.arm.com
