Chip designer Arm and Meta Platforms said on Wednesday they have agreed to a multi-year partnership to improve the power efficiency of artificial intelligence systems used across Meta’s products and infrastructure. The collaboration will see Meta’s AI software, including its PyTorch machine learning framework and ExecuTorch runtime, optimized for Arm’s chip architecture to boost performance and lower energy use. Meta said its recommendation and ranking systems, which power Facebook and Instagram, will also use Arm’s Neoverse-based data center chips. Both companies said they have worked to fine-tune AI compilers, libraries and frameworks for Arm-based systems, with the resulting code contributed to open-source projects. The companies said the changes will help improve performance per watt in data centers and on devices running AI applications. The deal extends ongoing cooperation between Arm and Meta on hardware and software co-design, as tech companies seek more efficient ways to run increasingly power-hungry AI workloads.