Zhi Tong Cai Jing·2026-03-11 23:58Core Viewpoint - Meta Platforms plans to launch four generations of self-developed AI chips by the end of 2027 to support its growing AI computing needs and reduce reliance on external chip suppliers [1][2]. Group 1: Chip Development Plans - The company will introduce the MTIA 300, MTIA 400, MTIA 450, and MTIA 500 chips, part of the "Meta Training and Inference Accelerator (MTIA)" series, aimed at internal AI training and inference tasks [1]. - The MTIA 300 is already in mass production for content ranking and recommendation model training, while the MTIA 400 has completed lab testing and is being gradually deployed [1]. - The MTIA 450 and MTIA 500 are expected to achieve large-scale deployment by 2027, with the MTIA 450 anticipated to launch in early 2027 and the MTIA 500 approximately six months later [1][2]. Group 2: AI Development and Chip Procurement - The rapid pace of AI development has necessitated an accelerated chip development timeline, with Meta currently relying heavily on external chip suppliers like NVIDIA and AMD, with procurement agreements reaching hundreds of billions [2]. - Meta has made significant investments in AI to build competitive large models and AI products, leading to unprecedented computational demands [2]. Group 3: Internal Chip Development Strategy - Self-developed chips will be optimized for specific applications, enhancing efficiency and reducing costs by eliminating unnecessary features [3]. - The chip development process is costly and lengthy, typically requiring billions in investment and several years from design to mass production [3]. - Despite challenges, including the cancellation of the advanced AI training chip project "Olympus" due to design difficulties, Meta remains committed to developing AI processors for training large AI models [3].