Xin Lang Cai Jing·2026-03-05 14:58Core Insights - Meta Platforms is seeking to expand its custom chip development efforts to support artificial intelligence model training [1][2] - The company has already utilized self-developed chips for tasks such as content rating and recommendations, with plans to extend these efforts to more complex AI models [1][2] Group 1 - The CFO, Susan Li, indicated that some workloads at Meta have highly customized characteristics, making in-house chip production a viable option [1][2] - Custom chips may enhance the performance of AI systems executing tailored recommendations and other large computational tasks [1][2] - This initiative aims to reduce Meta's reliance on third-party chip suppliers like NVIDIA and AMD, who are currently the largest players in the AI hardware industry [1][2] Group 2 - Meta has reportedly begun testing its first internally manufactured AI training chip in early 2025 [1][2] - The company initially planned to expand generative AI workloads into more areas by 2026, but encountered issues with certain complex chip designs [1][2] - This move is part of a broader trend among large tech companies to reduce AI infrastructure costs and gain better control over performance through in-house chip development [1][2]