Meta测试自研AI训练芯片，试图减轻对英伟达依赖

Core Viewpoint - Meta is actively testing its self-designed AI training chips, marking a significant step in reducing reliance on hardware suppliers like Nvidia [3][4][5] Group 1: AI Chip Development - Meta has begun small-scale deployment testing of its AI training chips, which are specifically designed for handling AI workloads and manufactured by TSMC [3] - Previous custom AI chip projects by Meta were primarily for running models rather than training them, and many were canceled or scaled back due to not meeting internal expectations [3][4] Group 2: Strategic and Financial Implications - The self-developed AI training chips hold major strategic and financial significance for Meta, which has projected capital expenditures of $65 billion this year, largely for purchasing Nvidia GPUs [4] - Reducing even a small portion of costs through in-house chip development would represent a substantial financial victory for Meta [5] - This move reflects a broader trend among tech giants pursuing self-developed AI hardware capabilities to mitigate supply chain risks and enhance flexibility and cost advantages in their AI strategies [5]