深夜，英伟达重挫！

Group 1 - Meta is considering using Google's Tensor Processing Units (TPUs) in its data centers by 2027 and may rent TPU capacity from Google Cloud next year [5] - Currently, Meta relies almost entirely on NVIDIA GPUs for AI computing to serve over 3 billion daily active users [5] - Google's TPUs are designed to accelerate machine learning tasks, particularly those using TensorFlow, and have been sold to major clients like Salesforce and Midjourney [5] Group 2 - The potential deal would be a significant endorsement of Google's decade-long investment in TPUs, with the latest generation claiming a fourfold performance increase over its predecessor [6] - AI startup Anthropic has announced a partnership with Google to deploy up to 1 million TPUs for training its AI model, Claude, with a projected capacity of 1GW by 2026 [6] Group 3 - Despite NVIDIA's dominance in the AI chip market, competitors like AMD are also entering the field, with AMD's MI355X chip showing competitive performance [7] - Google’s TPUs represent a direct challenge to NVIDIA's market position, as many large AI companies are accelerating their own chip development [7] - NVIDIA remains a key player, as its GPUs can handle a wider range of workloads compared to TPUs, maintaining its status as a major supplier to Google [7]