深夜，英伟达重挫！

Core Viewpoint - Meta is considering using Google's Tensor Processing Units (TPUs) in its data centers by 2027, which could significantly impact the AI chip market and challenge Nvidia's dominance [5][6]. Group 1: Meta's Shift - Meta is currently reliant on Nvidia GPUs for AI computing but is exploring the rental of TPUs from Google Cloud as early as next year [5]. - The use of TPUs would mark a significant endorsement of Google's decade-long investment in this technology, potentially allowing Google to capture at least 10% of Nvidia's annual revenue, which amounts to several billion dollars [5][6]. Group 2: Google's TPU Technology - Google's seventh-generation TPU, named "Ironwood," boasts a performance increase of four times compared to its predecessor and a nearly 30-fold improvement in energy efficiency since the first Cloud TPU launched in 2018 [6]. - Major clients for Google's TPU include Salesforce, Midjourney, and Anthropic, with Anthropic planning to deploy up to 1 million TPUs for training its AI model Claude, representing a multi-billion dollar expansion [6]. Group 3: Nvidia's Market Position - Despite the emergence of competitors like AMD and Google, Nvidia remains the dominant player in the AI chip market, with its GPUs being preferred for their versatility in handling a wide range of workloads [8]. - AMD's latest MI355X chip shows competitive performance but lacks the software ecosystem and data transfer capabilities that Nvidia offers [8]. - Research indicates that while Google develops its own chips, it still relies heavily on Nvidia as a major customer due to the flexibility that GPUs provide for various algorithms and models [8].