谷歌AI芯片获大单:Anthropic将使用100万个TPU训练大模型
Feng Huang Wang·2025-10-23 23:06

Core Insights - Anthropic's Claude model will utilize up to 1 million Google AI chips for training, valued at several billion dollars, aiming to enhance performance in the rapidly evolving AI sector [1] - Google, as an investor in Anthropic, will also provide additional cloud computing services, highlighting the significant demand for computing power in training, deployment, and ongoing inference of generative AI products [1] Company Developments - The transaction coincides with Google's expansion of the availability of its proprietary Tensor Processing Units (TPUs), which were previously used mainly for internal purposes [1] - Google is currently renting out TPUs through its cloud services, indicating a strategic move to monetize its AI hardware capabilities [1] Technology and Efficiency - Anthropic selected TPUs due to their high cost-effectiveness and superior efficiency, supported by the company's prior experience in training and deploying models using these processors [1]