Xin Lang Cai Jing·2026-03-02 05:18Core Insights - Cathy Wood, known as "Wooden Sister," emphasizes the increasing competition faced by Nvidia, predicting that custom AI chips may capture over one-third of the computing market by the end of this decade [1][3]. Group 1: Market Predictions - Ark Invest's Director of Next Generation Internet, Frank Downing, forecasts that by 2030, more than one-third of the computing market will consist of custom chips, defined as non-GPU chips that serve as alternatives to Nvidia and AMD products [1][3]. - Downing notes that the boundaries within the industry are becoming "blurred," indicating a shift in market dynamics [3]. Group 2: Competitive Landscape - Downing highlights that while Google's TPU is well-known, Amazon is emerging as a significant competitor, referred to as a "sleeping giant" [4]. - A chart shared by Downing illustrates that traditional server market share is rapidly being consumed by accelerated computing, with Application-Specific Integrated Circuits (ASICs) expected to rival GPUs by 2030 [4]. Group 3: Strategic Partnerships - Prior to Wood's comments, Amazon and OpenAI announced a long-term partnership, with Amazon committing to invest up to $50 billion and expanding existing computing agreements to $100 billion over the next eight years [5][6]. - OpenAI is set to utilize Amazon's Trainium custom chips, including a next-generation version expected to launch in 2027, consuming approximately 2 gigawatts of Trainium computing power [6]. - Google continues to position its Tensor Processing Units (TPUs) as alternatives to Nvidia's GPUs, with reports indicating that Meta has agreed to lease TPUs for advanced AI development [6].