Major events of the week:
Related to the AI sector in the US stock market：
How can Google's TPU replace NVIDIA's GPU and continue the AI narrative? Demand shifts from model training to model inference.
Major model developers like Meta have started using Google chips. To be honest, the previous narrative dominated by NVIDIA has become outdated (or overvalued). The market needs a new narrative to speculate on. Google’s specialized chips offer cost advantages over NVIDIA’s chips, as NVIDIA’s chips, while versatile and highly functional, are expensive. In contrast, Google’s chips can be specifically tailored for products.
For instance, Sora previously incurred a loss of $100 for generating each short video. This led Sora to switch to Google (Google is also more cost-effective in terms of power consumption amid the electricity shortage in the US). However, this is not a zero-sum game; it merely indicates that NVIDIA’s dominant position has been challenged.
China surpasses closed-source US artificial intelligence models in market share within the open-source AI domain.
MIT: Downloads of China’s open-source AI models have risen to 17% globally, surpassing the 15.8% market share held by developers such as Google, Meta, and OpenAI.
In reality, the biggest profit driver for current US AI companies lies in stock appreciation and premium valuations. The ecosystem of China’s open-source models has impacted the financing logic of US AI companies, primarily due to the emergence of low-cost models like DeepSeek, which affects capital expenditure > construction of computing power centers >...