According to analysis from AI research firm Epoch AI, Google likely possesses the world's largest AI computing power. This advantage stems from Google's custom Tensor Processing Units (TPUs), which offer computational capabilities equivalent to at least 600,000 Nvidia H100 GPUs. Epoch AI researchers noted: "Given Google's extensive fleet of TPUs, coupled with their NVIDIA GPUs, Google likely holds the most AI computing power of any single company."
Image source note: The image was generated by AI, provided by the image licensing service Midjourney
Despite Google's lead in overall capacity, Nvidia remains the dominant player in AI chip sales. Since early 2022, Nvidia has sold AI chips with computational power equivalent to about 3 million H100 GPUs. Most of these chips have gone to the four major cloud service providers, with Microsoft being the largest customer.
Other buyers include cloud companies like Oracle and CoreWeave, AI companies such as xAI and Tesla, Chinese tech companies, and governments building national AI infrastructure.
Epoch AI cautions that its estimates carry significant uncertainty, as the AI chip landscape is rapidly evolving with companies developing new processors and ramping up production.
Reports indicate that Nvidia's Blackwell series GPUs are fully pre-sold. According to Barron's, citing a conversation between Morgan Stanley analysts and Nvidia CEO Jensen Huang, the supply for the next 12 months of Blackwell is already sold out.
In addition to Nvidia and Google, other companies like AMD, Intel, Huawei, Amazon, Meta, OpenAI, and Microsoft are also developing their own AI chips. It is reported that China is encouraging domestic companies to purchase more chips from Chinese suppliers like Huawei.
This bar chart shows the AI computing power of leading tech companies, with Google leading due to its own TPUs. This distribution could have significant implications for the development and application of future AI models. | Image: Provided by EpochAI
As AI technology continues to advance, the competition in the AI chip market is intensifying. Google's TPUs and Nvidia's GPUs play a crucial role in providing powerful computing capabilities, which are essential for training and running complex AI models. With more companies entering this field, we can expect AI chip performance to continue improving while costs decrease, further driving the development and proliferation of AI technology.