【英文】Air Street资本报告：2024年AI现状报告（212页）

英文研究报告 2024年10月30日 16:23 管理员

The real large-scale GPU cluster growth has come from H100s. The largest continues to be Meta’s 350k H100s, followed by xAI’s 100k cluster and Tesla’s 35k. Meanwhile, Lambda, Oracle and Google have been building large clusters summing over 72k H100s. Companies including Poolside, Hugging Face, DeepL, Recursion, Photoroom and Magic have built over 20k worth H100 capacity. Moreover, the first GB200 clusters are going live (e.g. 10,752 at the Swiss National Supercomputing Centre), while OpenAI will have access to 300,000 by the end of next year.