A news summary based solely on the provided article body: Analysts at Epoch AI estimate that by the end of 2025, OpenAI will have the equivalent of about 1.7 million H100 GPUs in computing power, a 17-fold increase from 100,000 in 2023 and 400,000 in 2024. Despite this massive hoarding, Anjney Midha, founder of AMP, stated that for cutting-edge labs like xAI, GPU utilization may be less than 10%. Epoch AI analyst Josh You noted that cutting-edge labs have not used most of their AI computing power, as the speed of purchases far exceeds consumption. Fujitsu's 2024 "State of AI Infrastructure Report" found that more than 75% of enterprises have a GPU utilization rate of less than 70% even at peak load. VentureBeat made a more radical judgment based on similar data, stating that 95% of AI infrastructure spending is wasted.