**[Viewpoint: GPU Rental Prices May Increase Over 10x in the Coming Years]** According to monitoring by Beating, a well-known AI podcast host, Dwarkesh Patel, has put forward an unconventional prediction: GPU computing power rental prices may increase more than 10 times in the coming years. The smarter the model, the more valuable the work a single GPU can accomplish, and AI companies will be more willing to pay high prices to secure computing power. He uses software engineering as an example. If the computational power of a single H100 GPU can continuously operate at the level of a human software engineer, its annual rental price could theoretically exceed $250,000, approximately 15 times the current spot price. Signs of price increases are already emerging. The one-year rental price of an H100 GPU rose from $1.70 per hour in October 2025 to $2.35 per hour in March 2026, an increase of nearly 40%. Epoch AI estimates that global AI computing power will grow approximately threefold by 2025, but the demand for computing power from leading laboratories is growing even faster. If this projection holds, weaker models may lose their cost advantage due to consuming more expensive computing power, and low-value applications like image and short video generation may be the first to be squeezed out. However, this is just a scenario analysis. With the emergence of a large number of AI programmers, the market price for engineering work may also decline, meaning the $250,000 calculation might not hold true. [Original Link]