Opinion: GPU Rental Prices Could Increase Over 10-Fold in the Coming Years

According to Data Beating monitoring, prominent AI podcast host Dwarkesh Patel has made a bold prediction: in the coming years, GPU compute rental prices may increase by over 10 times. The smarter the model, the more valuable the work a single GPU can perform, leading AI companies to be more willing to pay a premium for compute power.

He used software engineering as an example. If a single H100 equivalent compute power can continuously run at the level of a human software engineer, its annual rental theoretically could exceed $250,000, around 15 times the current spot price.

Signs of price increase are already visible. The one-year rental price of an H100 has risen from $1.70 per hour in October 2025 to $2.35 per hour in March 2026, nearly a 40% increase. Epoch AI estimates that global AI compute will triple by 2025, but the demand for compute from leading labs is growing even faster.

If this projection holds true, weaker models may lose their price advantage due to the higher cost of compute, and low-value image and short video generation may be the first to be phased out. However, this is just a scenario. With a large influx of AI programmers, the market price for engineering work could also decrease, casting doubt on the $250,000 figure.