NVIDIA GPU Rental Rates Soar as AI Models Optimize Compute Footprint

GPU rental rates are on the rise for NVIDIA's high-end graphics processing units (GPUs), including the A100, H100, and B200 models. According to data from Silicon Data, hourly rental rates for these GPUs have increased significantly since January 2026. The H100 GPU, which was previously renting for just under $2 per hour in January, is now approaching $3. Meanwhile, the B200 GPU, which was going for around $5 per hour at the start of the year, is now hovering between $5.50 and $5.80.

This trend suggests that AI models are becoming more efficient to run on existing compute footprints, allowing older NVIDIA GPUs to retain their value. In fact, one theory behind this price appreciation relates to the Jevons paradox-type paradigm, where increased adoption of a new technology leads to proportionally higher uptake.

For hyperscalers like Microsoft, this development means that they can comfortably extend the estimated useful life of their NVIDIA GPUs from 15 to 25 years. This will reduce the quantum of GPU cost expensed each year via depreciation, giving an additional boost to their net income.