According to Silicon Data, the rental prices for A100, H100, H200, and B200 in the non-Hyperscaler market are approximately $1.65, $2.72, $3.29, and $5.61 per GPU hour, respectively. The H100 has rebounded significantly from $2 at the end of last year, while the B200 and H200 have also strengthened concurrently. Based on a 36-month forward rental curve, the estimated residual value of the H100 has risen from $14,000 in October last year to about $20,000 currently (residual value refers to model valuation, not actual transaction price). The rental trend supports Jensen Huang's logic of assetizing computing power: once GPUs stably generate inference rental income, they can be financed through future cash flow collateral, similar to airplanes and ships. On August 10, NVIDIA, in collaboration with Apollo, BlackRock, and others, promoted over $500 billion in AI infrastructure financing, with some loans secured by equipment. NVIDIA may provide up to 25% guarantees based on residual value. If this model is established, it will reduce Neo-cloud financing costs and stimulate GPU demand. However, it is important to note that Silicon Data's own second-hand market data shows that the listing price of a 3-year-old H100 is only 20%-30% of the peak price of new models, indicating significant technological depreciation. The stable rental income from the A100 over six years only suggests that older models still hold value in low-cost scenarios such as inference and fine-tuning. A more accurate assessment is that the economic lifespan of GPUs may exceed the previously assumed 2-3 years, with strong inference demand offsetting some technological depreciation, which is crucial for the collateral valuation in the $500 billion financing plan.