Valuation of GPU Pool

Basic Asset Evaluation

The performance of GPU devices varies, and while some devices demonstrate higher performance metrics, comparing devices specialized for deep learning and machine learning with general high-performance gaming devices can be challenging. Therefore, evaluating basic assets based on the market price and demand of the devices is the most effective method. The hourly rental cost for each GPU device model should be calculated using the following formula:

• Current market price of the GPU device (P): The purchase price of a single device

• Quantity of rental devices (N): The total number of GPU devices to be rented

• Rental hours (H): The total rental hours

• Expected annual return (R): The expected annual return rate

Hourly rental cost = Device price * (1 + expected annual return rate / 100) * quantity of rental devices / 12 * 720

Bidding

GPU providers can adjust the expected annual return rate to measure their GPU's hourly rental cost. If they aim for an annual return rate of 100%, they should set the R value to 100 in the rental cost formula. If they wish to lower the rental cost to attract more clients, they can set the expected annual return rate below the market average. Conversely, if demand increases, providers can raise the expected annual return rate to claim a fair usage fee. This fair bidding approach helps balance supply and demand, finding a reasonable equilibrium through competition.

Value of the GPU Pool

The total value of the GPU pool can be estimated based on the hourly rental cost agreed upon through bidding. For example, if the A100 model in the GPU pool is priced at $2 per hour and there are 100 units available, the hourly value of the GPU pool is $200. Annually, this amounts to $1,752,000.

While the market price of the A100 is fixed, there can be a discrepancy between its actual value and market price. However, by measuring value based on the hourly rental cost through bidding, a more accurate and flexible valuation of the GPU pool is possible. Using the annual production value of the GPU pool as a basis, the return on investment (ROI) for STO investors can be calculated. Additionally, by incorporating the average monthly utilization rate, the reliability of the GPU pool’s annual production value and ROI can be assessed.

Annual production value of the GPU pool = (Hourly rental cost of GPU device * 24 * 365) * average monthly utilization rate

The annual production value of the GPU pool is updated daily at 9 AM. The hourly rental cost average is extracted to calculate the average production of the GPU pool over the last three months and annualized for valuation purposes.