Morgan Stanley estimates that the ROIC for GPU leasing, operating the API with proprietary computing power, and renting third-party computing power are approximately 31%, 46%, and 25%, respectively. However, this is not the company-level return on investment already achieved by cloud vendors, but rather an incremental model after 1GW of GB300 capacity has entered stable operation. The return is based on utilization, pricing, and the cost of building computing power per unit: the baseline scenario assumes approximately 410,000 GPUs, 75% utilization, and a rental fee of $8.5 per hour; if the rental fee is between $7 and $10, the ROIC for GPU leasing will fluctuate between 23% and 39%.
