Goldman Sachs internet analyst Eric Sheridan noted in an October 10 report that the U.S. AI hashrate leasing market exhibits a "short-term expensive, long-term cheap" characteristic. Short-term spot leasing generates an annualized revenue of approximately $40 billion to $50 billion per gigawatt, which is double that of long-term contracts ($20 billion to $30 billion per gigawatt), and 2 to 4 times the break-even threshold for hyperscale cloud providers' self-built capacity (approximately $12 billion per gigawatt). The analysis suggests that cloud providers like Google are leasing hashrate at a high premium primarily because their self-built capacity cannot keep up; once capacity catches up, the spot premium will disappear. JPMorgan Chase analyst Gokul Hariharan also pointed out that current spot prices cannot generate profit through service-paying customers and are mainly used for model training, with funding sourced from equity or debt financing. The Goldman Sachs TMT team listed the credit spreads of Oracle and Broadcom as key indicators for monitoring the risk of debt-driven AI capital expenditure, with Oracle's five-year CDS reaching an all-time high.