The U.S. market for leasing AI computing power is marked by a stark contrast between short-term and long-term costs. The annual spot leasing price for short-term commitments hovers between $40 billion and $50 billion per gigawatt, doubling the long-term contract price of $20 billion per year. For supercomputing cloud operators, the breakeven point stands at roughly $12 billion per gigawatt annually. According to Goldman Sachs' analysis, hyperscale cloud providers such as Google opt for high-priced leased computing power mainly because their self-built capacities are inadequate. However, once their in-house capabilities are sufficient, the premium on spot leases will disappear.
