Goldman Sachs report reveals US AI compute rental market dynamics

On October 11, 2026, Wall Street CN reported on the current state of the US AI computing power rental market based on a report by Goldman Sachs internet analyst Eric Sheridan included in a SpaceX quarterly preview. The report indicates that the market exhibits a clear short-term expensive and long-term cheap characteristic. In terms of annualized revenue per gigawatt, short-term spot rentals are priced at approximately $40 billion to $50 billion, while long-term contracts are around $20 billion to $30 billion. This means short-term spot prices are twice as high as long-term contracts and roughly four times the breakeven threshold of about $12 billion for hyperscalers building their own capacity. Goldman Sachs analyzes that hyperscalers like Google are paying steep premiums to rent computing power primarily because their self-built capacity cannot keep up with demand in the short term, though this spot premium is expected to fade as leading companies expand their capacity. Regarding specific enterprises, SpaceX announced two new computing hosting contracts in the past quarter, informing investors that the pricing was at the high end of the historical discussion range of $30 billion to $50 billion per gigawatt. SpaceX is not an isolated case; Nebius disclosed in August that its third-quarter short-term compute contracts were priced at roughly $40 billion to $50 billion per gigawatt, and CoreWeave recently disclosed short-term contract pricing of around $40 billion per gigawatt. Goldman Sachs forecasts that SpaceX's computing scale will grow from 1.4 gigawatts in Q2 2026 to about 7 gigawatts by the end of 2027, and reach approximately 10.6 gigawatts by the end of 2028, expanding over sevenfold within four years, which demonstrates that overall industry supply is increasing rapidly.