AI Compute Spot Prices Reach $61 Billion Per Gigawatt
Financial analysts report a steep price inversion in AI compute rentals, with short-term spot rates far exceeding long-term contracts and sustainable investment hurdles.
Reports from Goldman Sachs and Deutsche Bank reveal a significant price inversion in the AI compute market. Short-term spot rentals for a gigawatt of capacity have reached between $40 billion and $61 billion annually, while long-term contracts floor at approximately $20 billion. This scarcity premium exists because hyperscalers cannot build data centers fast enough to meet demand.
Goldman Sachs analysts warn that these rates are unsustainable. The current spot pricing is roughly four times the $11.6 billion per gigawatt revenue hurdle required for hyperscalers to achieve a 15% return on investment. Much of this high-cost capacity is currently funded by venture-capital-backed startups for model training rather than profitable end-user inference.
Several providers are currently monetizing this peak demand. Space Exploration Technologies Corp. is managing the Colossus compute cluster at the high end of the $30-50 billion range, with some deals estimated by Deutsche Bank as high as $61 billion per gigawatt. Neocloud providers CoreWeave and Nebius Group have disclosed short-term deals between $40 billion and $50 billion per gigawatt, while Black has reported three-year contracts above $20 billion.
Analysts highlight significant duration risk in the market. Many high-value hosting deals include 90-day cancellation clauses, which could lead to price volatility as total AI compute capacity is projected to multiply through 2028.
