AI compute pricing is moving up. Nebius, the Europe-headquartered AI cloud company, reported third-quarter results on November 13 showing GPU hour prices up more than 20 percent quarter on quarter, with annual contract value on newly signed deals exceeding 25 million dollars per megawatt, above the 20-to-25-million range of the previous quarter. Quarterly revenue grew 571.3 percent year on year to 980.8 million dollars, with the AI cloud business contributing about 99 percent of revenue, up 643 percent.
Beyond price, the shape of new contracts signals where bargaining power sits. Management said on the earnings call that customers are committing to longer terms, higher prices and larger prepayments; the company signed AI cloud agreements worth billions of dollars in the quarter, with enterprise buyers including Databricks, Datadog, Snowflake and Palantir on the list. Elastic hourly rental is giving way to multi-year, locked pricing, as buyers pre-purchase future capacity certainty.
Delivery constraints are closing in, though. Morgan Stanley kept its hold rating and 144-dollar price target after the print, flagging that strong AI demand has a flip side: data center delivery delays. Power, land and supply chain timelines are becoming the real bottlenecks on the supply side.
GPU rental pricing is the most direct thermometer of the AI supply chain: closer to real consumption than chip futures, and more market-driven than list prices from hyperscalers. When per-megawatt contract value, prepayment ratios and contract lengths all rise together, pricing power over scarce compute is shifting from buyers to sellers. Whether that transmits to compute rental prices in China is worth watching.