Server manufacturers have reportedly informed their customers that prices for some of Nvidia's flagship AI server chip systems are expected to climb by roughly 17%, according to two people familiar with the matter.

The price increase primarily targets the Grace Blackwell 300 and Vera Rubin 200 chip systems, which are scheduled for delivery next year. Based on current pricing, this adjustment would add at least $5 billion to the cost of building a 1-gigawatt data center.

Rising costs for memory and other components are cited as partial drivers behind Nvidia's price hike. Cloud service providers and other data center developers are already grappling with a wave of cost overruns, from unexpected billions in expenses at Oracle's facilities to widespread misconceptions about AI data center economics. Some U.S. states are also considering eliminating tax incentives, which would further inflate construction costs. OpenAI's Stargate data center project is likewise facing higher price tags as cost pressures mount across new builds.

It remains unclear whether cloud firms purchasing Nvidia chips will absorb some of the increase themselves or pass the expense on to end customers renting computing power. One person familiar with the situation indicated that cloud providers will almost certainly transfer the pricing pressure downstream. Meta and OpenAI are among the companies buying chips for their own AI computing operations, with Nvidia planning to offer around $100 billion in credit support for OpenAI's large Nvidia server campus in Ohio.

Following this anticipated increase, a Vera Rubin rack equipped with 72 GPUs will cost approximately $8 million. This year, Nvidia, OpenAI, and other companies developing their own AI server chips have been racing to sign long-term memory supply agreements with storage vendors to hedge against future shortages or price spikes. Nvidia's memory demand, however, far exceeds that of other AI chip makers.

Recently, Nvidia has invested in and established partnerships with three core service providers for AI data center land and power connectivity, aiming to avoid situations where chips are delivered but facilities aren't ready to run them. An Nvidia spokesperson declined to comment on the matter, though Bloomberg News previously reported on the chip price increase exceeding 15%.