NVIDIA AI chip prices may rise by about 17%, server manufacturers notify customers

NVIDIA CEO Jensen Huang showcased the Vera Rubin chip system this year.

According to sources from two companies, server manufacturers have notified their clients that the prices of some flagship AI server chip systems from NVIDIA are expected to rise by approximately 17%.

The price increase mainly targets the Grace Blackwell 300 and Vera Rubin 200 chip systems, which are scheduled for delivery next year. Based on current chip system quotes, the cost of building a 1 gigawatt-scale data center will increase by at least $5 billion after the price hike.

Sources indicate that the rising costs of memory and other components are part of the reasons for NVIDIA's price adjustments. Cloud service providers and other data center builders are facing various cost overruns: Oracle's data center has encountered unexpected costs in the billions, there are many economic misconceptions regarding AI data centers circulating in society, some states in the U.S. plan to eliminate tax incentives, raising construction costs, and OpenAI's Stargate data center project is also facing rising costs, with various cost pressures piling up for new projects.

It is still unclear whether cloud vendors purchasing NVIDIA chips will absorb some of the price increase costs themselves or pass the costs onto end customers renting computing power. A person familiar with the price increase situation stated that cloud service providers will almost certainly pass the price increase pressure onto downstream customers.

Companies like Meta and OpenAI purchase chips for their own AI computing businesses. (NVIDIA plans to provide approximately $100 billion in credit support for OpenAI's large NVIDIA server park in Ohio.)

After this anticipated price increase, a rack equipped with 72 GPUs of the Vera Rubin system will cost about $8 million.

This year, companies like NVIDIA and OpenAI, which develop their own AI server chips, are competing to sign long-term memory supply agreements with storage suppliers to avoid risks of future supply shortages or price surges. However, NVIDIA's demand for memory far exceeds that of other AI chip manufacturers.

Recently, NVIDIA has completed investments and established partnerships with three core service providers for AI data center land and power access. This move aims to prevent situations where chips are delivered but the supporting facilities are not ready, preventing the chips from being operational.

A spokesperson for NVIDIA has not yet commented on this. Bloomberg News previously reported on the chip price increase of over 15%