Some of Nvidia’s largest customers have reportedly been told that servers using its AI chips could cost more than 15% extra, as rising memory chip prices put pressure on the AI hardware supply chain.
According to the report, the price increases will apply to systems delivered from early next year. They will affect servers using Nvidia’s latest Vera Rubin and Grace Blackwell processors.
The size of the price increase will depend on the processor generation and memory configuration.
Reuters said it could not immediately confirm the report. Nvidia did not respond to a request for comment made outside regular business hours.
Companies that manufacture servers for major data center providers, including Microsoft, Google and Oracle, have recently started informing customers about the expected price increases.
The potential impact is significant. Blackwell-based NVL72 GB200 racks have previously been priced between $2.8 million and $3.4 million, while early inquiries for Vera Rubin-based NVL72 systems have reached between $5 million and $7 million. A 15% increase could therefore add hundreds of thousands of dollars to the price of a single system.
The price hikes are mainly being driven by a sharp increase in DRAM prices. DRAM is the type of memory that helps Nvidia’s AI chips run workloads effectively. Samsung, SK Hynix and Micron produce most of the world’s DRAM, but their production has not been able to keep up with the growing demand driven by artificial intelligence.
This shortage has given memory makers more power to raise prices across the hardware industry. It has also reportedly forced companies such as Apple and Qualcomm to increase prices in recent months because of chip shortages.
Nvidia reported a gross margin of nearly 75% last quarter and has generally been able to absorb higher costs rather than pass them on to customers because of its strong position in the AI chip market.
