Loading page…
The reported increase would reach systems based on Vera Rubin and Grace Blackwell. The key variable is not only which Nvidia generation a buyer selects, but how much memory is paired with it.
Listen to this story
AI-server buyers may face materially higher Nvidia system bills early next year, with reported increases exceeding 15% in many configurations. The pressure is attributed to constrained, increasingly expensive DRAM—the memory paired with Nvidia accelerators—not to a newly announced Nvidia list price. Systems using Vera Rubin or Grace Blackwell could see different adjustments depending on chip generation and memory capacity.
Contract manufacturers reportedly notified customers including Microsoft, Google, and Oracle about forthcoming increases.
The reported threshold is above 15% in many cases, not a uniform surcharge across every server.
Samsung Electronics, SK Hynix, and Micron dominate DRAM production cited in the report.
Some of Nvidia’s largest customers have reportedly been told that prices for AI-chip servers will rise by more than 15% in many cases. The increases, tied to higher memory-chip costs, are expected to affect systems shipped early next year, putting a critical component alongside the accelerator itself in the economics of an AI server purchase.
The change has not been publicly confirmed by Nvidia. Bloomberg News attributed the information to people familiar with nonpublic customer communications; Reuters said it could not immediately verify the report. Nvidia representatives did not respond to requests for comment.
Nvidia’s accelerator processors are central to computers that create and run AI software, but their effectiveness depends on the dynamic random access memory, or DRAM, paired with them, Fortune reported. DRAM is the working memory used by a system while it operates. Fortune attributed the current price pressure to surging demand for AI infrastructure and memory supply that has not caught up; it identified Samsung Electronics, SK Hynix and Micron Technology as accounting for most global DRAM production.
That connection explains why a memory-cost increase can change the price of a completed server rather than merely the price of an adjacent component. Bloomberg’s sources said the higher memory costs were driving the reported increases. CNBC likewise described memory chips as essential to Nvidia’s GPUs and systems.
The reported increase is not a single universal surcharge. Its size will depend on the Nvidia chip generation and the memory configuration, according to people familiar with the process. That means a buyer cannot infer the final increase from the processor family alone.
The systems named in the report include configurations using Nvidia’s Vera Rubin and Grace Blackwell chips. Contract server manufacturers serving large data-center operators, including Microsoft, Google and Oracle, recently notified customers about the forthcoming increases, according to the people cited by Bloomberg.
The reported pass-through comes as major buyers are still building data-center capacity around Nvidia hardware. Amazon, Microsoft, Google and Meta are pursuing in-house chip programs, Fortune reported, while remaining dependent on Nvidia purchases for their data-center build-outs. Access to memory supply is therefore relevant both to the price of Nvidia-based servers and to how quickly customers can pursue alternatives.
The immediate unresolved issue is the exact pricing by system and configuration, since the report describes a threshold reached in many cases rather than a published price schedule. Buyers of affected systems may see memory configuration become a more consequential part of their server quote.
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.