Nvidia's AI server prices set to jump more than 15%

Some of Nvidia's largest customers have been told that servers containing its AI chips will cost more than 15% more often, as soaring memory chip prices ripple through the entire AI hardware supply chain, Bloomberg News reported Saturday, citing people familiar with the matter.

These price hikes are set to be applied to systems delivered from early next year and will affect systems with processors based on Nvidia’s leading processors, Vera Rubin and Grace Blackwell, as the report states. 

The amount of the hike The processor generation and memory configuration will determine the hike amount. on the generation of the processor and memory configuration.

According to Reuters, the information couldn’t be confirmed immediately, as Nvidia didn’t give a response after an attempt was made to contact them during non-business hours.

Recently, those companies which manufacture servers for large data centre providers, including Microsoft, Google, and Oracle, began to notify their clients about the upcoming price hikes.

The size of the problem that stands on the verge of realisation is significant. For example, Blackwell-based NVL72 GB200 racks have had prices around $2.8 million-$3.4 million, while early enquiries for Vera Rubin-based NVL72 have reached $5 million-$7 million, which means that the 15% rise will add hundreds of thousands of dollars to the price of one single system.

The driver behind the hikes is a sharp rise in DRAM prices, the memory type that determines how effectively Nvidia's accelerator chips can actually run AI workloads. Samsung, SK Hynix and Micron together produce most of the world's DRAM, and despite ramping up output, none have kept pace with surging AI-driven demand.

That imbalance has handed memory makers unusual leverage over hardware pricing across the industry, a dynamic that's also pushed Apple and Qualcomm to raise their own prices in recent months due to chip shortages.

Nvidia posted a gross margin near 75% last quarter and has typically been able to absorb rising input costs rather than pass them fully onto customers, given its dominant position in AI silicon.