SANTA CLARA, CALIFORNIA / RankWire.AI / – Nvidia is planning to implement price hikes exceeding 15% on many of its AI server configurations scheduled for shipment in early 2027. These adjustments will impact systems built on the Vera Rubin and Grace Blackwell platforms, with the magnitude of each increase depending on the chip generation and memory configuration. The company has not issued a blanket announcement for a universal price hike across all server products, but major data center clients have been informed of the new pricing through server manufacturers.
Major data center operators such as Microsoft, Google, and Oracle have received these pricing updates from server manufacturers. These clients purchase large volumes of AI infrastructure for their cloud and data center services. A significant factor driving the price increases is the rising cost of memory, which constitutes a major input in server pricing. AI deployments require substantial amounts of high-bandwidth memory along with GPUs, CPUs, networking gear, and storage, keeping several memory categories in tight supply during 2026.
TrendForce forecasts that contract prices for conventional DRAM will increase by 13% to 18% in the third quarter of 2026. The same research predicts NAND Flash contract prices will go up by 10% to 15% during the same period. They noted that server DRAM remains undersupplied as suppliers prioritize capacity for AI applications. While long-term supply agreements have restrained some price hikes, overall server memory costs continue to stay high.
Memory costs climb as demand for AI servers remains elevated
In May, Nvidia announced that Vera Rubin had entered full-scale production with server manufacturers and supply-chain partners globally. The company stated that products utilizing Rubin would be available during the second half of 2026. Vera Rubin integrates the Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, and Spectrum-6 networking technologies. It replaces Grace Blackwell as Nvidia’s latest rack-scale platform optimized for large AI workloads. System manufacturers are developing Rubin-based offerings aimed at cloud providers, AI laboratories, and hyperscale data centers.
Meanwhile, Grace Blackwell continues to play a significant role in Nvidia’s current data center portfolio. The GB200 NVL72 configuration links 36 Grace CPUs with 72 Blackwell GPUs within a single liquid-cooled rack. Designed to function as a unified NVLink computing domain, it supports large AI models. The reported price increases for servers vary based on configuration, with memory capacity and chip generation being key factors. As a result, the 15% figure does not indicate a uniform increase across all Nvidia server models.
Expansion of Vera Rubin production across Nvidia’s server range
Memory suppliers have shifted more production capacity toward server and high-performance products to meet AI demand, which has absorbed much of their output. According to TrendForce, this shift has reduced supply for certain PC and consumer DRAM products. Despite strong server shipment volumes through 2026 and ongoing inventory buildup by buyers, the market researcher anticipates that server DRAM supply will remain constrained into 2027 as demand growth surpasses new supply. These market dynamics underpin the pricing adjustments announced by Nvidia.
Nvidia approaches this pricing phase with record-breaking data center sales, reporting a total revenue of $81.6 billion in its fiscal first quarter ending April 26. Data Center sales hit an all-time high of $75.2 billion, representing a 92% increase year-over-year. The company projected a second-quarter revenue of $91 billion, with a margin of error of plus or minus 2%, during its May earnings announcement. Results for the fiscal second quarter will be disclosed on August 26.
}{