SANTA CLARA, CALIFORNIA / RankWire.AI / – Nvidia is planning to implement price increases exceeding 15% on many AI server configurations scheduled for delivery in early 2027. These hikes will affect models utilizing the company’s Vera Rubin and Grace Blackwell platforms. The extent of each increase depends on the chip generation and memory configuration. Nvidia has not publicly declared a blanket price hike across all its server offerings. Additionally, server manufacturers have communicated the new pricing details to major data center clients.
Major data center operators, such as Microsoft, Google, and Oracle, have been informed of the price adjustments by server manufacturers. These corporations purchase large volumes of AI infrastructure for cloud and data center operations. The rise in memory costs has emerged as a significant factor influencing the reported increases in server prices. AI deployments demand large quantities of high-bandwidth and server memory along with GPUs, CPUs, networking gear, and storage. Throughout 2026, this heightened demand has kept several memory categories in tight supply.
TrendForce projects that conventional DRAM contract prices will increase by 13% to 18% in the third quarter of 2026. It also forecasts NAND Flash contract prices to go up by 10% to 15% during the same timeframe. The market research firm highlighted that server DRAM remains in undersupply as suppliers focus on capacity expansion for AI-related applications. While some long-term supply agreements have limited the scope of price hikes, overall server memory costs are still elevated.
Memory Costs Climb Due to Persistent AI Server Demand
In May, Nvidia announced that Vera Rubin had entered full production with server manufacturers and supply-chain partners globally. The company stated that products based on Rubin would become accessible in the second half of 2026. Vera Rubin incorporates 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 architecture designed for large AI workloads. System manufacturers are developing Rubin-based products aimed at cloud providers, AI research labs, and hyperscale data centers.
Grace Blackwell remains a key component of Nvidia’s current data center lineup. The GB200 NVL72 configuration links 36 Grace CPUs with 72 Blackwell GPUs within a single liquid-cooled rack. Designed to operate as a unified NVLink computing domain, this setup supports large AI models. Price increases reported for servers vary depending on configuration, with memory capacity and chip generation playing significant roles in pricing. Consequently, the 15% figure does not apply uniformly to all Nvidia server models.
Expansion of Vera Rubin Production in Nvidia’s Server Ecosystem
Suppliers of memory have shifted production toward server and high-performance products as AI demand continues to absorb capacity. According to TrendForce, this shift has reduced the supply of some PC and consumer DRAM products. Despite robust server shipments through 2026 and ongoing inventory accumulation by buyers, the supply of server DRAM is expected to remain tight into 2027 as demand growth surpasses new supply. These market conditions underpin the pricing increases reported by Nvidia.
Nvidia is entering this pricing phase with record-high data center sales. In its fiscal first quarter ending April 26, the company reported total revenue of $81.6 billion. Data Center revenue hit an all-time high of $75.2 billion, a 92% increase year-over-year. Nvidia forecasted second-quarter revenue of around $91 billion, plus or minus 2%, when reporting results in May. The company is scheduled to release its fiscal second-quarter earnings on Aug. 26.