Nvidia plans to increase prices for servers containing its artificial intelligence chips by more than 15 percent in many cases, according to communications the chipmaker sent to major customers including Microsoft, Google and Oracle.
The price adjustments will affect systems equipped with Nvidia's Vera Rubin and Grace Blackwell AI chips shipping early next year. The specific increase will depend on chip generation and memory configuration.
The move reflects sharply rising costs for high-bandwidth memory (HBM), which Nvidia must source from suppliers facing rare demand. SK Hynix, Samsung and Micron have already committed their 2026 production capacity, and both SK Hynix and Samsung raised HBM3E prices by nearly 20 percent before 2026 began.
The semiconductor industry is experiencing what analysts call "RAMageddon"—a severe supply crunch in the DRAM market. Conventional DRAM contract prices jumped between 90 and 95 percent quarter-over-quarter in the first quarter of 2026, with further increases of 58 to 63 percent projected for the second quarter.
Memory suppliers are reallocating production capacity toward HBM and specialized server products to meet surging AI demand. HBM production is inherently more resource-intensive than conventional DRAM, consuming approximately four times the wafer area for equivalent memory capacity.
The scale required by modern AI systems is enormous. Nvidia's Rubin GPU ships with up to 288GB of HBM4 per package. A single NVL72 rack-scale system combining 72 of these GPUs integrates more than 20TB of HBM into one rack, before accounting for the LPDDR memory attached to its Vera CPUs. Memory is now one of the largest line items in an AI server's bill of materials.
For cloud providers, these increased server costs translate directly into higher capital expenditure for AI infrastructure buildouts, pressuring margins and forcing harder choices on capital allocation as they weigh compute costs against AI service revenue potential.
Nvidia's pricing power rests on its dominant position in the AI chip market, where its proprietary GPU architecture remains essential for training and deploying large language models. The company faces a structural constraint: the demand for its chips that drives its market position also fuels the memory supply crisis now forcing it to raise prices.
