Nvidia has notified some of its largest customers that server prices for AI chips including Vera Rubin and Grace Blackwell will rise more than 15% for systems shipping early next year, with increases varying by chip generation and memory configuration. The price hikes are driven by soaring memory chip costs, which are essential for GPU systems and now represent over 60% of platform cost.
Memory expenses on Vera Rubin have surged 435% versus Grace Blackwell, pushing memory allocations to 74.7 TB per rack—equivalent to several thousand smartphones' worth of DRAM. This reflects unprecedented AI infrastructure demand straining the global memory supply chain.
For data center operators and cloud providers, these increases compound capex pressure at a moment when demand for compute remains insatiable. Vera Rubin volume ramps Q4 2026, and Nvidia's pricing signals it will pass memory inflation downstream rather than absorb it, cascading through cloud and enterprise infrastructure budgets.