NVIDIA AI server prices set to rise more than 15% as memory costs surge

51

TL;DR:

NVIDIA customers reportedly face AI server price increases of more than 15% as soaring memory costs hit Blackwell and Vera Rubin systems. The increases could make already-expensive AI infrastructure projects costlier from early 2027.

Article:

NVIDIA customers are being warned of price increases of more than 15% on some AI servers as sharply higher memory chip costs feed into the price of next-generation computing infrastructure. The increases are expected to affect systems shipped from early 2027, including configurations built around NVIDIA’s Grace Blackwell and forthcoming Vera Rubin platforms.

The timing matters because memory has become another constraint in the AI hardware supply chain. Demand from hyperscale data centres and AI workloads has tightened DRAM and NAND supply, giving memory suppliers stronger pricing power just as cloud operators are committing billions of dollars to new computing capacity.

TrendForce expects conventional DRAM contract prices to rise 13% to 18% quarter-on-quarter in the third quarter of 2026, while NAND Flash contract prices are forecast to increase 10% to 15%. The research firm said, “The DRAM market will remain extremely tight in the third quarter of 2026.”

The reported NVIDIA server increases will vary by chip generation and memory configuration. Server manufacturers supplying large data-centre operators have already passed pricing information to customers, according to the report. Nvidia had not publicly commented on the reported increases at the time of publication.

The cost pressure shifts the AI investment debate beyond GPU availability. Memory, storage, power and cooling are increasingly determining the total cost of deploying AI infrastructure.

That could favour the largest cloud operators, which have stronger procurement leverage and capital budgets, while making high-end AI capacity harder to finance for smaller enterprises and AI companies. If memory prices remain elevated into 2027, the economics of training and inference will face another test before the next generation of AI servers reaches scale.