Servers built around Nvidia’s newest AI chips are expected to cost more than 15% extra in many configurations because of an ongoing memory shortage, according to a Bloomberg report summarized by The Decoder. Systems using Vera Rubin and Grace Blackwell hardware are among those affected.
The reported increase is being driven by higher DRAM prices from major suppliers Samsung, SK Hynix, and Micron. Contract manufacturers serving Microsoft, Google, and Oracle have reportedly warned customers about higher prices on shipments planned for early next year. Nvidia had not commented on the report.
The added cost would land on cloud providers including Amazon, Microsoft, Google, and Meta, as well as AI developers such as OpenAI and Anthropic. Several of these companies are designing their own accelerators, but they still depend heavily on Nvidia systems for large-scale model training and inference.
Memory is a substantial part of an AI server, not a minor accessory. Modern accelerators need large pools of fast memory to hold model parameters and move data efficiently, so shortages can raise the price of an entire deployment. If the reported increases persist, buyers may delay installations, seek alternative configurations, or demand better utilization from existing clusters. The figures remain reported expectations for upcoming shipments rather than confirmed universal price changes.