Nvidia Raises AI Server Prices Over 15% Amid Memory Shortage

Nvidia Corp. is increasing the prices of its artificial intelligence servers by more than 15% in many cases, with hikes taking effect on systems shipped early next year. The price increases specifically target systems featuring Grace Blackwell and Vera Rubin chips, with final costs varying by chip generation and memory configuration. Contract manufacturers have already notified major data center operators, including Microsoft Corp., Alphabet Inc.'s Google, and Oracle Corp., of the forthcoming costs.

The price surge is driven by soaring costs for dynamic random access memory (DRAM), which has granted significant market leverage to manufacturers SK Hynix Inc. and Micron Technology Inc. This supply bottleneck was highlighted by Elon Musk, who noted that while customer demand for compute capacity is growing at 200%, unit volume for memory chips is increasing by only 20% year-over-year. Musk and Space Exploration Technologies Corp. plan to deploy 15 to 20 gigawatts of compute capacity for AI data centers by the end of 2027, a scale exceeding the peak energy demand of New York City.

While major customers such as Amazon and Meta are developing in-house chip programs to reduce dependency, they remain reliant on Nvidia for current data center build-outs. These cost increases add complexity to global AI infrastructure projects already hampered by labor shortages and capital market tightening. Investors are currently pricing in these increases ahead of Nvidia's second-quarter financial results scheduled for August 26.