Nvidia’s AI boom has made it the industry’s price-setter. Now, a shortage of the memory that feeds its chips is forcing the chip giant’s customers to absorb a steeper bill.

The warning first surfaced as a broader sign that AI inflation was spreading beyond consumer electronics. Nvidia had already raised prices on gaming GPUs, while the companies assembling servers for Oracle, Microsoft and other cloud operators faced higher component costs. One report said “some of Nvidia’s biggest customers” had been told server prices would rise by more than 15%.¹

Those increases are expected to hit systems shipped early next year, affecting servers built around Nvidia’s flagship Vera Rubin and Grace Blackwell platforms. The final increase will vary by chip generation and memory configuration, according to people familiar with the communications.²

The pressure point is DRAM. Samsung, SK Hynix and Micron dominate production of the memory needed to pair with Nvidia’s accelerators, but expanded output has not caught up with the AI infrastructure surge. That shortage has handed memory makers unusual leverage—even over Nvidia, whose 75% gross margin and still-scarce chips have long insulated it from much of the sector’s cost pain.²

For cloud giants, the move complicates already expensive data-center plans strained by project delays, labor shortages, tighter capital markets and local resistance. Amazon, Microsoft, Google and Meta are pursuing in-house chips, yet remain dependent on Nvidia for their near-term build-outs.²

The immediate question is whether customers can secure enough memory to keep expanding—or whether the cost squeeze gives rivals their first meaningful opening. Nvidia, which did not respond to requests for comment, is expected to face renewed scrutiny when it next reports earnings.²