Nvidia’s next-generation AI servers are about to get noticeably more expensive, and the culprit isn’t the GPUs. A global memory shortage is driving up DRAM costs from Samsung, SK Hynix, and Micron, and those increases are flowing straight through to systems built around Nvidia’s upcoming Vera Rubin and Grace Blackwell chips. Bloomberg reports prices will rise more than 15 percent in many cases, with the hikes hitting shipments scheduled for early next year.
Contract manufacturers have already started warning their customers. Microsoft, Google, and Oracle — all of whom rely on third parties to build their Nvidia-based servers — have been told to expect the higher bills. Nvidia itself hasn’t commented on the pricing shift, but the mechanics are straightforward: memory is a major component cost in AI servers, and when DRAM prices spike, the entire bill of materials climbs with them.
The timing is awkward for the cloud giants. Amazon, Microsoft, Google, and Meta are pouring unprecedented capital into AI infrastructure while simultaneously racing to develop their own custom silicon to reduce dependence on Nvidia. OpenAI and Anthropic are in the same boat. Yet for now, none of them can walk away. A 15 percent price increase on already-expensive AI servers lands squarely on the balance sheets of the very companies trying to break free from Nvidia’s orbit — and they’ll pay it, because the alternative is falling behind in the AI race.
This is the second squeeze in under a year for AI hardware buyers. Memory pricing has been volatile since late 2023, when AI demand began absorbing a huge share of HBM and high-capacity DRAM output. What’s different now is that the shortage is hitting mainstream server memory, not just the exotic high-bandwidth variants. That suggests the supply crunch is broadening, and it raises an uncomfortable question: if DRAM costs are already forcing double-digit price hikes on systems that haven’t even shipped yet, what happens when demand from AI labs and cloud providers scales up again later this year?
