Three memory firms now set the AI price floor

The AI server bill is no longer set by the chip designer. It is set by three memory firms Nvidia cannot squeeze.

A Business Times Singapore report citing Bloomberg puts the new floor at more than 15% on AI servers shipping in early 2027, including Grace Blackwell and Vera Rubin systems. Microsoft, Alphabet's Google, and Oracle are the customers being notified now, while AI capital budgets for next year are still being finalized. The 75% gross margin Nvidia runs, the strongest in semiconductors, still cannot absorb the increase.

The reason is a tightening market for DRAM, the working memory paired with AI processors. Three firms, Samsung, SK Hynix, and Micron, supply most of it, and their components are now scarce enough to, in effect, set the price floor for what the rest of the stack charges. Apple and Qualcomm have already said they are passing chip shortages into prices. Nvidia, sitting on the most powerful pricing position in the industry, is now on the receiving end of the same mechanism.

This pattern — when a single input becomes scarce enough to set the price floor, pricing power shifts to that bottleneck — appears to be a reusable dynamic in semiconductor supply chains. In this case the captive supplier is the most profitable chip designer in the world. The lever has moved to memory.

The next data center budget will be set in Seoul and Boise, not Santa Clara.