Nvidia isn't eating the AI server price hike anymore. Microsoft and Google are

A memory chip shortage just forced Nvidia to do something it usually avoids: pass the bill straight to its biggest customers.

Aliteq
Ravi Malhotra · Hardware Editor

The short version

Nvidia has warned server makers supplying Microsoft, Google and Oracle that Vera Rubin and Grace Blackwell AI server systems will cost more than 15% extra on units shipping in early 2027.

The short version

The cause is a DRAM and HBM memory shortage, not a change to Nvidia's own chip pricing — server-grade DRAM roughly doubled in price during Q1 2026 alone.

The short version

Samsung, SK Hynix and Micron control most global DRAM production and are reportedly using that leverage to push combined DRAM, NAND and HBM prices up 80-90% quarter over quarter.

The short version

Nvidia runs at roughly a 75% gross margin and is choosing to pass the memory-cost spike through to hyperscalers rather than absorb it itself.

The short version

Deloitte doesn't expect meaningful new memory capacity until 2029-2030; Gartner expects the shortage to persist through at least the first half of 2027.

My honest take

Nobody selling you a ChatGPT, Gemini or Copilot subscription is going to email you saying 'sorry, DRAM got expensive.' It'll show up quietly instead — a couple dollars added to a subscription tier,…

Aliteq

Read the full story

Nvidia isn't eating the AI server price hike anymore. Microsoft and Google are

Read the full story on Aliteq