Nvidia sets the terms for nearly everyone in AI hardware. It turns out there's an exception: the handful of companies that make the memory chips its own products depend on.
Some of Nvidia's biggest customers have been told that server prices containing its AI chips are rising more than 15% in many cases, according to people familiar with the notifications. The increases take effect on systems shipping in early 2027.
The Actual Numbers
The price hikes hit systems built around Nvidia's flagship Vera Rubin and Grace Blackwell chips, with the exact increase depending on chip generation and memory configuration. Companies that build servers under contract for data center operators including Microsoft, Google and Oracle have already notified their own customers.
Nvidia didn't respond to requests for comment.
The company also recently raised prices on its consumer gaming GPUs, with Tom's Hardware reporting increases as steep as 39% on some RTX 50-series cards.
Why Nvidia Can't Just Absorb the Cost
Nvidia's chips need dynamic random access memory, DRAM, to actually function, and Samsung, SK Hynix and Micron control most of the world's supply of it. Demand has outrun their production increases badly enough that memory makers now hold real pricing power over a company that otherwise dominates its own market almost completely.
That's a notable shift. Nvidia runs a 75% gross margin and charges tens of thousands of dollars per chip precisely because TSMC's manufacturing supply can't keep up with demand.
Even that level of market control wasn't enough to let Nvidia eat rising memory costs without passing them along. AI accelerators originally grew out of PC gaming chips that sold for a few hundred dollars each. The gap between that starting point and today's pricing shows how much room demand alone has created for every link in the supply chain to extract more value, not just Nvidia itself.
This is the receipt for the claim MAIN covered from Micron's CEO Sanjay Mehrotra this week, that AI demand has permanently changed memory pricing power. Nvidia being forced into a 15% price hike rather than absorbing the cost itself is exactly the kind of evidence that argument needed.
This Isn't Only an Nvidia Problem
Apple and Qualcomm have both said recently they're raising prices because of chip shortages too. Amazon, Microsoft, Google and Meta are all building their own in-house AI chip programs to reduce Nvidia dependence, but every one of them still needs memory from the same three suppliers regardless of whose logic chip it pairs with.
The timing adds pressure to an already complicated AI data center buildout. Project delays, labor shortages, tightening capital and community pushback have already slowed plans in multiple states, and rising hardware costs on top of that squeezes the economics further.
Nvidia reports fiscal second-quarter earnings next week, a closely watched signal for how the rest of the AI infrastructure trade is holding up.
What This Means for Miami
This is the concentration risk J.P. Morgan's Bill Eigen has been warning Miami investors about all year, now showing up as a real cost increase rather than a hypothetical.
A handful of memory suppliers now have enough leverage to raise the price of the entire AI hardware stack, regardless of how dominant the chip designer sitting above them is. That's precisely the kind of single-point-of-failure risk Eigen has argued gets underpriced when investors focus only on the biggest, most visible name in a supply chain.
For Miami's data center developers and AI infrastructure investors, this is a direct input cost increase worth modeling into any project still in planning, not just a headline about Nvidia's margins. Hardware budgets built on last year's pricing assumptions are already out of date.