Nvidia has told major customers that prices for AI server systems are set to rise by more than 15%, according to Korean media citing Bloomberg, as higher memory costs move through the AI hardware supply chain.
The increases are expected to apply to systems shipping from early next year, including next-generation Vera Rubin products and existing Grace Blackwell-based systems. Microsoft, Google and Oracle were among the large customers reportedly notified.
The important Korean angle is memory. Nvidiaโs accelerators depend on large amounts of high-performance DRAM and advanced memory, markets in which Samsung Electronics and SK hynix are major suppliers. If memory prices continue rising, the benefit flows upstream to chipmakers while the cost is pushed downstream to AI infrastructure buyers.
A gain for memory makers, a cost for AI builders
That creates a two-sided effect for Korea. Stronger memory pricing can improve earnings and negotiating power for Korean semiconductor companies. But more expensive Nvidia systems can also increase the cost of building AI data centers and deploying large-scale models.
Cloud companies have some alternatives, including custom accelerators, but Nvidia remains deeply embedded in AI infrastructure. That limits how quickly major customers can avoid price increases.
The next question is where the extra cost lands. Hyperscalers can absorb some of it, slow capital spending, renegotiate contracts or pass part of it on through higher cloud and AI-service prices.