Nvidia is warning big customers that servers built around its AI chips will cost over 15% more, bad news for the cloud giants already spending heavily on AI and good news for memory makers.

• The price rise starts with systems shipping early next year, covering Nvidia's top Vera Rubin and Grace Blackwell chips.
• The cause is a jump in memory chip costs, which Nvidia cannot push down or absorb itself.
• Samsung, SK Hynix and Micron now hold the upper hand, since AI demand for memory is growing faster than they can build it.
• Microsoft, Google and Oracle get the bill, and Apple and Qualcomm have already raised product prices over the same shortage.
• Nvidia is still highly profitable, with margins near 75%, and has also raised prices on gaming graphics cards.

Outlook: How hard customers push back depends on whether they can lock in their own memory supply, and their in-house chip plans still leave them buying from Nvidia for now.