The most powerful company in AI just told its customers the bill is going up more than 15 percent — because of memory
Server makers building machines for Microsoft, Google and Oracle have started telling those customers that systems containing Nvidia's AI chips will cost more than 15 percent extra in many cases, on gear shipping from early 2027. The increases cover the flagship Vera Rubin and Grace Blackwell systems, and vary by chip generation and memory configuration. The stated cause is not Nvidia's silicon but the DRAM around it: memory prices have surged as Samsung, SK Hynix and Micron fail to keep up with AI demand. Nvidia did not comment.
Why it mattersNvidia has spent three years as the company that sets prices rather than pays them. On this one it is passing a cost through, which means the memory makers now have leverage over the most valuable chip company in the world. Everything downstream — cloud rental rates, the price of a million tokens, what a startup pays to train — sits on top of that bill.
Read in the app — free, in 9 languages
Related stories
The day after his robot company's stock closed 460 percent up, the founder said the breakthrough might be ten years out2026-08-22Nearly a third of American tech job ads are now AI jobs — and in the Bay Area it is 57 percent
2026-08-22Brazil put $444 million into AI machines and deliberately split the order between China and America
2026-08-22Apple is cutting the people who built the old Siri, because the new one needs a different kind of engineer
2026-08-22Anthropic is not buying these chips. Broadcom is borrowing up to $100 billion so somebody else can buy them and rent them out.
2026-08-22