Nvidia has told its largest customers that AI hardware is about to get meaningfully more expensive.
According to GIGAZINE, the company notified major customers it will raise prices on AI-related products by more than 15%, potentially touching its "Vera Rubin" and "Grace Blackwell" lines. The Information, citing server makers, puts the increase on flagship AI chips at 17%. CNBC-TV18 reported the news broke over the weekend, with chips costing 15% more by the end of 2026.
The stated culprit isn't greed so much as supply. Reporting surfaced via MSN warns that AI servers could cost more than 15% extra as memory prices surge — a shortage in one crucial component rippling outward through entire data centers. TweakTown reports the hikes on Vera Rubin and Grace Blackwell servers begin early next year as memory costs climb.
And this isn't the first increase. Network World and CIO.com note the 15% hike comes on top of an even larger increase in July. Consultants and analysts told CIO.com that IT leaders likely have little choice but to accept continually rising AI prices, since switching suppliers — where it's even possible — probably won't help them sidestep the underlying costs.
The market reaction was split. The Korea Herald reports that memory makers Samsung and SK hynix gain pricing power as Nvidia's server prices rise, though CNBC-TV18 noted a sharp sell-off in Korea's Kospi. Meanwhile, Indian IT stocks rallied: the Nifty IT index rose over 1%, per reporting from MSN and Zeebiz, as investors revisited worries about AI infrastructure costs, stretched valuations and massive capital expenditure — with some asking whether AI bubble fears are coming true.
Why it matters: the cost of building AI is now rising faster than the cost of the chips alone, and those increases eventually reach the companies — and customers — paying for AI services.