GPU and RAM prices are creeping up

Graphics cards are getting squeezed by the same part that used to be the boring line item: memory. A modern gaming card is a graphics processor wrapped in fast memory chips, and when those memory chips jump in price, the whole board gets more expensive. (trendforce.com) That memory is called dynamic random-access memory, which is just the short-term workspace a chip uses while it is doing a job. On a graphics card, the memory has to move huge textures and frames every second, so vendors use premium versions like Graphics Double Data Rate 7 instead of the slower sticks that sit in a laptop. (nvidia.com) The price shock is not subtle anymore. TrendForce said on February 2, 2026 that conventional dynamic random-access memory contract prices for the first quarter of 2026 were now expected to rise 90% to 95% from the previous quarter, and it said personal computer dynamic random-access memory prices were expected to at least double quarter over quarter. (trendforce.com) Spot markets are still moving around day to day, but the broader direction is up. TrendForce’s public price page on April 9, 2026 showed several dynamic random-access memory contract categories above their prior session averages, including Double Data Rate 4 16 gigabit chips up 13.46% and Double Data Rate 4 8 gigabit chips up 13.04% in the latest contract update. (trendforce.com) The reason reaches far beyond gaming computers. Artificial intelligence servers use enormous amounts of high-bandwidth memory, which is a stacked form of dynamic random-access memory built like a high-rise instead of a ranch house so data can move faster between layers. (micron.com) Once that high-bandwidth memory market tightened, memory makers started steering capacity toward the parts with the fattest margins. SK hynix said on January 28, 2026 that its record results were driven by artificial intelligence memory and other high-value products, which is corporate language for “the best fabs are busy making the expensive stuff first.” (news.skhynix.com) Micron said on its March 18, 2026 earnings call that its high-bandwidth memory was sold out for calendar 2026. When the premium lanes are fully booked that far ahead, the cheaper memory used in consumer hardware stops looking like a market where sudden bargains appear. (fool.com) That is why graphics card prices can rise even when the graphics chip itself has not changed. Board partners still have to buy memory, power components, and coolers, and if one major input spikes, the final shelf price moves even if the official launch price stays printed on the box. (trendforce.com) NVIDIA’s own launch prices show how little room there was to begin with: the GeForce RTX 5090 opened at $1,999 and the GeForce RTX 5070 opened at $549. Those numbers were already high enough that even a modest bill-of-materials increase can turn a “wait for a sale” plan into a very short-lived window. (nvidia.com 1) (nvidia.com 2) There was a brief stretch in late 2025 when some top-end cards touched list price again in the United States. Reports from September 2025 showed the GeForce RTX 5090 finally appearing at its $1,999 official price, which now looks less like a new normal and more like a pause between squeezes. (videocardz.com) So the practical read for buyers in April 2026 is simple. If your current card still runs the games or workloads you need, stretching one more year may beat jumping into a market where memory suppliers, not gamers, are setting the tempo. (trendforce.com)