The buyers who assemble the world’s phones, PCs, and servers are being asked to pay more for memory again. Samsung Electronics is negotiating with customers to raise average prices for conventional DRAM by about 20% in the third quarter of 2026, according to people familiar with the discussions, with prices for LPDDR memory used in mobile devices rising even more. If agreed, it would be the third consecutive quarter of large increases at the world’s largest memory maker, following a first-quarter rise of roughly 90% and a second-quarter increase of 50% to 60%.
The cumulative effect is staggering. Contract prices for standard DRAM have more than doubled over the first half of the year, and spot prices have moved even higher. The increase is the clearest sign yet that the shortage gripping the memory industry — long confined to the high-bandwidth memory that sits beside AI accelerators — has spilled into the commodity DRAM that every computer, phone, and data center still needs.
The cause is the same one driving everything else in the chip industry this year: artificial intelligence. AI servers consume far more memory than conventional servers, and the industry’s three suppliers — Samsung, SK Hynix, and Micron — have all but exhausted their capacity producing HBM for Nvidia and other accelerator makers. With fabs running flat out on the most profitable products, little capacity remains for the standard DRAM that phones and PCs rely on, and prices have responded accordingly.
Goldman Sachs analysts have described the global memory market’s supply-demand imbalance as the most severe in nearly 15 years, with AI servers and inference cards the core drivers. The bank expects prices to keep climbing through the rest of the year, and it argues that the shortage will not be resolved until the new fabs announced this year begin producing in 2027 and 2028. Until then, memory makers hold most of the negotiating power.
Samsung’s customers are pushing back. Phone makers face the sharpest pressure, since LPDDR memory is one of the largest single cost components of a premium handset, and several have told Samsung that increases of this size will force them to raise device prices or trim memory content in new models, people familiar with the discussions said. PC makers face similar math, and server buyers — already paying record prices for HBM — are being asked to absorb higher costs on the standard DRAM that fills the rest of their machines.
The memory makers have little reason to blink. All three are selling everything they can make, and Samsung’s own record profits this year have been built on the very price increases its customers are complaining about. The industry also remembers 2023, when a glut drove DRAM prices to unprofitable lows and forced the same companies to cut production; discipline since then has kept supply tight even as demand grew.
Analysts said the real question is not whether the third-quarter increase sticks, but how long the pricing power lasts. Memory demand is notoriously cyclical, and the industry’s history suggests that when new capacity arrives all at once — as it is scheduled to do in 2027 — prices can turn just as quickly as they rose. The companies now enjoying the boom are investing the profits back into the fabs that will eventually produce the glut, a pattern the memory business has repeated for four decades.
Contract negotiations in memory are slow, formal affairs. Samsung typically opens with a proposed price, customers respond with volumes, and the two sides settle somewhere between supply and demand — which is why the third-quarter round is being watched so closely. Spot prices, which move daily, are running well above contract levels, giving Samsung the stronger hand. Customers who resist the increase face the prospect of paying spot prices or going without allocation, and in a shortage, allocation is what matters.
The industry’s discipline is the other half of the story. After the 2023 collapse, all three memory makers committed to capacity restraint, and that commitment has held even as prices recovered — a departure from the industry’s historical pattern of overbuilding at the first sign of profit. The result is that supply has stayed tight while AI demand has grown, and analysts say the three companies are effectively coordinating through their own self-interest rather than any formal arrangement.
What could break the cycle is the capacity arriving in 2027. Samsung, SK Hynix, and Micron are all building fabs scheduled to open in the next two years, and Goldman Sachs has cautioned that the same imbalance that is producing record prices today could produce the opposite once that capacity comes online. For now, the direction of travel is set: the third-quarter increase is likely to stick, and customers are already planning around prices that would have seemed impossible two years ago.
For the broader technology industry, the increases are a cost story with wide reach. Memory is a small part of a server’s price but a large part of its availability, and every data-center operator, cloud provider, and device maker is now budgeting for memory prices that have tripled in a year. The AI boom built the shortage; the rest of the computing industry is paying the bill.
