TrendForce, a research firm that tracks the memory chip business, has published its outlook for the fourth quarter of 2026 (4Q26). It says demand from AI servers keeps pushing memory contract prices up. Shoppers and device makers are still feeling the strain. Contract prices are what big buyers pay chip makers under deals. For ordinary DRAM, the report expects a rise of 10–15% from the previous quarter. For NAND Flash, the type of memory used for storage, it expects a rise of 15–20%.
DRAM makers are still putting their best production lines toward high-end server chips. That keeps the whole market short of supply. Price rises are slowing down a bit, but they have not stopped.
For laptops, TrendForce says the notebooks sitting in stock increasingly show higher part costs in the fourth quarter. That is hurting sales to buyers. Even so, PC makers keep buying a lot of memory, because they expect DRAM to stay tight in 2027. Throughout 2026, DRAM makers have delivered the agreed amounts to PC makers and module makers every quarter. In 2027, though, PC memory supply could shrink as more production moves to servers.
Server memory is a different story. More server processors are now available, so cloud companies and server makers are buying more memory chips. They need RDIMM modules for general servers that run agentic AI work. Suppliers have made more of the smaller-capacity RDIMMs to match what customers want. Still, limits on parts, on packaging and testing capacity, and on how flexibly the front-end chip factories can switch products mean supply cannot catch up. Server DRAM should stay short in 4Q26. Some suppliers may raise prices more slowly than the market average, because their long-term agreements set how prices are decided.
Phone memory is under pressure too. Costs are rising and cheap old stock is running out, which holds back phone production and lowers demand. Suppliers keep moving capacity to servers, where demand is strong and profits are higher, so phone memory stays tight. Prices keep climbing, but more gently in 4Q26, since big increases were already counted in earlier quarters.
In graphics memory, some PC brands are looking again at the older GDDR6 type to cut the total cost of materials. Demand for the newer GDDR7 comes mostly from AI chips, and buying there is stronger than in PCs. Suppliers are cutting GDDR6 output and slowly leaving that market. GDDR7 output is limited because the best production capacity is going to uses like server DRAM. Graphics memory prices should keep rising.
For consumer DRAM, the new demand comes mainly from makers of SSDs. It adds to existing demand from home networking gear and TVs. A better outlook for QLC SSDs has lifted demand for the DRAM they need. The big suppliers are still holding to production cuts, so the market leans heavily on suppliers in Taiwan. Prices should keep rising, only more slowly.
Now to storage. For SSDs in PCs, makers can use finished products and channel stock built up in the first half of the year to cover fourth-quarter shipments. They mainly buy parts that are in short supply. PC brands are also putting smaller SSDs into mainstream models to save money. That means they buy fewer units and less capacity per device. Buyers already hold enough stock and need little, so NAND suppliers have become more flexible on price and are limiting how much they raise it.
Enterprise SSDs, the drives used in data centers, look very different. TrendForce expects demand measured in bits to grow by more than 80% year over year in 2026. Cloud companies keep building AI inference systems, and the focus is moving from training large language models to using them in real services. Agentic AI is also making data volumes balloon, because systems must fetch and cache information in real time. QLC drives are being used more in vector databases. Suppliers are adding QLC capacity and high-capacity products, and PCIe 6.0 is spreading along with the next AI server platforms. But most of the extra supply is already promised, so little is left for the open market. As a result, enterprise SSD price growth should speed up in 4Q26. It is the only memory category where the rise is bigger than last quarter.
For eMMC and UFS, the memory in phones and similar devices, phone brands are mostly using stock they already hold for fourth-quarter production, so they buy very little extra. Low-margin products such as TVs, set-top boxes and wearables have even less room to pay more. Suppliers want wider price rises, helped by the strong AI trend. Device makers answer by buying only the smallest amounts needed to keep good ties with suppliers. Talks have shrunk to confirming prices on fairly small orders.
Finally, NAND wafer prices are at record highs, and shops and other retail channels increasingly cannot absorb them. Module makers, the main buyers, have become more careful. Suppliers still decide where to put capacity by profit, so wafer supply stays tight. But weak buying by module makers should keep the price rise modest.
Text and photos: trendforce.com · 30.09 — original →
TrendForce, a firm that studies the memory chip market, says prices for memory will keep going up in the fourth quarter of 2026 (4Q26). The main reason is strong demand from AI servers. Regular DRAM contract prices are expected to rise 10–15% from the quarter before. NAND Flash contract prices are expected to rise 15–20% from the quarter before. Enterprise SSDs are the only kind of memory where the price rise should speed up, because big cloud service providers (CSPs) want so many of them.
DRAM makers keep giving their best production lines to high-performance server chips. That keeps the whole market short of supply. Still, the price rises are slowing down a little. For NAND Flash, AI demand is growing fast while demand from ordinary consumers is weak, but prices are going up in every segment.
For PC memory, laptops on shelves now carry higher part costs, and that hurts sales. Even so, PC makers keep buying a lot, because they expect DRAM to stay tight in 2027. That supports prices. Through 2026, DRAM makers have delivered the agreed amounts to PC makers and module makers every quarter. But PC DRAM supply could shrink in 2027, as more factory capacity moves to servers.
For server memory, more server processors are now available. So cloud companies and server makers are buying more RDIMM modules for general servers that run agentic AI work. Memory makers have put more low-capacity RDIMMs into their mix to match what buyers ask for. But shortages of parts, limits in packaging and testing capacity, and limited flexibility in the first stage of DRAM production mean supply still cannot catch up. Server DRAM should stay undersupplied in 4Q26. Some suppliers may see smaller price rises than the market average, because their long-term agreements (LTAs) set the pricing rules.
For phone memory, higher costs and the end of cheap stored stock are hurting smartphone production, so total demand for memory bits is held down. At the same time, AI products bring strong demand and better profit, so suppliers keep moving capacity to servers. That leaves phone memory tight. Prices keep climbing, but more slowly in 4Q26, because big rises were already counted in earlier quarters.
For graphics memory, some PC brands are looking again at GDDR6 to cut their total parts cost. Demand for GDDR7 comes mostly from AI chips, and buying there is stronger than in the PC market. Suppliers are cutting GDDR6 output and slowly leaving that market. GDDR7 output is limited too, because advanced capacity is going to uses such as server DRAM. So graphics memory prices should keep rising.
For consumer memory, new demand has come mostly from SSD makers, on top of the usual demand from home networking and TVs. Better hopes for QLC SSDs have raised demand for the DRAM parts they use. But major suppliers still hold to production cuts, so the market leans heavily on suppliers from Taiwan. Prices should keep rising, just more slowly.
For SSDs in PCs, makers can use finished products and channel stock built up in the first half of the year to cover fourth-quarter shipments. So they mostly buy only parts that are in short supply. PC brands are also cutting SSD sizes in mainstream models to save money, so both the number bought and the average size per device are falling. Buyers have enough stock and little need, so NAND makers are being more flexible on price and holding back the size of increases.
Enterprise SSDs are very different. Demand for their bits is expected to grow by more than 80% year on year in 2026, as cloud companies build more AI inference systems. Demand is also moving from training large language models to actually running AI. Agentic AI is also making data volumes explode, since systems must fetch and cache data in real time, and QLC use in vector databases keeps growing.
Suppliers are adding QLC capacity and more large-capacity products, while PCIe 6.0 is spreading with the next AI server platforms. But most of the extra supply is already promised, so little is left on the open market. That is why enterprise SSD price growth should speed up in 4Q26, the only memory category with a bigger rise than the quarter before.
For eMMC and UFS, the chips used mainly by smartphone brands, those brands are living off existing stock for fourth-quarter production, so they buy very little extra. Low-margin products such as TVs, set-top boxes and wearables can absorb price rises even less. Suppliers want wider price increases, riding the strong AI-driven momentum. But device makers are buying only the minimum needed to keep ties with suppliers. So talks have shrunk to confirming prices for fairly small orders.
Finally, NAND Flash wafer prices have hit record highs, and shops further down the chain increasingly cannot absorb them. So module makers, the main buyers, have become more careful about buying. Suppliers still hand out capacity by profit, which keeps wafer supply tight. But weak buying by module makers should limit overall price growth to a modest rise.
Text and photos: trendforce.com · 30.09 — original →
