If your product has DDR4, DDR5, LPDDR4 or NAND flash on the bill of materials, the DRAM shortage is already your problem, even if no buyer has flagged it yet. Memory makers have spent 2026 moving capacity toward high-bandwidth memory and server DRAM for AI data centers, and by early August industry sources were telling DIGITIMES that much of the capacity allocation for 2027 had already been negotiated. Most mid-volume OEMs hadn’t started planning next year’s builds.
At Thomas Instrumentation, we design and build electronics for industrial, medical and military customers, and memory is now the line we check first on every BOM that comes through the door. This post covers what’s short, why DDR4 is being hit from two directions at once, how allocation actually gets decided, and what an OEM with memory on the BOM should do in the next ninety days.
Why the DRAM Shortage Arrived All at Once
The short explanation is that AI servers buy memory in quantities the market wasn’t built for, and they pay more for it. High-bandwidth memory (HBM) also needs significantly more wafer input per bit than conventional DRAM, so every wafer a supplier moves to HBM removes more ordinary DRAM from the market than it adds in HBM. The CEO of Taiwanese module maker Apacer estimated in July that roughly 60% of DRAM capacity is now directed toward server applications, according to Tom’s Hardware.
Prices followed. TrendForce measured conventional DRAM contract prices rising about 93% to 98% quarter over quarter in the first quarter of 2026, then forecast another 58% to 63% for the second quarter. Increases have slowed since, to a forecast 13% to 18% in the third quarter and 10% to 15% for the fourth quarter. Slower increases are still increases. The price level isn’t coming back down; it’s settling higher.
Looking into next year, TrendForce estimates DRAM supply will fall short of demand by 1% to 2% in 2026 and expects the gap to widen in 2027. New fabs are coming, but TrendForce doesn’t expect meaningful production ramps until the second half of 2027, with substantial output arriving in 2028. SK hynix’s CEO went further in July, telling Reuters that 2027 will be the worst year in the industry’s history from a supply perspective, as reported by Tom’s Hardware. Memory makers have an interest in a tight market, so treat any single forecast with some skepticism. The direction, though, is consistent across every source we follow.
DDR4 and DDR5: What’s Actually on Your BOM
Most of the search traffic around this shortage is about DDR4 and DDR5, and for good reason. They’re the DRAM generations in most computers, industrial controllers and higher-end embedded devices. They’re also being affected in different ways.
DDR5: Available, but Competing With Servers
DDR5 is the generation suppliers are actively making and investing in. The problem is who gets it. Server DDR5 RDIMMs have seen some of the steepest price increases in the market, and component-level DDR5 for embedded and industrial designs competes for whatever capacity is left after server customers are served.
DDR4: Squeezed From Both Ends
DDR4 is the more serious story for anyone with an installed base of long-life products, because it’s tightening for two separate reasons.
The first is demand. As newer memory got expensive, enterprise buyers stepped up DDR4 purchasing, and Taiwan’s Commercial Times reported first-quarter DDR4 price jumps of up to 50%, as cited by TrendForce. Demand has stayed solid across industrial and networking equipment.
The second is supply, and it’s structural. Samsung, SK hynix and Micron all set end-of-life plans for DDR4, and TrendForce reported in January that Samsung was holding firmly to its DDR4 exit despite the higher prices. SK hynix, meanwhile, converted its Wuxi fab from the older 1z process to the 1a node, with about 90% of that site’s capacity now on 1a. Every wafer that moves to a newer node for DDR5 and server memory is a wafer that no longer makes DDR4.
Industrial, medical and defense products are where this lands hardest. Those designs were often qualified years ago around a specific DDR4 part, and they’re expected to keep shipping for years more. Taiwanese suppliers such as Nanya and Winbond are filling some of the gap with mature-node DDR4 and DDR3, but they’re selling into the same tight market, and their pricing reflects it.
LPDDR4, DDR3 and Graphics Memory
The squeeze doesn’t stop at DDR4. Avnet Silica’s third-quarter market report says DDR4 and LPDDR4 are now under allocation, with DRAM lead times of 3626 weeks or more. High-density DDR3 has seen acute shortages, and TrendForce reports that suppliers are scaling back GDDR6 while gradually exiting that market.
NAND Flash: Tight Now, With Relief Penciled In for Late 2027
NAND is following a similar path on a slightly different timeline. TrendForce estimates a 4% to 5% NAND supply deficit for 2026 and expects the balance to turn positive in the second half of 2027 as new capacity comes online. Until then, prices keep climbing, with NAND contract prices forecast to rise another 15% to 20% in the fourth quarter.
Two details matter for embedded and industrial designs. Enterprise SSDs for AI infrastructure are absorbing most new supply, which leaves eMMC and smaller-density parts with less attention from suppliers. And legacy NAND is disappearing faster than the rest: TrendForce expects Samsung’s last MLC NAND shipments in June 2026 and global MLC NAND capacity to fall 41.7% year over year. If your storage choice was made for endurance years ago, confirm what it’s built on.
So “sold out through 2027” needs one qualification for NAND. Capacity is largely committed, and relief is forecast for late 2027, but relief tends to reach the largest buyers first.
How Memory Allocation Actually Gets Decided
Allocation isn’t a queue. Suppliers decide where bits go based on margin and commitment, and right now both favor the biggest customers.
- Long-term agreements come first. Micron and SK hynix have signed long-term supply agreements that commit volume over multiple years, typically with a price floor and ceiling. Those volumes are set aside before anyone else is served.
- Willingness to pay sets the pace. TrendForce notes that cloud providers accepted price increases, which pushed other customers to follow simply to protect their allocations.
- Agreed volumes are honored; new volume is scarce. Suppliers have been shipping agreed volumes to PC makers and module makers each quarter. The word that matters there is “agreed.” Uncommitted demand is the first thing turned away.
- Everyone downstream gets squeezed further. Apacer’s CEO warned that supply from major DRAM makers to independent module makers could fall to just 30% of 2026 levels next year.
Most mid-volume OEMs buy memory through distributors or module makers, which puts them in exactly the part of the market being cut. During allocation, a quoted lead time is a statement of intent. It isn’t a delivery date.
What to Do in the Next 90 Days
These steps are ordered by when to start them, not by importance. Several should run in parallel.
Weeks 1 to 2: Pull Every Memory Line Off Your BOMs
List every DRAM, NAND, eMMC, NOR and SSD part across your active products. Capture the manufacturer part number, generation, density, speed grade, temperature grade, package, current supplier, lifecycle status and your expected 2027 quantity. Most teams find at least one memory part they’d forgotten was on a legacy board.
Weeks 2 to 4: Ask for Allocation, Not Lead Time
Go back to each distributor and ask a different question: how much of this part is allocated to our account for 2027, and on what schedule? Get the answer in writing. A lead time tells you when a supplier hopes to ship. An allocation tells you what they’ve actually committed.
Weeks 3 to 6: Cover 2027 Where the Numbers Support It
For parts with confirmed end-of-life dates or thin allocation, compare the cost of buying next year’s needs now against the cost of a line-down. Account for cancellation terms, since many memory orders are now non-cancelable and non-returnable. Plan for storage too. Moisture-sensitive memory needs controlled handling, and climate-controlled component storage is part of how we protect customer inventory between purchase and build.
Weeks 4 to 8: Start Qualifying a Second Source
Qualification takes longer than people expect, particularly for medical and defense products. If a second DDR4 or eMMC source exists for your design, begin the paperwork now so it’s ready if the first source slips.
Weeks 6 to 12: Flag the Products That Need a Longer-Term Memory Plan
Some products will ship well past 2027 on memory that’s being discontinued. You don’t need to decide their future this quarter, but you should know which ones they are. Our PCB design team can review a board’s memory implementation and tell you what options exist before the decision becomes urgent.
Throughout: Share a Firm Forecast With Your Contract Manufacturer
In a constrained market, firm demand gets served before tentative forecasts. The earlier your manufacturing partner sees real quantities, the earlier they can secure parts against them. We’ve written more about supply chain challenges in electronics manufacturing and how planning reduces them.
Where Thomas Instrumentation Fits
We’ve been designing and manufacturing electronics since 1971, today from our facility in Cape May Court House, New Jersey, and much of our work is long-life industrial electronics built around exactly the kind of memory now under pressure. Because design, sourcing and assembly happen under one roof, a memory problem doesn’t bounce between three vendors. Our electronics manufacturing services include supply chain management, so our buyers watch allocation on your parts alongside our own.
Frequently Asked Questions About the DRAM Shortage
How long will the DRAM shortage last?
TrendForce expects DRAM supply to stay tight through 2027, with new fab capacity ramping in the second half of that year and substantial output arriving in 2028. Some memory makers expect demand to outpace supply for longer. Plan for constrained memory through at least the end of 2027.
Is DDR4 being discontinued?
The three largest DRAM makers, Samsung, SK hynix and Micron, have all announced end-of-life plans for DDR4. Smaller suppliers, including Nanya and Winbond, continue to produce mature-node DDR4, but supply is shrinking while demand from industrial and networking equipment remains steady.
Why are DDR4 prices rising if it’s an older technology?
Supply is falling faster than demand. Fabs are converting DDR4 capacity to newer nodes, and buyers priced out of DDR5 and server memory have shifted purchases toward DDR4, which pushed first-quarter 2026 DDR4 prices up by as much as 50%.
Will NAND flash prices come down in 2027?
TrendForce forecasts the NAND market will move from shortage toward balance in the second half of 2027 as new capacity comes online. Prices are still expected to rise through the end of 2026, and any easing is likely to reach high-volume buyers first.
Get Ahead of Your Memory Risk
If DDR4, DDR5 or NAND is on your BOM and you haven’t confirmed your 2027 allocation, the time to do it is now, while there’s still room to act. Send us your BOM and we’ll flag the memory lines at risk. Contact the Thomas Instrumentation team to start the review.
