DDR5 is on allocation.
Here's what that actually means.

Our Monday dashboard shows DDR5 contract pricing up 150 to 200% year on year, and next to that number sit the two words buyers dread most: on allocation. If your builds depend on DDR5 and your last confirmation came back thinner than your forecast, this piece is for you. Here is how rationing works, why it happened, how long the tightness is likely to run, and the moves that keep a production line fed while the queue clears.

What does "on allocation" actually mean?

Allocation is the memory industry's polite word for rationing. A maker looks at the orders on its book, accepts it cannot fill them all, and decides what share of real output each customer receives. You no longer buy what you need. You receive what you are given.

In practice that means confirmed orders get trimmed, lead times quoted in weeks become quarters, and the next contract round is priced at whatever the market will bear. And the queue is sorted by who you are, which we'll get to below.

Why did DDR5 get scarce? HBM is eating the wafers.

Nobody's fab burned down. This shortage is a deliberate capacity trade. Samsung, SK Hynix and Micron are steering wafer starts toward high-bandwidth memory for AI accelerators, because HBM earns far more per wafer than commodity DRAM. HBM is stacked and packaging-heavy, and it consumes roughly three times the wafer capacity per gigabyte of plain DDR5.

The numbers move fast. TrendForce puts HBM at about 23% of global DRAM wafer capacity, up from roughly 10% in 2023, and AI data centers are expected to absorb around a fifth of total DRAM wafer output in 2026. Every wafer that becomes HBM is a wafer that never becomes DDR5. Mainstream memory is being made from what's left over.

Who gets served first when supply is rationed?

The pecking order is unsentimental. Hyperscalers and AI infrastructure buyers sit at the front, followed by large OEMs holding multi-quarter contracts. Big EMS companies that can commit to volume and swallow the new pricing come next.

Everyone else, meaning smaller OEMs, smaller EMS firms and anyone buying quarter to quarter, gets reduced allocations, pushed-out deliveries, or a referral to the spot market. That last group is where our phones start ringing.

We've seen this queue before

This desk has quoted memory since 2004, which means we worked through the 2017 to 2018 DRAM squeeze. The pattern then was the same: suppliers protected their largest accounts, prices climbed for seven straight quarters, and the buyers who came out ahead were the ones who qualified alternates and secured stock early instead of waiting for relief that kept sliding right.

One thing is different this time. In 2017 the driver was smartphone and server demand catching up with underinvestment, and new capacity eventually closed the gap. Today the pull is structural: as long as AI accelerators sell, HBM will keep first claim on wafers, and conventional DRAM stays the residual.

How long will DDR5 stay tight?

Nobody serious is calling an early end. TrendForce expects contract prices to keep climbing through 2026, with quarterly jumps in the 55 to 60% range at the sharpest points. The broader analyst consensus sees no meaningful relief before the second half of 2027, and several forecasts push normalization into 2028 if AI demand holds.

The planning assumption we give our own customers is blunt: treat today's price as the floor for the next twelve months, and treat every allocation letter as provisional.

The playbook: six moves that keep lines running

1. Stretch the forecast. Twelve to twenty-four months of visibility, with firm early orders on the critical parts. Allocation follows the order book; makers serve demand they can see.

2. Protect your legacy parts too. DDR4 is retiring while everyone stares at DDR5, and it's up 70 to 90% year on year on our board. Audit which products can't leave DDR4 before 2028 and price a last time buy with phased deliveries. It's usually cheaper than chasing allocation every quarter.

3. Qualify a second source. Two manufacturers and two channels per critical SKU. Teams re-qualifying second-tier makers for industrial DDR4 are finding supply the tier ones no longer offer.

4. Carry a real buffer. The just-in-time reflex is a liability in a rationed market. Sixty days of safety stock on critical memory is the current baseline; 90 to 180 days where a stockout stops a line.

5. Sign the long-term agreement if you're offered one. Buyers locking supply two to three quarters ahead, at banded pricing, are the ones still receiving full quantities.

6. Use the independent channel, with a lab behind it. When the franchised pipe runs dry, brokered stock fills the gap, and that's where counterfeit risk lives. Every suspect lot we ship can run through our Malmö lab, up to the full 10-step AS6081 protocol under a x2145 microscope. Around 80% of our business is sourcing to order; allocation seasons are what the desk is built for.

Holding memory instead of hunting it?

The same market has a flip side. If your stores hold uncommitted DDR4 or DDR5 from a cancelled project or an over-order, it is worth more today than the day you bought it. Our excess desk turns that stock into cash and puts it on a warranty-backed shelf for a buyer whose line is waiting.

Frequently asked questions

Is DDR4 a safe fallback while DDR5 is on allocation?

No. DDR4 capacity is being retired as makers migrate wafers, and contract prices are up 70 to 90% year on year. Treat it as its own supply risk and plan a last time buy.

When will DDR5 prices come down?

Consensus says prices rise through 2026, with no meaningful relief before the second half of 2027. Several analysts say 2028. Budget accordingly.

Can a smaller OEM get DDR5 without a long-term agreement?

Yes, with a buffer strategy and a vetted independent channel. Insist on tested lots; an untested brokered module is a risk your line shouldn't take.

Should I sell excess memory now?

If it's uncommitted, yes. This is a seller's market, and surplus memory is the rare corporate asset that appreciated this year.