I Watched DDR5 Hit 1,400 RMB — Here’s Why You Shouldn’t Panic

I. First, How Bad Is It

Search for a 16 GB DDR5 desktop module on JD.com. Self‑operated big brands hover around 1,400 RMB. Laptop modules are similar: 1,300 to 1,400 RMB for a standard 16 GB stick. Second‑tier brands are a bit cheaper — Lexar’s 24 GB module sells for just over 1,400 RMB, the best value you’ll find right now. High‑capacity SKUs are simply out of stock. 32 GB singles, 64 GB kits — most show “notify when available.”

The distribution channel is even wilder. A friend in computer wholesale told me upstream quotes now “change twice a day.” The price sheet you get in the morning is obsolete by afternoon. Refusals to release inventory are routine. You place an order at the morning price, and by the afternoon they tell you there’s no stock, or they’ll only ship at a marked‑up price. Some sellers are straight‑up telling customers, “Maybe check back in a couple of months.”

That’s not even the craziest part. The DDR5 surge has pulled DDR4 up with it — 16 GB DDR4 3200 is now over 800 RMB on JD. Even DDR3, stuff that should have been e‑waste ages ago, has doubled in price. A few years back, selling it wouldn’t even cover shipping. Now it’s actually worth selling.

Han Lu said something the other day that’s been making the rounds: “A 10,000 RMB PC today gives you what 5,000 RMB bought two years ago.” Not an exaggeration.

Here’s my own story. A while back, I helped a few friends offload their spare memory — everyone heard prices were going up and pulled sticks out of old builds to cash in. I sold mine too, but I sold way too early. A 16 GB Kingston Fury DDR5 stick — I let it go for 650 RMB. Two weeks later, the same stick was selling for 950 on Xianyu. Another week, it broke 1,000. By stock market logic, if you didn’t make it, you lost it. I completely missed the spike. Painful.

Looking back, though, I was lucky to move them at all. Once prices climb higher, high‑priced listings on Xianyu just sit there — no buyers. You see the price, but you can’t actually sell into it. The ones who cash out are the real winners; unrealized gains don’t count. A friend of mine held onto four sticks, watched JD prices climb, and felt great. He listed them at 1,100 each on Xianyu. Two weeks, over a dozen inquiries. Not a single actual payment.

II. How to Tell Whether This Shortage Is Real

Watch three prices: contract price, spot retail price, and second‑hand transaction price.

Contract price — what the chipmakers (Samsung, SK Hynix, Micron) sign with their big customers (Dell, HP, Lenovo and other OEMs). Large volume, stable contracts, executed monthly or quarterly. This price isn’t public, but analysts grill executives about it on quarterly earnings calls, and management gives vague directional guidance.

Spot retail price — the price you see on JD and Taobao. The only real‑time window ordinary consumers have into the market.

Normally, these two don’t drift far apart. Contract price is the main army; spot price is the scout. They pull each other along. A spread of 5% to 10% is healthy — supply and expectations are stable.

This time it’s different.

TrendForce’s July 3 survey shows the DRAM market remains tight in Q3 2026, with general‑purpose DRAM contract prices expected to rise 13% to 18% quarter‑over‑quarter. Manufacturers keep shifting capacity toward server applications, squeezing PC DRAM supply, while AI server demand supports high‑capacity memory procurement.

Contract prices are rising. Spot prices are rising even faster. The spread is widening.

Right now, a 16 GB DDR5 standard module on JD is 1,400 RMB — up nearly 70% from three months ago. Over the same period, contract prices have increased about 25% to 30%. Spot inflation is more than double contract inflation.

What does that tell us? The distribution channel is in serious instability. Downstream distributors have fewer and fewer modules in stock, and they’re pricing entirely by nerve. You think 1,400 is expensive today? Tomorrow it might be 1,500. With that expectation, distributors are even less willing to sell at lower prices — they’d rather hold and wait for tomorrow’s higher price. It’s a game of chicken, and no one wants to be the last sucker to sell cheap.

Until the spread closes, retail prices aren’t coming down. When spot price growth narrows to match contract price growth, that’s the first real signal we’ve hit a top.

Second‑hand transaction price — primarily on Xianyu.

You can estimate it from what Huaqiangbei recyclers are offering. Right now, DDR5 single 8 GB sticks fetch 400 to 500 RMB in recycling, 16 GB about 900 to 1,100. DDR4 is cheaper, but many recyclers have stopped taking it entirely — DDR4 8 GB used to go for 200 RMB in recycling, now it’s dropped to around 150. Notice: recycling price going down doesn’t mean retail price is going down. It means the recyclers themselves are afraid of catching a falling knife and are buying more cautiously.

Recyclers buy low to sell higher, so Xianyu listings from individuals will always be above recycling price.

But the real metric is actual person‑to‑person transaction prices — not the bait prices dealers list, but prices where real trades happen. How do you tell if a sale actually closed? Check the seller’s profile: a genuine individual seller will also have an old phone, an old GPU, some unused headphones for sale — not a feed full of memory modules. If their entire store is memory sticks and the description says “brand new sealed, with receipt, free shipping,” nine times out of ten it’s a dealer pretending to be an individual.

At high prices, volume dries up — the classic “price without liquidity.” Almost no one actually buys brand new retail on JD unless they have a specific need: must have a receipt, must have the latest factory warranty, must go through official channels.

So second‑hand transaction prices tell you where the real end‑user demand ceiling is. That ceiling is moving up, but slower than spot prices. End‑users are already voting with their feet — not buying, waiting, or finding alternatives.

III. The Number to Watch Isn’t Price — It’s Inventory

Price is the symptom. Inventory is the root.

How many DDR5 chips do Samsung, SK Hynix, and Micron have left? That determines whether they’re in a rush. Plenty of inventory: they clear it at a discount. Bottom of the barrel: they jack up prices at will.

Public data shows inventory levels are currently about two to four weeks. This figure comes from “inventory turnover days” in each manufacturer’s financial reports and industry analyst tracking. Normal levels are eight weeks or more. At the bottom of the 2023 down‑cycle, inventory topped twelve weeks — manufacturers were practically begging OEMs to take chips off their hands.

What does two to four weeks actually mean?

Think about this: from wafer start to packaging and testing to shipping, a single DDR5 chip takes about three months to produce. Two to four weeks of inventory means the chipmakers are essentially empty‑handed — every chip they make is sold the moment it’s finished. No safety buffer whatsoever. If a production line goes down, or a major customer places a sudden rush order, the entire supply chain tightens instantly.

Even a buyer as massive as Apple is scrambling for memory. Recently, they floated the idea of sourcing from Chinese manufacturers. Whether they actually plan to buy or just want leverage against Samsung and Hynix in negotiations, no one knows. Either way, it tells you one thing: the upstream shortage is making even the biggest customers anxious. Apple is the kind of buyer that suppliers used to line up to serve. Now they’re proactively hunting for channels. That signal is louder than any price data.

When inventories get back above eight weeks — that’s when there’s a real basis for price cuts.

This data isn’t public. Ordinary people can’t get real‑time numbers. But pay attention to quarterly earnings calls. Analysts always ask about inventory. Changes in how executives phrase their answers are signals. Listen closely: “Inventory levels are healthy” means it’s on the high side, likely to drop. “Inventory is at low levels” means more increases ahead. “We are actively adjusting capacity” means even they don’t know what to do.

IV. When Will Prices Actually Drop — Don’t Hold Your Breath for a Year

The patient waiters never lose — as long as you refuse to admit defeat, keep waiting, and eventually there will be a chance to buy at a relative low.

But how long are we talking?

The most reliable window is DDR6 launch.

The historical pattern of the memory industry: one to three months before a new generation launches, the previous generation gets cleared out at the best discounts. The logic is simple: DDR6 takes over the premium tier — enthusiasts and OEM new‑product orders switch directly to DDR6. DDR5 moves from “latest generation” to “previous generation,” its pricing space compresses, and manufacturers are finally motivated to clear inventory at lower prices.

Look at history: DDR4 launched in 2014. DDR3 started dropping sharply in the second half of 2015, with the deepest discounts hitting in 2016. DDR5 launched at the end of 2021. DDR4’s true clearance fire sale didn’t happen until 2023 — nearly two years later. The timing plays out almost the same way every time.

The problem: DDR6 isn’t here yet.

JEDEC hasn’t even finalized the DDR6 specification. Given the generational cadence, from spec finalization to mass production typically takes 18 to 24 months. Current public information puts DDR6 commercial availability around 2028. That puts the clearance window somewhere around late 2027 to early 2028.

That means, from today, we’re looking at another year and a half to two years of high‑level consolidation.

There may be small fluctuations in between — a slow seasonal quarter, some manufacturer doing a small volume release to hit targets, prices pulling back 5% to 10%. But the big downward turning point isn’t there.

There’s only one core reason: AI.

V. How Much of This Is AI’s Fault — Let’s Be Fair

AI demand is real. It’s not made up.

HBM (High Bandwidth Memory) has gross margins three to five times higher than consumer DDR5. When Samsung and Hynix shift wafer capacity to HBM, they’re effectively imposing an “opportunity cost tax” on PC DRAM — the profit they lose from not using that capacity for HBM has to be recovered by raising consumer prices. That’s the “marginal cost transfer of capacity shift” in economics terms. In plain English: the money you don’t make on HBM, you have to make back on DDR5.

And that’s not even the whole story.

Here’s a change I’ve observed firsthand: a medium‑scale development project that used to take two or three people grinding for a month — now, with an engineer who knows prompt engineering and project structure, paired with a capable AI model, core work can be wrapped in half a week. Especially for smaller companies that don’t handle sensitive data, AI integration is almost frictionless.

I did a data cleanup project last month originally scoped for two weeks — no company data involved, just my own side work. I paired DeepSeek with some fixed scripts and wrapped it in two or three days. It was dumb at first — field mappings kept drifting — but once I broke the logic into seven or eight small chunks and validated each one separately, it stabilized. The final output was cleaner than what I’d have written by hand, because its formatted output doesn’t make mistakes.

The efficiency gain was about ten to one. I finished in two evenings what was scheduled for two weeks. The rest of the time went to checking logic gaps and writing documentation.

Of course, if you can do more work, you find more work to do. When you’re more efficient, your boss piles on more. Project complexity increases — you start taking on requirements you wouldn’t have touched before. Pace accelerates, workload expands. That’s a separate conversation.

Back to memory. The compute demand from AI isn’t entirely a bubble. It’s consuming more and more HBM and DRAM capacity, and there’s no sign of it stopping in the near term. None of the major manufacturers are willing to expand capacity aggressively — memory is a capital‑intensive industry. A single new production line costs billions of dollars and takes three to five years to break even. What if the cycle turns down right after you finish expanding? In 2019, Micron mis‑timed an expansion and posted a quarterly loss of over $2 billion. No one wants to relive that lesson.

So consumer memory getting squeezed — no short‑term fix.

However, AI demand alone wouldn’t impact the entire consumer memory market to this degree. Consumer DDR5 accounts for less than 30% of total DRAM market. HBM and server DRAM are the real volume drivers. AI is eating into incremental capacity, not existing supply. The real culprit is that this has been weaponized as a fear tool for this cycle. Fear feeds on itself — distributors hoard, downstream buyers scramble, quotes jump. Fear moves faster and further than the underlying supply‑demand reality.

Think of it this way: AI is the lighter. Fear is the gasoline. The flame is real. But what’s really burning is the gasoline.

VI. But the Impact on You Is Smaller Than You Think

After all that anxiety‑inducing talk, here’s something reassuring.

Most people don’t actually need DDR5.

Think about it calmly.

If you have an older DDR4 platform, find a used CPU that fits it — say an i7‑11700K or a Ryzen 7 5700X, tray CPUs are basically at rock‑bottom prices right now — and pair it with a set of DDR4 3200. JD has new 16 GB sticks for about 800 RMB. More expensive than before, sure. But for daily office work, gaming, video editing — how much real‑world difference can you actually feel versus a DDR5 system?

For 90% of people, the difference is imperceptible.

Office work? No contest. Word and Chrome don’t care about memory generation. Fifty tabs open — DDR4 3200 versus DDR5 6000 — Chrome couldn’t care less.

Gaming? Be honest with yourself: if you sink your budget into the CPU and GPU, is memory frequency really your performance bottleneck? At 4K, the GPU is the absolute bottleneck. A 20% difference in memory bandwidth might show up as a 2% difference in frame rates. At 1080P competitive gaming, CPU single‑core performance matters way more than memory bandwidth. The number of games that genuinely eat memory bandwidth can be counted on one hand.

DDR4 still too expensive? Keep going down.

DDR3 budget builds: E5 Xeon “e‑waste” plus a cheap Chinese motherboard — a whole system for a few hundred RMB. Runs Linux servers, NAS, soft routers, perfectly capable. The nice word for it is “Tualatin spirit” — the less nice word is “trash picking” — but it’s not like it doesn’t work.

I know one guy who built a “server” for 300 RMB: an E5‑2650 v2 for 50, an X79 clone board for 120, four 8 GB DDR3 server pulls for 80 total, power supply and case from a scrapped office PC. Running Plex media server, Transmission downloader, Home Assistant smart home hub — four months without a reboot. His exact words: “While you guys are stressing over memory prices, I’m watching 4K movies.”

What about performance? Sure, Jellyfin transcoding is a no‑go — but he doesn’t transcode. Clients direct‑play, the server just streams. CPU utilization sits below 20% year‑round. That’s the real definition of “good enough.”

VII. How to Build a PC During an Up‑Cycle

One rule: spend money to save time, spend time to save money. Depends how urgent you are.

Three scenarios:

First: you need a new platform now, non‑negotiable.

Don’t buy 16GB×2 high‑frequency RGB kits — that’s the highest‑premium segment, the exact product hoarders are flipping. Check JD: G.Skill Trident Z Royal 16GB×2 6000 goes for 2,300. Same brand, same capacity, but standard 5600 module — 1,400. What’s the difference? Lights. Lights cost 900 RMB.

Buy a single 24 GB or 48 GB DDR5 5600/6000 standard module — no RGB, not flagship overclocking SKUs, average bin chips that hoarders don’t bother with, lowest premium. Run a single stick for now, add a matching one when prices drop next year for dual‑channel. Single‑channel does lose some performance, but for the vast majority of use cases, you won’t notice. The scenarios where you can actually feel the difference are benchmark software and some niche professional productivity tools — gaming differences are within 5%, and the gap shrinks as your GPU gets stronger.

Why 24 GB instead of 16 GB? Because 24 GB is the per‑side capacity limit for the new die generation. It’s only 20% more expensive than 16 GB but gives 50% more capacity — better value. And 24 GB modules target workstations and content creators, not the segment hoarders care about.

Second: old platform user, no rush to upgrade.

DDR4’s current price is “emotional premium” — it’s riding DDR5’s coattails, not driven by actual demand. DDR4 production lines matured years ago. There’s no capacity shortage. It’s purely channel sellers seeing DDR5 rise and marking up accordingly.

Real individual sellers on Xianyu (not dealers) are still pricing the old way. Hunt for “switching to DDR5, selling off DDR4” listings from graduates or enthusiasts — their prices are way below JD new stock. How to hunt: search “DDR4 16G DDR5 upgrade sale,” filter by “individual” not “professional seller,” sort by newest, check listings from the last hour. These sellers usually throw up a low price and move on — they don’t follow the market. Their goal is cash recovery, not profit. This is probably the best value scavenging opportunity right now.

Third: the ultimate money‑saving move — don’t upgrade your software.

Windows 12’s AI PC requirements mandate higher memory specs. If you don’t install it, don’t chase new systems, your memory demand doesn’t suddenly balloon. As long as you stay on 2024‑2025 versions of your software stack, DDR4 3200 remains a “sweet spot” speed.

OS vendors and software vendors are the real memory demand drivers. Don’t upgrade to their new versions, and they can’t reach into your wallet. Windows 11 eats 1.5 GB more memory than Windows 10. Windows 12 is projected to eat another 2 GB on top of that. Adobe raises memory requirements with every major version. Don’t chase the new releases, and all those extra overhead costs don’t apply to you.

The real meaning of this advice: the biggest victims of this price surge are “habitual upgraders.” If you’re comfortable with your system versions sitting two years back, your hardware requirements can sit two years back too. And two years ago, memory prices were half of what they are now.

VIII. The Memory Industry’s Iron Law: Three Years Lean, Three Years Fat

Samsung, SK Hynix, Micron — they’ve been in this business for decades. They know this rhythm better than anyone.

In the 2019 down‑cycle, Micron posted a quarterly loss of $2.3 billion. Stock dropped from $60 to just over $30. In the 2021 up‑cycle, Micron posted $2.2 billion in quarterly profit. Stock doubled. In 2023, another brutal price war — SK Hynix lost 3.4 trillion won (about $2.6 billion) in a single quarter. And then? 2024 AI demand exploded. HBM made them back everything they’d lost, plus interest.

That’s the destiny of the memory industry. Three years of famine, three years of feast. In down‑cycles, they sell at a loss. Sometimes the market leaders start price wars themselves, bleeding out until they’re nearly dead, then getting swallowed up — look at Qimonda. Look at Elpida. Both disappeared that way. In up‑cycles, they double and triple prices, recovering every loss from the lean years, plus interest.

When to raise prices, by how much, what pretext to match market sentiment — it’s all textbook. Look back at news coverage from every memory cycle. The reasons are always the same: insufficient capacity, strong demand, earthquakes, fires, blackouts. Look closer: reasons are just reasons. The real driver is always two words: cycle.

There’s one hidden variable in this cycle that many people overlook: domestic Chinese memory from ChangXin Memory Technologies (CXMT).

CXMT already proved its mass‑production capability in the DDR4 era. By 2023, CXMT’s DDR4 chips were showing up in large volumes on Chinese‑brand memory modules, pulling entry‑level DDR4 prices down significantly. DDR5 R&D and production line development have been progressing — slower than market optimists hoped, but the direction is clear.

Once domestic DDR5 chips pass major OEM validation and scale up — making it into Lenovo and HP supply chains, or showing up on JD’s self‑operated domestic memory brands — the “cartel of scarcity” maintained by the three foreign giants will collapse overnight. Pricing power no longer belongs to just three players. The price‑fixing alliance won’t hold. At that point, you won’t need to wait for DDR6. Prices will start loosening — maybe even crashing.

Domestic substitution isn’t just a slogan. It’s a real capacity variable. The only question is timing. Based on publicly available project timelines, that inflection point could arrive in the second half of 2027.

IX. One Last Thing

New build, must‑have: go with the “single standard module transitional” strategy. Don’t touch the highest‑premium flagship SKUs. Once you’ve plugged your stick in, just use it. Don’t open Task Manager every day to stare at memory frequency. Forget about it, and it won’t hurt you.

Not urgent: push your build plan directly to Double 11, 2027. That might be the year of the most brutal price war in consumer memory history — DDR6 launch approaching, DDR5 inventory needing clearance, domestic capacity potentially hitting volume. Triple pressure stacking up — that’s when the patient waiters finally get their victory day.

The patient waiters never lose.

They just have to wait long enough.

And in this long wait, what you’re saving isn’t just a thousand RMB. It’s the anxiety of being held hostage by market fluctuations. You can look at any price peak and calmly say: “Doesn’t matter to me.”

Then close JD, open your old computer, and go back to watching that 4K movie.