DDR5 Up Another 10-15% in a Week, Huaqiangbei GPUs Double in Ten Days: The Channel ‘Financialization’ Inflection Point Beneath Diverging Spot and Stock Prices#

Chasun | October 22, 2026

Over the past week, the memory market has produced two curves pointing in exactly opposite directions. On the channel side: US system integrator Stince Built notified customers that DDR5 memory prices would rise another 10%-15% within the week, now formally in effect; in Huaqiangbei’s GPU wholesale stalls, the RTX 5090 has climbed a cumulative 15,000 yuan in ten days. On the capital markets side: A-share storage stocks have slumped together since July, with Biwin Storage down a cumulative 66.74% and a host of storage supply-chain companies off more than 60%. A price-hike notice, sealed warehouses, and crashing stock prices all landing in the same seven days is a rare sight in the memory industry’s recent cycles.

The same commodity rising on the spot market while its makers’ securities fall is not market noise — it is two pricing logics coming uncoupled. Rather than repeat the supply-side narrative, this piece focuses on a less-discussed question: when the channel treats “expected price increases” as an asset to be managed, what kind of market does the ordinary PC builder actually face?

First, the key waypoints of this week’s action, strung into a single line for reference below:

The three dates interlock: rumors ignite expectations, spot prices take the baton and climb, and upstream players back the increases by “tearing up orders” — the full chain this article sets out to unpack.

1. Beyond the Price-Hike Notice: The Unconventional Signal Inside a Channel Announcement#

Start with the most telling announcement of this run. Stince Built is a US system integrator that ships mainly to complete-system customers. Around October 21, the company relayed two things to its customers: first, that upstream had notified it DDR5 memory prices would rise another 10%-15% within the week, now formally in effect; second, that one unnamed memory supplier, after issuing its price-hike notice, outright canceled every previously effective open order awaiting delivery, forcing channel buyers to reorder at new prices 10%-15% above the originals.

A caveat on sourcing: “another 10%-15% within the week” comes from a single integrator’s notice to its own customers — a first-hand channel source, but limited in coverage, and not an industry index. What truly merits dissection is the second move: the supplier would rather absorb a breach-of-contract dispute than keep pending orders locked at the old price. That is extraordinarily rare in a normal business cycle, because the cost of tearing up effective orders is certain (contract disputes, damaged customer relationships), while the payoff is merely the expectation of “higher prices later.” The math only works if upstream judges that prices will keep climbing in the short term, and that the climb will very likely exceed the cost of default.

For the downstream, one last cushion remains in this announcement: Stince Built has committed to honoring all confirmed orders at original prices, absorbing the supplier-side increase itself. But the company also conceded that a 10%-15% jump exceeds what conventional margins can absorb, and that system prices will have to rise over the longer term. In other words, the industry understanding that “placed orders don’t get repriced” is still alive — how long it survives depends on when the increases punch through the integrator’s margin.

Taken apart, the announcement reveals three signals of deepening severity:

The cancellation of pending orders is such a strong signal because it also splits channel orders into two different fates. Two buyers may have placed the same order, but order status determines entirely different treatment:

Put differently, an “open order” means the price floats with the market; only a “confirmed order” locks the price. In a stable market that distinction hardly matters. In a market rising 10%-15% in a single week, it is a difference paid in real money.

2. The Huaqiangbei Spot Line: Memory Up Over 300% in a Year, GPUs Double in Ten Days#

The other side of the channel signal is the spot market’s hard transaction prices. Reporters from the Shenzhen Economic Daily’s Duichuang news app visited Huaqiangbei in person on October 21 and came back with numbers blunter than any announcement: memory module prices are up more than 300% over the past year. By generation, DDR4 16G modules went from under 200 yuan last October to a floor of 530 yuan for domestic brands and above 600 yuan for imported brands such as Samsung; DDR5 16G climbed from just over 200 yuan a year ago to more than 1,300 yuan, with a Kingston DDR5 16G module tagged at 1,300 yuan outright in one stall. Quotes for 32GB DDR5 kits were even more scattered, ranging from 2,100 to 3,800 yuan. Some merchants said pickup prices rose another roughly 5% in the past week and were holding near their yearly highs.

The GPU curve is steeper still. Mr. Liu, a Huaqiangbei GPU wholesale merchant, put it this way: the RTX 5090 went from 22,000 yuan on January 1 this year to 55,000 yuan, a 150% gain on the year; the most recent leg is even sharper — another round of violent surges began October 12, with prices rising almost exactly 1,500 yuan per day over the last ten days, a cumulative 15,000 yuan, roughly 50% packed into ten days. Over the same stretch the RTX 5080 rose from 8,000 yuan to 15,000 yuan, doubling in ten days.

That DDR4 is being bid up alongside DDR5 says more than DDR5’s own percentage. DDR5’s rise at least has a narrative to attach to — server demand and crowded-out wafer capacity — while DDR4 is a previous-generation product with no new demand story. When even the previous generation’s lowest domestic price lifts from under 200 yuan to 530 yuan, the spot rally is a broad-based advance across the whole category, not a single-segment supply-demand mismatch. That corroborates the “supplier tearing up orders” from Section 1: the second piece of evidence for channel financialization.

One easily missed detail: the field report noted that “pickup prices rose another ~5% in the past week and prices are holding near yearly highs.” The pickup price is what a stall pays the wholesale tier above it, and its rise propagates to the counter ahead of retail tags. Today’s shelf price, in other words, usually reflects the market of a few days ago; a ~5% weekly slope in pickup prices means retail restocking costs are still climbing day by day, and the rise in shelf prices most likely has not finished running its course.

The quote range itself is information: for the same 32GB DDR5 kit, quotes collected on the walk ranged from 2,100 to 3,800 yuan. The spot market has no unified posted price; the deal depends on negotiating leverage and pickup timing. That kind of price dispersion is unthinkable in the contract-price system, where procurement is priced quarterly and executed under agreements. For builders, it means the same kit of memory can get two completely different answers from two different stalls.

Two details deserve separate callouts. First, tracking data from European price-comparison platform DropReference shows the average price of a 32GB (2×16GB) DDR5-6000 CL30 desktop kit at about 120 euros in April 2025, surging to about 630 euros in early October 2026 — a more than fourfold rise in eighteen months. Huaqiangbei is no outlier; it is a cross-market move in the same direction. Second, Mr. Liu’s line that “high-end GPUs seem to have taken on a certain financial character” is the closest the field report comes to the essence of the matter — expanded in Section 5.

3. The Other End of the Divergence: What Storage Stocks Are Falling On#

Look only at the spot market and this is a one-way rally; switch the lens to the secondary market and the picture is completely different. From July through the October 21 close, A-share Biwin Storage fell a cumulative 66.74%, while Demingli, GigaDevice, Longsys, Puya, Shannon Semiconductor, Netac, and others lost more than 60%. Changxin Technology (CXMT) is down 25.22% in October, its total market cap slipping below 3 trillion yuan (about 2.78 trillion, dropping it to second place among A-shares).

Overseas markets are just as grim. In October, Samsung Electronics is down 17.76% and SK Hynix down 27.14%, drawdowns of 40.32% and 56.67% respectively from their June highs. On US exchanges from October 1 to October 20: Seagate fell 26.8%, SanDisk fell 21.52%, Western Digital fell 19.83%, and Micron fell 15.87%.

On one side, Huaqiangbei stalls where “a morning quote is dead by the afternoon”; on the other, storage stocks in a full rout. One industry, two markets, prices pointing in opposite directions.

Two details merit separate callouts. First, the decline spans three markets — A-shares, Korean stocks, and US stocks — and is concentrated in storage supply-chain names, which points not to a single-market flow disruption but to a synchronized repricing of the storage sector by global capital. Second, the drawdowns start from the June-era highs — SK Hynix is already 56.67% off its June peak, while spot prices were still setting yearly highs over the same period. The price discovery of a commodity and the price discovery of an asset parted ways within a single month.

The Changxin Technology numbers deserve a look of their own: down 25.22% in October, total market cap below 3 trillion yuan (about 2.78 trillion, slipping to second among A-shares). As one of the largest storage-sector listings in China, its drawdown against the stubborn Huaqiangbei spot market forms the most symbolic contrast of all — within the same month, the securities pricing and the spot pricing of domestic storage assets stopped pointing the same way entirely.

4. Two Pricing Logics: Why Spot and Stock Prices Tell Different Stories#

As for the cause of the divergence, the industry insiders and analysts the Shenzhen Economic Daily interviewed offered a clear framework: the memory industry runs on two pricing systems. Headline large electronics manufacturers buy directly from the original makers at contract prices refreshed quarterly, anchored in long-term supply-demand modeling; small and micro distributors and DIY consumers trade on the spot market at going rates, anchored in present-day panic and scarcity. The participants, cycles, and risk appetites of the two systems are entirely different, and spot channel transaction prices and secondary-market sector valuations have come apart accordingly. The interviewees’ conclusion: short-term spot volatility is not the same thing as the medium-to-long-term industry trend.

Put the contract-price side’s data on the table and you can see why capital markets are not panicking. In its October 12 semiconductor Q3 earnings preview, JPMorgan was positive on the long-term outlook for memory, judging that the supply-demand imbalance had intensified further over the past three months and that the spread of long-term supply agreements is changing the industry’s traditional price-cycle pattern; its global memory research team projects memory supply-demand gaps of roughly 20%, 19%, and 15% for 2026 through 2028. TrendForce, meanwhile, estimates DRAM contract prices will rise about 20% quarter-over-quarter again in Q4 2026 — a smaller gain than Q2’s roughly 80% and Q3’s roughly 40%, narrowing but still elevated; the reasons for the slowdown include an already-high base, weakened end-market purchasing power, and price-cap clauses in some long-term procurement agreements starting to take effect.

That dataset explains the stock logic: contract-price gains are narrowing, large customers are locked in by long-term agreements, and the supply-demand gap is slowly closing — valued on the quarterly contract-price system, the sector genuinely should not rise further. The spot logic is just as self-consistent: small and mid-tier merchants buying in Huaqiangbei never touch contract prices. They face a spot book where “don’t buy today, pay more tomorrow,” and a 300% annual gain and GPUs doubling in ten days are the only reality they can perceive.

Set the two systems side by side and the structure of the divergence is plain:

Worth noting are the contract price’s accelerator and its brakes. TrendForce attributes the core upward force to AI inference: larger models, longer contexts, and KV caches are all lifting server DRAM demand while HBM crowds out general-purpose DRAM wafer capacity; the reasons for the slowing gains are an already-high base, weakened end-market purchasing power, and price-cap clauses in some long-term procurement agreements starting to take effect. Between accelerator and brake, the estimated ~20% Q4 increase is really a composite signal: “still rising, but the marginal forces are weakening.”

Ordinary players live inside the spot system and perceive only the right-hand column; capital markets value the storage sector off the left-hand column. The two curves can point opposite ways in the short term — that does not mean either side is “wrong,” only that they were never pricing the same thing.

JPMorgan’s preview also contains an easily overlooked line: the spread of long-term supply agreements is changing the industry’s traditional price-cycle pattern. That sentence is key to understanding the divergence — when the largest players lock in capacity ahead of time through long-term agreements, the float left for the spot book gets thinner; the thinner the float, the wider the swing the same demand shock produces in spot prices. It also explains a counterintuitive pattern: why contract-price gains are narrowing (roughly 80% in Q2, 40% in Q3, an estimated 20% in Q4) while spot prices accelerated in the very same week. The two markets do not even share the most basic variable — float — so the divergence naturally feeds on itself.

5. Channel ‘Financialization’: From Torn-Up Orders to Sealed Warehouses and Waiting for Discontinuation#

Connect the moves of this run and a complete “financialization” transmission chain comes into view.

Step one is upstream rewriting delivery rules — the supplier canceling all pending orders mentioned earlier is, at its core, a refusal to “perform at low prices,” swapping contract thinking for inventory thinking. Step two is mid-chain hoarding — Huaqiangbei sellers are “sealing their warehouses and refusing to sell,” with the direct trigger being a rumor that began circulating October 10: NVIDIA would discontinue the RTX 5090 and 5090D v2, prioritizing the GB202 flagship chip for higher-margin professional cards. The rumor is unverified and NVIDIA has not confirmed it, but it was enough to shift market expectations — “it might get discontinued, hold for higher prices” became an openly stated reason to sit on stock in the stalls. Step three is the failure of the quoting system — the validity of a merchant’s quote compressed from “days” to “hours”; a rhythm of 1,500 yuan a day for ten days turns any quote sheet into an instant snapshot.

When “hold and wait for appreciation” displaces “fast turnover, move volume” as the channel’s dominant strategy, hardware completes its shift from commodity to quasi-financial asset: scarcity expectations are the fundamentals, the discontinuation rumor is the catalyst, sealed warehouses and reluctance to sell are the position-control technique, and the 1,500-yuan daily gain is the candlestick chart. Mr. Liu’s line that “high-end GPUs have taken on a certain financial character” describes exactly this.

Condense that transmission chain into a table and the moves and effects at the three links are plain:

Aggregate the corroborating data from around the world into a table and the breadth of the “financialization” becomes vivid:

Nor is this an isolated phenomenon. When original-makers’ capacity is sold out through 2027 and the entire chain, from system vendors to channel, is talking about “discontinued models” and “wait and it will cost more,” the channel’s behavior pattern is already indistinguishable from a trader’s — that is the full meaning of the word “financialization”: not how fast prices rise, but that the purpose of market participation has shifted from “moving the goods” to “holding the goods.”

6. The Ordinary Builder’s Ground-Level Judgment#

Both pricing logics hold on their own terms; the trouble is that ordinary builders stand exactly in the crack between them — no access to big-customer quarterly contract prices, no way to avoid transacting at spot-book prices. Below are this column’s opinion-based judgments drawn from the data above, for reference:

1. Confirmed orders are still protected; a must-have build calls for no panic-buying. Stince Built has committed to honoring confirmed orders at original prices — one of the few rules in the current channel that still works. Orders placed and confirmed through legitimate system vendors and first-party e-commerce storefronts have, in principle, locked in their transaction price; Section 1’s order-tiering table already showed that in today’s market, the word “confirmed” is worth a 10%-15% price gap.
2. The expected payoff of “waiting a bit longer” is shrinking. On the contract side, TrendForce estimates another ~20% quarter-over-quarter increase in Q4 2026; on the spot side, Huaqiangbei pickup prices are still rising week over week, and the European curve for 32GB DDR5-6000 CL30 kits — up more than 4x in eighteen months — has yet to show a down leg. Waiting for cheaper prices presumes a signal that “gains have turned negative,” and every visible data point currently reads “gains narrowing but still rising.”
3. Do not take part in stockpiling speculation. The discontinuation rumor is unconfirmed, and a 55,000-yuan RTX 5090 quote carries a large scarcity premium; if the rumor is debunked or supply recovers, speculative buyers at the top have no fundamentals underneath them. The spot-versus-stock divergence is itself the warning: spot prices can drift far from the industry trend, and the risk of chasing high and hoarding is entirely your own.
4. Evaluate the “spot surge” and the “industry trend” separately. Falling stocks do not mean spot falls immediately, and rising spot does not mean the gap widens forever — JPMorgan’s gap estimates (roughly 20%, 19%, and 15% for 2026-2028) narrow year by year, while the spot book has shown no reaction at all. Anchor your build decision to your own timing of need, not to the mood on either curve.

Whether DDR5 stabilizes in November, and whether this round of increases passes through fully to retail — the reporting answers neither. Until those questions land, any talk of “buying the dip” or “selling the top” is just guesswork.

Finally, three common misreadings bred by this market, cleared up:

1. “Storage stocks crashed, so spot prices must be about to fall.” The two pricing logics have different participants and different anchors; the spot book currently has no mechanism that follows stock prices. The reverse has held for the past four months — stocks stopped following spot long ago.
2. “The discontinuation rumor is everywhere, so the 5090 really is going away.” The rumor has not been confirmed by NVIDIA; paying a premium as if it were settled fact is the most fragile link in this round of “financialization” — once expectations reverse, that premium has no fundamentals holding it up.
3. “It is rising this hard, so stockpiling a little is a sure win.” Spot prices can drift far from the industry trend. Winning the hoarding game requires knowing the news before the stalls do and having more volatility tolerance than the supply chain — ordinary builders have neither.

Further Reading#

Three related pieces on this site complement one another in the order “official framing → retail impact → build decisions”; read on as needed:

• NVIDIA’s official framing of this shortage and the discontinuation rumors, in our report: NVIDIA Admits the GPU Shortage Is “Beyond Its Control”: DRAM Root Cause, New Highs at Every Price Tier, and a Restructured Build Economics
• The knock-on effects at GPU retail, in: The RTX 5090 Retail Market “Extinction”: How AI Demand Destroyed the Gaming GPU Price Structure
• The complete framework for build decisions, in: A Build Decision Guide for the Memory Price Storm: DDR5 Market, the New 32GB Mainstream, and the “Buy, Wait, or Switch” Strategy

Closing#

What is most worth recording about this run is not another few points of percentage gain, but the inflection point in the channel’s behavior pattern: a supplier tearing up effective orders, merchants sealing their warehouses for an unconfirmed “discontinuation,” quote sheets valid for hours. When hardware starts to be held on asset logic, price is no longer decided by cost and supply-demand alone — a layer of holders’ expectation games is stacked on top. That is the full weight of the word “financialization.”

For ordinary builders, what can be done is plain: tell must-haves from speculation, respect the protection boundary of confirmed orders, and do not put a build budget into a game in which you hold no informational edge. Spot prices will fall back, stock prices will recover — but both should happen after your build is done.

As for when this divergence converges, no public reporting has offered a timetable — whether spot returns toward the industry trend or stocks move toward spot, November’s action will deliver the first answer. Until then, watching two of the most visible signals is enough: whether suppliers resume performing at original prices, and whether the stalls’ sealed warehouses reopen.

Sourcing and Risk Notes: The “another 10%-15% within the week” figure in this article is a US integrator’s notice to its customers, not an industry index; the order-canceling supplier is unnamed; the RTX 5090/5090D v2 discontinuation is a rumor, unconfirmed by NVIDIA; Huaqiangbei quotes are individual merchants’ figures with no platform-level transaction data to back them. All opinion-based content in this article is labeled as such and does not constitute investment advice.