The market read the November 6 decline in memory stocks as the price-increase cycle peaking. With November revenue for Taiwan stocks due to be published on December 10, if Phison Electronics (8299) and ADATA (3260) still show double-digit year-on-year growth, does that count as the "peaking" thesis being overturned?

The market read the November 6 decline in memory stocks as the price-increase cycle peaking. With November revenue for Taiwan stocks due to be published on December 10, if Phison Electronics (8299) and ADATA (3260) still show double-digit year-on-year growth, does that count as the "peaking" thesis being overturned?

Not a reversal. The threshold for this test is more than thirty times lower than the two companies' current growth rates; whether it passes or fails carries no information.

What actually deserves attention has already been given by the Q3 reports: module makers' gross margins turned down across the board in the third quarter, while revenue is still hitting new highs. Year-on-year revenue is the last indicator to turn, not the first.

On November 6, memory did not lead the decline

First look at the actual trading that day. The weighted index closed at 39701.64, down 4.51%. Phison closed at 1500 yuan, down 3.85%, falling slightly less than the broader market; ADATA closed at 299.5 yuan, unchanged.

Only the wafer side fell more than the broader market: Nanya Technology -4.58%, Winbond -5.49%, also only about a point more.

The real selloff happened on November 3 and 5. Phison fell 9.87% and 8.77% on those two days, while the broader market fell 3.59% and 3.20% on the same days.

Over seven trading days, the weighted index fell from 47035.58 to 39701.64, down 15.59%; memory stocks fell 17% to 30%. This is the asset that rose the most in the first half falling more in a broader market pullback—normal amplification, not a memory-specific pricing move.

Reading this stretch as "the price-up cycle has peaked" is booking the broader market's losses to memory.

The double-digit growth threshold is so low it has no testing power

October revenue was reported on November 6 and 7; no need to wait until December. Phison's 315.30 yi hit a new high; ADATA's 170.63 yi is still below August's 189.25 yi.

Phison's November revenue last year was 70.22 yi; this October it was 315.30 yi. To keep year-on-year +10% in November, it only needs to do 77.24 yi—a 75.5% drop from October.

ADATA's November last year was 55.98 yi; this October it was 170.63 yi. To keep +10% it only needs 61.58 yi, a 63.9% drop from October.

In other words, even if Phison's November falls 20% quarter-on-quarter, year-on-year is still +259%; if ADATA falls 20% quarter-on-quarter, year-on-year is still +144%. These two are now running at triple-digit growth rates; setting the threshold at double digits tests nothing at all.

The year-on-year calculation uses last year's same month as the denominator. Last November happens to be the starting point of this round of price increases: Phison's 70.22 yi that month was only 22.3% of this October's; ADATA's 55.98 yi was 32.8%. When the denominator is smallest, the calculated growth rate is largest.

By next February, Phison's denominator rises to 121.98 yi. With the same 315.30 yi monthly revenue, year-on-year would be only +158.5%. The business hasn't gotten any worse; only the arithmetic has changed denominators.

There is also the exchange rate layer. The Taiwan dollar depreciated from 31.00 against the US dollar in November last year to 32.97, a 6.35% depreciation. Phison's October +346.3% is calculated in Taiwan dollars; converted to US dollars it is +319.6%, a difference of 26.7 percentage points; ADATA's +282.3% converted to US dollars is +259.5%, a difference of 22.8 percentage points.

So this question actually asks "will it collapse," not "has it peaked." Whether it has collapsed—just look at quarter-on-quarter; whether it has peaked—you have to look down at gross margin.

In the Q3 reports, module-layer gross margins have already turned down

Phison went from 65.35% in Q2 to 59.85% in Q3, down 5.50 percentage points. Its Q3 revenue of 860.76 yi was a record high, and gross profit of 515.18 yi was also a record high, but gross margin turned down.

Break it down quarter-on-quarter: revenue went from 678.88 yi to 860.76 yi, +26.8%; gross profit went from 443.65 yi to 515.18 yi, only +16.1%; operating profit went from 263.70 yi to 328.98 yi, +24.8%. For every extra yuan of business, new gross profit is only sixty-one cents.

Transcend fell harder: 77.03% to 56.50%, 20.53 percentage points; revenue simultaneously fell from 188.15 yi to 160.67 yi, -14.6% quarter-on-quarter.

Innodisk's gross margin went from 65.01% to 49.47%, -15.54 percentage points; revenue was still growing, 258.88 yi, +15.4% quarter-on-quarter.

Apacer 51.25% to 40.08%, -11.17 percentage points, revenue -11.3% quarter-on-quarter.

ADATA's Q3 report is not out yet; the statutory deadline is November 14. But its Q2 gross margin had already fallen from 55.70% to 42.11%, -13.59 percentage points in one quarter.

Of the six, only Team Group went against the trend: 42.77% to 49.27%, +6.50 percentage points.

Monthly revenue's peaking order gave the signal earlier than gross margin.

Transcend and Apacer peaked in April, ADATA in August, Innodisk in September, and Phison was still hitting new highs in October. The layer closest to retail spot quotes turns first; the further upstream and the more quarterly-priced, the later it turns. This sequence itself is the footprint of price passing from spot to contract, from terminal to upstream.

Phison's October was +2.9% quarter-on-quarter, its slowest month of 2026. The preceding nine months were +20.0%, +16.7%, +50.2%, +10.3%, +13.0%, +8.9%, +9.3%, +4.1%, +8.4%. This is the first time it has fallen below 3%.

The five companies other than Team Group share one shape: revenue year-on-year is still between +148% and +455%, while gross margin has already turned. October revenue year-on-year: Phison +346.3%, ADATA +282.3%, Transcend +191.6%, Innodisk +454.8%, Apacer +147.7%.

The top is not read from revenue. Revenue is "shipment volume times price"; price is still at a year-on-year high, so revenue year-on-year can rise for another two or three quarters. Gross margin is "selling price minus inventory cost"; once price stops rising, that spread is gone.

Upstream is still expanding, but that is another chain

Nanya Technology's third-quarter gross margin was 85.67%, another 6.18 percentage points above the second quarter's 79.49%; operating profit was 1068.94 yi, operating margin 79.98%. Winbond went from 66.25% to 70.57%, +4.32 percentage points.

Inventory days are also falling: Nanya Technology 140 days, 116 days, 99 days; Winbond 129 days, 116 days, 110 days. Goods on hand are getting thinner, not piling up.

In October revenue, both were still up quarter-on-quarter: Nanya Technology 464.76 yi, +3.1%; Winbond 293.61 yi, +3.9%.

At the wafer end, there is no sign of a top through October; gross margin, inventory days, and monthly revenue are all moving up.

I cannot get first-hand data on contract prices. This sits in TrendForce's paid database; public channels only have secondhand accounts, so I do not use it as evidence and use only the three traceable numbers from financial reports: gross margin, inventory days, and monthly revenue.

But one thing needs to be separated out here: Nanya Technology and Winbond make DRAM; Phison makes NAND controllers and modules. Two chains, two kinds of business. So "upstream expanding, modules contracting" may not be an upstream-versus-downstream difference but rather DRAM being tighter than NAND.

I cannot rule out this alternative explanation, only narrow it. ADATA does both DRAM and NAND lines, and its second-quarter gross margin already fell 13.59 points; Transcend likewise does both, and its third-quarter fell 20.53 points. The timing of the decline differs, but the direction is the same. The module layer, whatever the product, is being squeezed.

To separate this completely, I need ADATA's third-quarter report. The answer will be clear before November 14.

Which line the money leaks from

The transmission chain is short. Wafer prices rise; module makers still hold inventory bought before the price increase; selling prices rise while costs do not, so gross margin is artificially high; once inventory turns over, materials are bought at the new price; selling price and cost converge again, and gross margin returns to a processing level.

On the statements this lands in "operating cost," not "operating expense," and not "asset impairment." This matters — it means the effect seeps out slowly, not as a one-day charge.

There is an accounting detail worth spelling out. Inventory is measured on the statements at the lower of cost and net realizable value; when prices rise, inventory is not revalued upward. That excess profit does not appear in "fair value changes"; it shows up directly in gross margin at the moment the goods are sold, when low-cost inventory is matched against high-price revenue.

Once restocking is complete, this effect disappears automatically. It has nothing to do with whether management performs well; it is determined by accounting rules. No one can recreate an inventory price spread that has already disappeared through operations.

Phison can verify this direction. Third-quarter gross margin fell 5.50 percentage points, while the operating expense ratio actually fell from 26.46% to 21.48%, so the operating margin fell only from 38.84% to 38.22%, 0.62 percentage points. The expense side is not broken; what is broken is the spread on the cost side.

Phison's inventory at the end of the third quarter was 1117.74 yi, 93.5% of shareholders' equity of 1195.19 yi. This batch of goods is all its elasticity: if NAND rises another ten percent, it is profit; if it falls ten percent, it is impairment.

But its inventory days fell from 355 days to 294 days; it is selling faster than it is buying. Of the six companies, four saw inventory days fall; only Apacer and Team Group rose, with Team Group rising from 296 days to 305 days. So the claim of "can't sell, goods piling up" does not hold up in the data.

The real change is not inventory but pass-through. Upstream took the entire price increase; module makers buy materials at the new price and cannot add that much back when selling.

There is one more layer of time lag to remember. Monthly revenue counts deliveries; from customer order to delivery to revenue recognition, one to two months pass, and order cuts follow the same rhythm. So good November revenue measures September and October orders; the real temperature in November will show up on the statements at the earliest in January or February next year.

What is called a peak is actually three different peaks. Wafer contract prices no longer rising is the price peak; module makers' gross margin no longer expanding is the gross margin peak; monthly revenue no longer hitting new highs is the revenue peak.

The timing of the three peaks does not follow a fixed order. Transcend's revenue peak was in April, and its gross margin peak was in the second quarter; Phison's revenue peak has not yet appeared, while its gross margin peak has already passed. Since the order is not fixed, one cannot use "revenue is still rising" to negate "gross margin has already peaked."

The opposition's strongest argument, and the three numbers I concede

A smart person would say: upstream is still expanding gross margin, supply is still tight, Phison's inventory days are still falling, so the third-quarter gross margin decline is just low-priced inventory being used up; in the fourth quarter, inventory will be replenished at new prices and selling prices will continue to rise, so gross margin will expand a second time.

An even harder point: Team Group's third-quarter gross margin rose 6.50 percentage points against the trend, showing that "the module layer collectively peaking" is overstated. Phison itself also said AI-related business accounted for 65% of October revenue, with the mix shifting toward enterprise-grade; enterprise-grade gross margin is inherently lower than the windfall consumer modules enjoy when prices are being raised, so the 5.5-point drop may be mix, not price.

There is also one fact: on November 6, the day of the decline, Phison announced a plan to buy back 2,000 treasury shares.

None of these three is false. But what they rebut is "gross margin peaking means a continuous decline ahead," not "year-on-year revenue can prove there is no peak." On the former I may be wrong; the latter is arithmetic.

The first conceded timing point is ADATA's third-quarter report before November 14. If its gross margin holds above 42.11% without falling, then "the module layer collectively turning down in the third quarter" shrinks to only four companies, and the conclusion must be narrowed to "branded modules turn first, industrial control and niche modules later by one to two quarters."

The second is Phison's fourth quarter, to be published before March 2027. If single-quarter gross margin returns above 62% and inventory days remain below 300 days, then my reading of it as "low-priced inventory used up" is wrong; the real reason is product mix, and those 5.5 points are not a cycle signal.

The third is November revenue, which in practice is mostly reported between December 3 and 7, with no need to wait until the 10th. If Phison's quarter-on-quarter growth exceeds +8% and ADATA's exceeds +15%, while contract prices are confirmed to keep rising in the fourth quarter, then the judgment "the price cycle has peaked" is void.

Conversely, if ADATA's fourth-quarter revenue really declines slightly as the company itself said at its November 6 earnings call, while fourth-quarter year-on-year revenue is still triple-digit, then the flaw in this question is fully demonstrated: year-on-year revenue and cycle direction can be completely opposite.

One sentence

November year-on-year revenue in double digits proves these two companies have not collapsed; it does not prove the price-raising cycle has not peaked. To see the peak, look at quarter-on-quarter single-quarter gross margin, inventory days, and contract prices—one of the three has already turned.

Sources: Phison Electronics 2026 third-quarter consolidated financial report (Market Observation Post System, filed 2026-11-08); Phison Electronics October 2026 consolidated revenue filing (disclosed 2026-11-07, monthly revenue filings have no announcement number); ADATA Technology 2026 second-quarter consolidated financial report and October 2026 consolidated revenue filing (disclosed 2026-11-06, no announcement number); Transcend Information, Innodisk, Apacer Technology, and Team Group 2026 third-quarter consolidated financial reports; Nanya Technology and Winbond Electronics 2026 third-quarter consolidated financial reports; Taiwan Stock Exchange 2026-11-06 centralized market daily trading information (TAIEX and closing prices of individual stocks); ADATA Technology 2026-11-06 investor conference fourth-quarter outlook, Phison Electronics 2026-11-06 investor conference and treasury share buyback plan (MoneyDJ, cnyes 2026-11-06 reports).