Phison Q2 FY26: The Margin Came Out of the Warehouse, and the Warehouse Now Costs 2026 Prices

A record nobody should argue with

Start with what holds, because most of it does. Revenue was NT$67.89B, and the three monthly releases, April NT$20.21B, May NT$22.83B and June NT$24.85B, add up to it to the thousand. Gross profit was NT$44.34B (about US$1.40B). Operating income was NT$26.37B (about US$0.83B), a 38.8% margin. The first half alone brought in NT$108.86B (about US$3.44B) against NT$72.66B (about US$2.30B) for all of 2025. Every US$ figure here uses about NT$31.61 per US$, the company's own Q2 average rate, and is approximate.

Two things about the EPS line, then I'll leave it alone. About NT$14.21 of the NT$118.57 came from investing gains, mostly a NT$3.70B (about US$0.12B) share of profit from equity-method investees. And the company is no longer mainly a controller business. Controllers were 6% of Q2 revenue. The CEO put it himself on the call: "yes, we are module house," per a third-party transcript (unverified). That matters for everything below, because a module house earns its gross margin on NAND it bought.

The warehouse is the position

Here is the balance sheet the margin came from. Inventory was NT$91.79B at June 30, against NT$72.20B in March and NT$35.61B in December. That is 3.90 times the quarter's operating costs of NT$23.52B. The company's own inventory days were 297. For a buyer of NAND in a price spike, the warehouse is the position: the gross margin is the spread between what the flash cost when it went onto the shelf and what the finished drive clears at now. A 65.3% spread on a module is not a design margin. It is the holding period, paid out.

Now read the date stamp on the shelf. Of the NT$91.79B, at least NT$56.18B (about US$1.78B), or 61.2%, was added between January and June. Whatever cost method the auditor signs off on, most of what Phison sells in the third quarter was bought in the first half of 2026, at first-half prices, not at the prices of the 2025 stock that fed the first half's margins. The CEO also said the company "never revalue[s]" inventory, it only writes it down (unverified, third-party transcript). So there is no revaluation gain waiting to flatter Q3. The cost layer is just catching up.

One detail cuts the other way and I'll give it its due. Write-downs took about 1.10% off Q2 gross margin and about 1.50% off Q1. A company writing inventory down in a rising market is being conservative, and a Q3 with smaller write-downs gets a small head start on 65.3%.

The average is still rising. The increment is not.

The deck gives six quarters of gross margin: 30.93%, 29.06%, 32.41%, 41.67%, 61.31%, then 65.31%. That looks like a line still going up. Read the ingredients, not the level. Gross profit rose NT$19.22B in Q2 on revenue up NT$26.92B, so the margin on the extra revenue was 71.4%. In Q1 the same calculation gave 86.0%, and in 4Q25 77.7%. The average keeps climbing only while the increment stays above it. At 71.4% against 65.3%, the gap is 6 points and closing.

The CEO described the same curve in his own words on the call. NAND prices would keep rising, "but the ratio getting smaller and eventually saturate," and on memory makers' margins, "Price definitely will still going up, but 85%, more than enough" (both unverified, third-party transcript). Note that I am not the one calling the price curve a flattening one. He is. A module house that sells the price spike on a ten-month shelf earns the most when price growth is fastest, and that was Q1.

The AI label and the incremental dollar

The release says "the structural wave of storage demand driven by AI continues to accelerate." The deck shows where Q2's growth actually went. AI ecosystem modules, the enterprise SSDs, boot drives and aiDAPTIV, were 38% of revenue in Q1 and 38% again in Q2, on growth of 68%, about in line with the company as a whole. Embedded ODM modules, the PC and mobile OEM drives and gaming, grew 109% and went to 33% of revenue from 25%. The fastest-growing line in an AI quarter was drives for PCs and phones.

That is not a knock on the AI business, which grew 68% in a quarter. It is a point about what the margin was earned on. Embedded ODM drives are NAND sold to OEMs at the same spot and contract prices everybody else pays, and they took the largest share of the growth. The composite hides the component: a margin that rose while the mix tilted toward commodity-priced OEM drives is price, not product.

Price-long, cash-late, and who paid for the shelf

Differencing the statements isolates the quarter. First-half operating cash flow was negative NT$14.87B (about US$0.47B); Q1 alone was negative NT$13.85B (about US$0.44B). So Q2 was negative NT$1.02B, on pre-tax income of NT$31.07B (about US$0.98B). By the price-long, cash-late test, a narrowing outflow while the balance keeps building is what a stockpile converting into cash looks like. That is the generous reading. The composition says something else.

Inventories took NT$20.34B (about US$0.64B) and receivables NT$15.94B (about US$0.50B). Suppliers did not carry it. Accounts payable fell NT$7.01B (about US$0.22B) after rising NT$11.67B in Q1, which is what paying the flash makers faster looks like. What held the quarter near zero was other payables, up NT$14.07B (about US$0.45B) to NT$31.66B, which the deck describes as including the cash dividend, income taxes, salaries and accrued employee compensation. Those are bills for profits already booked. They come due.

So who paid for the shelf? Bondholders. Bond issuance raised NT$25.20B (about US$0.80B) in the quarter, NT$31.25B for the half, and bonds payable went to NT$25.41B from NT$5.14B in March. Part of it went to repay NT$12.84B (about US$0.41B) of short-term loans, and cash rose NT$10.61B (about US$0.34B).

The dividend is the company's own cash forecast

The board proposed NT$60 a share (about US$1.90), "approximately 32% of 1H26 EPS of NT$187.43." The release gives the reason in plain words: it "will maintain a more prudent cash position to ensure operational flexibility and reduce potential financing costs," citing "continued raw material supply constraints." On the call the CEO said the old policy tried "to match 35% of the net profit, but this time we decide not to do that," and then, flatly, "We need cash" (unverified, third-party transcript). A company with ROE of 102.09% that cuts its payout ratio is telling you the profit has not reached the bank. It is on the shelf, and the shelf keeps getting restocked at higher prices.

What the peers did with the same shortage

Three Taiwan storage names, one quarter, three answers. ADATA pushed volume, grew revenue 46.15% QoQ and gave gross margin back to 42.11% from 55.69%. Team Group held its stock and let revenue fall 16.11%. Phison grew revenue 65.7% and still took gross margin up to 65.3%, because its shelf was deeper and older than either. Silicon Motion, the merchant controller vendor that holds no NAND, printed a 50.2% non-GAAP gross margin. A module house out-earning the pure controller designer on gross margin is being paid for the flash it held, not for the silicon it designed.

Grading June

The Q1 FY26 deep dive of June 5 said the NT$72B inventory line, not EPS, would decide what came next. On the frame, that held. The inventory paid rather than bit: it grew to NT$91.79B, write-downs shrank to about 1.10% from 1.50%, and the margin rose. One correction to that page: it cited FY2025 EPS of NT$39.74 from a third-party aggregator; the company's decks give NT$41.98 basic. And the AI share it watched did not move. It was 38% in Q1 and 38% in Q2.

The strongest case against me

Here is the other side at full strength, and it's not weak. The CEO said prices "will still going up," which on his own reading means the spread keeps widening for a while even as the increment shrinks. He said the company bought NAND wafers from Shenzhen module houses in June and July at prices "much cheaper than the NAND contract price," which is exactly the kind of cheap layer that could lift Q3. He said the inventory is short, not long: "most likely we're still short in some part by end of this year and the early next year" (all three unverified, third-party transcript). The release says the inventory is "mostly applied to the Non-Retail market," where pricing is contracted. And Q2 absorbed about 1.10% of write-downs that a clean Q3 would not. Believe all of that, and a Q3 margin above 65.3% is the ordinary result of a price curve that hasn't turned.

I still land on the other side, for one reason. Price rising is not enough. It has to rise faster than the cost of a shelf that was 61.2% restocked in 2026, and the increment already fell from 86.0% to 71.4% while prices were rising fast. Cheap spot wafers are a quarter of opportunity. A 297-day shelf is the whole book.

What decides it

One number in one report: Q3 gross margin, expected in early November. At or below 65.3%, the holding period has caught up and I'm right. Above it, price is still outrunning the cost layer and I'm wrong.

Two earlier reads. September revenue is due by October 10, 2026. July was NT$27.16B (about US$0.86B) and August NT$28.28B (about US$0.89B), up only 4.10% on July after gains of 10.32%, 12.96%, 8.86% and 9.29%. Two months sum to NT$55.44B (about US$1.75B). A September equal to August puts Q3 at NT$83.71B (about US$2.65B), +23.3% QoQ against +65.7% in Q2, and slower revenue growth is the condition under which the cost layer catches price. Then, in the Q3 statements, trade payables against NT$13.64B and operating cash flow against negative NT$1.02B. If the accruals that carried Q2 are paid and inventory still grows, the bondholders are funding a second turn of the shelf. What I cannot tell from anything published is which company sits behind the NT$3.70B equity-method gain; the filing pages I have do not name it, and nobody on the call asked.

The trade (analysis, not advice)

Phison at this point is a NAND-price instrument with a controller company's R&D budget. R&D was NT$15.10B (about US$0.48B) in the quarter, up 74.1%, and that spend does not fall when the spread narrows. The setup into November is asymmetric the wrong way for anyone paying for the margin line: a Q3 above 65.3% mostly confirms what the record already shows, while a flat or lower margin, with operating expenses climbing, lands on operating leverage from both sides. Anyone who wants the flash cycle without holding someone else's price risk on a ten-month shelf can own it one layer up, through picks-and-shovels capture at the makers that set the contract price.

Bottom line

Phison printed the best quarter in its history, and the margin that made it came out of a warehouse stocked before the prices rose. That warehouse is now NT$91.79B, most of it bought this year, with trade payables falling rather than rising under it, and financed with NT$25.20B of bonds. My view is that Q3 gross margin prints at or below 65.3%. The report due in early November settles it.