Macronix (2337.TW) Q2 2026 — The Mature-Node Fab Caught the Cycle, and the Depreciation Bill Waits Until 2028

The setup

Macronix is the Hsinchu specialist that spent two decades being the wrong kind of memory company. It does not make DRAM. It does not chase leading-edge 3D NAND layer counts. It makes NOR flash, SLC/MLC NAND, eMMC and mask ROM on trailing-edge nodes — 55nm NOR, 48-layer and 96-layer 3D NAND — for automotive, industrial, networking, and embedded customers who care more about a twenty-year supply guarantee and a qualification file than about cost per bit. For most of the last cycle that positioning meant structurally thin margins and, in 2025, a full-year loss of NT$3.31B on NT$28.88B of revenue.

Then the AI buildout arrived at the boot layer. Every accelerator, switch, and NIC in a rack needs high-density NOR to hold firmware, and NVIDIA's generational cadence has been multiplying that content per box; every AI server needs SLC NAND and eMMC in places where a TLC drive is the wrong part. At the same time Samsung exited MLC, tightening a niche that already had almost no marginal supply. The result is a shortage in exactly the mature-node parts nobody had built capacity for in a decade, priced monthly, at a company with a fully depreciated fab. That is the whole print.

The numbers

Records on every line that matters. Net sales of NT$19,125M (US$604.8M) rose 83% sequentially and 181% year-on-year. Gross profit was NT$12,320M (US$389.6M) at a 64.4% gross margin — up 188% QoQ and 1,058% YoY, and 23.6 percentage points above Q1's 40.8%. Operating income was NT$8,968M (US$283.6M) at a 46.9% operating margin, up 364% QoQ. Net income after tax reached NT$7,736M (US$244.6M), or EPS of NT$3.91, against book value of NT$31.28 per share. Operating cash flow was NT$7.8B and free cash flow NT$7.3B.

The quarter reconciles cleanly against the monthly disclosures, which is worth doing because Macronix reports revenue twelve times a year and several figures are in circulation. April NT$5.913B + May NT$6.256B + June NT$6.956B = NT$19.125B — exactly the reported quarterly net sales, with each of the three months setting a record in turn. Q1 backs out the same way: NT$29.593B first-half less NT$19.125B leaves NT$10.468B. If you have seen a larger top-line number attached to this print, it is almost certainly one of three other measures — first-half revenue of NT$29.59B, trailing-twelve-month revenue of about NT$45.5B, or a forward Q3 figure (local coverage has framed Q3 as capable of challenging NT$20B, which is commentary, not company guidance). The quarter itself is NT$19.125B.

The cumulative figure is the one that reframes the company. First-half revenue of NT$29.59B (+128.8% YoY, against NT$12.94B) already exceeds all of fiscal 2025's NT$28.88B, and first-half EPS of NT$4.80 has erased the entire prior-year loss in two quarters. Q1 was the hinge: it ended a run of ten consecutive quarterly losses, carried gross margin to 40.8% and swung operating margin from negative to 18.5%. July did not slow down: monthly net sales of NT$7.725B, +11.0% MoM and +180.5% YoY, the first month above NT$7B in company history — a pace that annualizes to roughly NT$23.2B for Q3, another ~21% above Q2 if it merely holds.

One housekeeping note that matters for anyone comparing write-ups. The same NT$19.125B has been published in at least three different dollar figures — US$604.8M, US$595.6M and US$589.2M — and none of them is wrong. The company's own release converts the entire P&L at roughly NT$31.62/US$, which is within a rounding error of the 2026 average of 31.627; press coverage converts at spot on its own publication date, and USD/TWD ran as high as 32.41 in mid-July. On a 181% growth rate the FX choice is noise, but it is enough to make two accurate write-ups look like they disagree. The NT$ figures are the authoritative ones.

One line deserves to be pulled out rather than buried. CFO Yeh Pei-Fu disclosed that approximately NT$900M of the quarter's gross profit came from a reversal of prior inventory write-downs. Strip it out and gross margin is roughly 59.7% rather than 64.4% — the write-back is worth about 4.7 points. That does not make the quarter less real; a reversal is the accounting system admitting that inventory written down in the 2025 trough is now worth more than it was carried at, which is itself a price signal. But it is a non-cash, non-repeatable component of a headline number, management confirmed there is more of it available in the second half, and any model that extrapolates 64.4% forward without separating the two is extrapolating the wrong base. GM Lu Chih-Yuan was direct that the rest is real: "The majority comes from ASP; price increases are the most important factor."