The setup
Nanya sold fewer bits in the third quarter and still grew revenue 61.9%, because price rose more than 60% and the deck says why: “Q3’26 ASP increases primarily reflect renewed long-term agreements.” That one footnote is the quarter. In Q2 the price was a rent collected on a market the top three left short. In Q3 the rent is written into contracts with customers — and with shipments guided flat until Q4’27, the contract price is the only variable left. Our test: if October–December monthly revenue averages above NT$44.55B, Q3’s monthly average, the contract rent is holding. If it averages below, it has already rolled.
The Q2 dive closed on a question: could a supplier with 1.6% of the DRAM market, on TrendForce’s 1Q26 count, turn a rent it does not control into a node position it does? Four weeks after that print the board committed up to NT$346.6B (~US$10.84B at ~31.96 TWD/USD, approximate) to Fab 5A. Q3 does not answer the node question. It answers the funding question, and it answers it early.
The headline: revenue NT$133,649M (~US$4.18B, +611.7% on the year), net income NT$87,627M, EPS NT$25.40. The line underneath it matters more. Bit shipments fell by a mid-single-digit percentage on the quarter, and management says they “stay flat until new fab wafer output Q4’27.” Four more quarters with no volume lever. Whatever Nanya earns between now and then, the contract book earns it.
The numbers
| Net sales | 133,649 · +61.9% QoQ · +611.7% YoY |
| Cost of goods sold | 19,167 · +13.2% QoQ |
| Gross profit · gross margin | 114,482 · 85.7% (Q2: 79.5%) |
| Operating expenses (R&D 5,145 · SG&A 2,443) | 7,588 · 5.7% of sales |
| Operating income · operating margin | 106,894 · 80.0% (Q2: 73.7%) |
| EBITDA · EBITDA margin | 110,002 · 82.3% |
| Income tax | 21,705 · NT$9.6B more than Q2 |
| Net income · net margin | 87,627 · 65.6% |
| EPS · book value per share | NT$25.40 · NT$118.99 |
| ASP QoQ · bit shipment QoQ · FX | >+60% · down mid-single digit · favorable low-single digit |
| Operating cash flow · capex · free cash flow | 96,017 · 20,783 · 75,234 |
| Net cash (cash 288.5B less LT debt 18.4B) | 270,100 (Q2: 199,200) |
| Incremental gross margin (derived) | 95.6% (Q2: 96.5%) |
| Implied D&A, EBITDA less op income (derived) | 3,108 · 2.3% of sales |
| Capex as share of operating cash flow (derived) | 21.6% |
Reported lines from Nanya’s Q3’26 press release and investor conference deck (). The three copper rows are derived from those disclosures, not company-stated.
The company prints the revenue bridge on one slide and leaves it there. Revenue +61.9%. ASP “increased >60%.” Bit shipment “decreased mid-single digit%.” Exchange rate “favorable low-single digit%.” Read the ingredients, not the level. The headline barely moved from Q2’s +68.2%, and that stability is what hides the change: in Q2 price carried a flat volume line; in Q3 price had to carry a falling one. The same growth rate now rests on one variable instead of two.
Then the cost line. Cost of goods sold rose 13.2% to NT$19,167M while revenue rose 61.9%. Gross profit rose NT$48.9B on NT$51.1B of extra revenue — an incremental gross margin of 95.6%, after Q2’s 96.5%. The company calls it “mainly due to higher ASP.” Decode: an incremental line that far above the 85.7% average means the average has not finished climbing. Nothing in Q3’s cost structure caps the margin. Only the price can.
Depreciation explains why. EBITDA of NT$110,002M less operating income of NT$106,894M gives implied D&A of NT$3,108M — 2.3% of sales, down from 3.6% in Q2. The asset base did not grow; the denominator did. Price does the work the capex hasn’t done yet: the plant that earned this quarter was paid for during 2023–2025, when capex ran NT$13.2B, NT$16.1B and NT$13.4B, and it is now carrying an 80.0% operating margin.
One cost line did jump. R&D rose 59.3% to NT$5,145M and SG&A 56.3% to NT$2,443M, and the deck gives one reason for both: “concurrent and future employee compensation.” At 5.7% of sales it does not move the margin. The word that matters is future: the company is already accruing for the people Fab 5A will need.
The rest of this dive is for paid subscribers.
The headline numbers are above. The 6 sections below carry the mechanism, the peer read, the valuation work, and the dated tests that decide it.
- What’s actually driving it
- How peers read this
- Valuation reality check
- What we’d watch from here
- The trade
- Bottom line
Already a subscriber? Sign in.
What’s actually driving it
Two things, and the deck keeps them apart.
The rent moved onto paper. The Q4 outlook slide uses LTA three times: suppliers’ “long-term capacity planned based on LTA commitments,” LTAs “supporting market stability,” and “new fabs planned generally aligned with future AI projects under LTAs.” Management frames this as stability. Read it as allocation. A commitment is an allocation claim before it is a cost: every mature-node wafer a customer locks up is a wafer no one else can bid on, which is why the contract price can climb more than 60% in a quarter in which the seller shipped less. Q2’s price was a shortage. Q3’s is the shortage under contract.
AI reached the mature nodes, and now the company says so. AI infrastructure — servers, networking, storage, system management — is “more than 20%” of revenue. The Q2 dive had to infer that mechanism from TrendForce; Q3 discloses it. HBM crowds out DRAM, and the server OEM pays the rent: Nanya makes no HBM, yet the boards around every accelerator still need conventional DRAM, and the deck says AI “drives DRAM demand beyond HBM and LPDDR5, across networking, storage, and system management.”
Read the same slide for what management lists without stressing. Supply to “smartphones, PCs, and general consumers” is “suppressed”; buyers are moving to “down-specification or extended lifecycles”; “recycled DRAM partially supports low-end market.” That is a consumer market at its price ceiling, described by the seller. The 20% AI share is what is contracted. The other ~80% is closer to that ceiling than the margin suggests.
The capex wave has started, and most of it is still ahead. Q3 capex was NT$20,783M, against NT$4,046M in Q2 and about NT$2.8B in Q1 (from the NT$27.6B nine-month total). The board’s 2026 ceiling of NT$69.7B implies up to NT$42.1B in Q4 alone. Treat a one-period change as a phase, not a decision, until you have the series — and this series is the build starting: three years of maintenance, then a fivefold quarterly jump, then the ceiling allows the same again. Equipment moves in Q1’27, ramping to 30,000 wafer starts a month by the end of 2028 and 50,000 by 2030, with a 3D wafer packaging and test fab at Pingtung by the end of 2028.
How peers read this
Put Nanya’s 85.7% gross margin beside Micron’s Q4 FY26 non-GAAP 87.0%. A mature-node supplier with no HBM sits within 1.3 points of the U.S. leader. In this cycle, gross margin no longer separates the leader from the follower. What separates them is who is underwriting the next build: Micron disclosed customer deposits as collateral behind take-or-pay floors; Nanya now says new capacity is “generally aligned with future AI projects under LTAs.” Different instruments, one direction — the customer is funding supply it wants reserved.
For Winbond and PSMC, Nanya’s quarter is a read on contract duration: if LTAs set the price, it falls more slowly than spot would, which supports the cohort. For the module houses and distributors that buy this DRAM, the read inverts — a contracted input is a cost they cannot shop around this cycle.
Valuation reality check
Nanya closed at NT$527 on , before the results, which came out after the Taipei close. On 3,450M shares that is about NT$1,818.2B (~US$56.89B). Annualised Q3 EPS of NT$25.40 is NT$101.60 — about 5.2x. At about 4.4x book of NT$118.99, with net cash of about NT$78.29 a share (14.9% of the price), the market is pricing the quarter as a peak, not as a run rate.
The balance sheet is where the Q2 framing breaks. Net cash went from NT$199.2B to NT$270.1B in one quarter. Fab 5A’s first phase is NT$346.6B for 35.9K wafer starts a month across 2026–2029; net cash already covers 77.9% of it. Add one more quarter of Q3’s NT$75,234M free cash flow and the total is NT$345.3B — 99.6% of the budget. The Q2 dive called Fab 5A a bet against Nanya’s own windfall. On these numbers the windfall pays for the bet before the fab produces a wafer.
What it does not pay for is the depreciation. The carrying cost is senior to the shareholder, and Fab 5A’s starts when the tools move in. As an author’s illustration only: NT$346.6B straight-line over five years is about NT$17.3B a quarter, roughly 5.6 times Q3’s NT$3,108M of D&A. Nanya has not disclosed an asset-life schedule, so take the direction, not the decimal — the 2.3% depreciation line that made Q3 possible is the line Fab 5A resets.
The best case against this. A bear would say we have the order wrong: contracts did not lift the price, the price lifted the contracts. Customers sign LTAs at the top of a shortage to secure supply, and an LTA renewed at a 60%-higher ASP is a ceiling written down, not a floor. If spot rolls, a renewal clause or a volume shortfall can reprice the book faster than the word “long-term” implies, and a supplier that already guided flat volume has nothing to offset it. The 5.2x multiple says the market agrees. That is a serious case and the deck does not refute it — it discloses no LTA duration, no price-reset terms and no minimum volumes. We still land on the other side, for a reason in the cash flow: Nanya says its new capacity, up to 50,000 wafer starts a month by 2030, is planned against “future AI projects under LTAs,” and a supplier does not size a fab to agreements its customers signed expecting the price to fade inside a quarter. The monthly releases will settle it faster than the argument can.
What we’d watch from here
- October–December monthly revenue vs NT$44.55B. Q3’s NT$133,649M averages NT$44.55B a month. With shipments flat by guidance, a Q4 average above that line means contract price is still rising; below it, the rent has rolled. This is the call.
- Q4 capex vs the NT$42.1B implied. A Q4 far short of the ceiling means the build slid into 2027 — and with it the depreciation reset and the Q4’27 output date.
- The AI share above 20%. Rising means the LTA book is migrating toward demand that is not price-capped. Stalling while consumers keep down-specifying means the mix is the risk.
- 1C customer sampling in Q1’27. Node progress is the only route from a rent to a position. 1D, 1E and EUV are “on track”; 1C qualification is the first dated milestone.
The trade
Long-only, evidence-led, and paced by the calendar rather than the print. At about 5.2x annualised Q3 earnings the market already prices a peak; the question for a holder is not whether Q3 repeats but how slowly it fades. The monthly revenue releases, around the 10th, are the checkpoints — each month above NT$44.55B is a month the contract thesis holds. A holder who wants to stay exposed through the build and be paid for the volatility could overwrite an existing position, sized to the monthly checkpoints rather than the Q4 print. We have not priced strikes here.
The risk to the long view is not execution. It is the ~80% of revenue that sits outside the disclosed AI share, already strained on the company’s own slide, giving way faster than the AI share can grow into it.
Bottom line
Q2 showed a rent. Q3 shows the rent under contract, an AI share the company now discloses instead of one we had to infer, and a balance sheet one quarter away from covering Fab 5A’s first phase. Volume is flat for a year by management’s own guidance, so the next four quarters test one variable — the contract price. The call: October–December monthly revenue averages above NT$44.55B. Wrong if it averages below.