A 7.5% Margin at Record Scale, Financed on Wiwynn's Own Balance Sheet: Initiating at Outperform, 40% Below the High
Key Takeaways
- Revenue was already public; the operating margin was the news, and it held. Monthly filings had put third-quarter revenue of NT$434.6bn (~$13.2bn, +62.9%) in view by October 8. What the board added on November 6 was operating income of NT$32.5bn, up 60.5% on the quarter at a 7.47% margin, as operating expenses fell to 1.40% of revenue and more than offset a 39bp slip in gross margin to 8.87%. Diluted EPS of NT$38.99 (~$1.18) beat the NT$34.8–35.2 consensus by about 11%.
- Interest is now a real cost line. A NT$0.8bn non-operating gain a year ago became a NT$2.0bn loss, almost all of it interest on the bank debt that carries the build. It cost NT$2.77 of EPS, and basic EPS of NT$41.46 grew 49.1% against 65.8% for operating income.
- Wiwynn is financing the ramp on its own balance sheet, and now with equity. Inventory reached NT$308.4bn (~$9.3bn, 45% of assets), materials bought on others' behalf NT$110.3bn, and nine-month operating cash flow was negative NT$56.8bn. The same board meeting approved up to 18 million new shares for a GDR. With the November 6 bond conversion, committed dilution is about 7% of the September share count, and about 11% if every convertible converts.
- The fourth quarter starts higher, with a supply caveat. October revenue of NT$165.1bn fell 4.3% from September but ran 14% above the third quarter's monthly average. The board warned that tight raw-material supply "may cause short-term monthly revenue fluctuations" and named US midterm policy as something it will monitor.
- Rating: Initiating at Outperform, with a 12-month value range of NT$1,850–2,050. At NT$1,570 the shares are 40% below their September high and trade at about 11.5x our 2026 EPS and 9.9x our 2027 diluted EPS of about NT$158, an estimate below the Street's. The drawdown came on macro and policy days while estimates rose. Funding and customer concentration are the real risks, and they are why our range sits well under most sell-side targets.
Results vs. Consensus
Two conventions govern this section. First, Wiwynn's revenue is not news on print day. Taiwan-listed companies publish revenue for every month by the tenth of the following month, and July, August and September revenue summed to the reported NT$434.618bn to the thousand. That total was public about four weeks before the board met. A revenue "beat" against a consensus figure set before the monthly filings is stale arithmetic, and we do not score it as one. The information the board released on November 6 was the cost structure: gross margin, operating expenses, the non-operating line, tax, and therefore earnings per share. Second, the prior-year and prior-quarter EPS comparatives are the company's adjusted figures (NT$27.80 for 3Q25, against NT$82.92 as originally reported; NT$26.97 for 2Q26, against NT$80.43 as originally reported).
The quarter against expectations
| Metric | 3Q26 actual | Consensus | Beat/Miss | Magnitude |
|---|---|---|---|---|
| Revenue | NT$434.6bn (~$13.2bn) | NT$400.4–402.6bn | Pre-disclosed | Public from monthly filings by Oct 8; consensus stale |
| Gross margin | 8.87% | No reliable consensus | n/a | (39)bp QoQ; +5bp YoY |
| Operating income | NT$32.5bn (~$0.98bn) | No reliable consensus | n/a | Margin 7.47%; +65.8% YoY |
| Net income | NT$23.4bn (~$0.71bn) | No reliable consensus | n/a | +51.6% YoY |
| EPS, diluted (NT$) | 38.99 (~$1.18) | 34.83–35.22 | Beat | +10.7% to +11.9% |
| EPS, basic (NT$) | 41.46 (~$1.26) | n/a (consensus is set on diluted) | n/a | +49.1% vs adjusted NT$27.80 |
| Non-operating line | (NT$2.0bn) | n/a | Drag | NT$2.77 per share after tax |
On EPS the size of the beat depends on which consensus one uses. The two aggregator figures that agree, NT$34.83 and NT$35.22, are scored against diluted EPS, and against them NT$38.99 beat by about 11%. A third aggregator carried NT$39.7, which would make the same figure a 2% miss, and a late-October press report cited an unnamed institutional estimate near NT$45.5. We score the beat on the two agreeing sources and do not lean on it. The cleaner evidence is the direction of estimates after the print: the published survey median for 2026 EPS rose from about NT$125 before the results to about NT$136 by November 12, and 2026 consensus now sits at NT$133–136 across three sources. No two independent sources gave a consensus for gross margin, operating income or net income, so we show none rather than borrow one.
Quarter-over-quarter and year-over-year (NT$bn, Taiwan-IFRS)
| Line | 3Q26 | 2Q26 | QoQ | 3Q25 | YoY |
|---|---|---|---|---|---|
| Net revenue | 434.6 | 278.2 | +56.3% | 266.8 | +62.9% |
| Gross profit | 38.6 | 25.8 | +49.6% | 23.5 | +63.8% |
| Gross margin | 8.87% | 9.26% | (39)bp | 8.82% | +5bp |
| Operating expenses | 6.1 | 5.5 | +10.1% | 4.0 | +54.1% |
| Operating expenses / revenue | 1.40% | 1.99% | (59)bp | 1.48% | (8)bp |
| Operating income | 32.5 | 20.2 | +60.5% | 19.6 | +65.8% |
| Operating margin | 7.47% | 7.27% | +20bp | 7.34% | +13bp |
| Non-operating income (loss) | (2.0) | (1.2) | n/m | 0.8 | n/m |
| Income before tax | 30.4 | 19.0 | +60.3% | 20.4 | +49.5% |
| Effective tax rate | 23.2% | 21.1% | +2.1ppt | 24.3% | (1.1)ppt |
| Net income | 23.4 | 15.0 | +56.1% | 15.4 | +51.6% |
| Net margin | 5.38% | 5.38% | (1)bp | 5.78% | (40)bp |
| EPS, basic (NT$, adjusted) | 41.46 | 26.97 | +53.7% | 27.80 | +49.1% |
| EPS, diluted (NT$, adjusted) | 38.99 | n/a | n/a | 26.82 | +45.4% |
| Weighted average shares, basic (mn) | 563.6 | n/a | n/a | 554.3 | +1.7% |
In US dollars the quarter is revenue of about $13.2bn, operating income of about $0.98bn and net income of about $0.71bn. Nine-month revenue of NT$989.3bn (~$30.0bn, +50.3%) and net income of NT$52.4bn (+40.5%) already exceed full-year 2025 net income of NT$51.1bn, which the board summary notes.
Revenue: known, but the shape matters
Revenue of NT$434.618bn (~$13.2bn) rose 62.9% year over year and 56.3% quarter over quarter, a single-quarter record by a wide margin; the previous record was 4Q25's NT$292.4bn. The monthly path inside the quarter was steep: July NT$117.7bn, August NT$144.3bn, September NT$172.6bn, so September alone was 47% above July and 100% above September 2025. The second quarter had been flat on the first (NT$278.2bn against NT$276.5bn), which is why the sequential jump looks the way it does. Wiwynn's product disclosure is one line, "Hyperscale data center," which carried 100% of revenue, and the company reports a single segment. What the print adds to the monthly data is geography, covered under Segment Performance below.
Margins: gross down, operating up
Gross margin of 8.87% fell 39bp from 2Q26's 9.26%, the highest of 2026's three quarters (7.55% in the first), and was 5bp above the year-ago 8.82%. The year-over-year comparison is not like for like: management has said that since April some customers' memory has been bought on their behalf and kept out of revenue, which lifts the percentage margin (Key Topics, item 3). The sequential slip, on the same footing, is what a rack-scale ramp looks like: as more of each shipment is a full rack with high-value components inside, the gross margin percentage compresses even when gross profit per rack rises. Incremental gross margin on the NT$156.5bn of sequential revenue growth was 8.2%. Operating expenses then did the work. They rose only 10.1% sequentially to NT$6.10bn, so the ratio to revenue fell from 1.99% to 1.40%. Research and development was NT$4.12bn, up 45.9% on the year and 1.0% on the quarter; the second quarter's spending step-up was absorbed by a revenue base 56% larger. The operating margin of 7.47% is therefore a real operating-leverage result, and a narrow one: it rests on an OpEx ratio that cannot fall much further from 1.4%.
EPS and below the line
Basic EPS of NT$41.46 (~$1.26) rose 49.1% from the adjusted NT$27.80; diluted EPS of NT$38.99 rose 45.4% from NT$26.82. The diluted count adds 35.6 million potential shares from convertible bonds, a 6.3% gap between basic and diluted EPS, and the first visible per-share cost of how this ramp is being financed. The tax rate of 23.2% was 2.1 points above the second quarter and 1.1 points below a year ago; it is not a source of quality in either direction. Wistron, the parent, owned 33.71% of the shares at September 30, down from 35.45% at year-end as bond conversions added shares.
Nine months (NT$bn)
| Line | 9M26 | 9M25 | Change |
|---|---|---|---|
| Net revenue | 989.3 | 658.2 | +50.3% |
| Gross margin | 8.61% | 8.71% | (10)bp |
| Operating income | 70.1 | 47.5 | +47.8% |
| Operating margin | 7.09% | 7.21% | (12)bp |
| Non-operating income (loss) | (2.8) | 1.1 | n/m |
| Net income | 52.4 | 37.3 | +40.5% |
| EPS, basic (NT$, adjusted) | 94.04 | 67.34 | +39.6% |
Segment Performance
Wiwynn reports one segment and one product line, and its quarterly report does not name customers. The company discloses customer concentration only in the annual report. We do not estimate a customer or product split. What the quarterly report does give is revenue by destination, and the monthly filings give the run-rate. We use both.
"The Group’s core profession is to provide the products and services in data center, and there is no significant segment division."
— Wiwynn, 3Q26 consolidated financial report, note 14
Revenue by destination (NT$bn)
| Region | 3Q26 | Share | QoQ | YoY | 2Q26 | 3Q25 |
|---|---|---|---|---|---|---|
| America | 362.7 | 83.5% | +59.4% | +64.9% | 227.6 | 220.0 |
| Europe | 40.7 | 9.4% | +80.5% | +164.7% | 22.5 | 15.4 |
| Asia | 29.1 | 6.7% | +22.3% | +104.5% | 23.8 | 14.2 |
| Other | 2.1 | 0.5% | (49.6)% | (87.6)% | 4.2 | 17.2 |
| Total | 434.6 | 100.0% | +56.3% | +62.9% | 278.2 | 266.8 |
America: the engine, and the concentration
Shipments to the Americas were NT$362.7bn, 83.5% of revenue, up from 81.8% in the second quarter and 82.1% for 2025 as a whole. Wiwynn's customers are a short list of US cloud platforms, and the 2025 annual report sizes the dependence directly: three customer groups took 48.58%, 29.46% and 18.27% of 2025 revenue, 96.3% combined. The destination data says the third quarter did not dilute that concentration; it intensified it.
"In 2025, the main sales were to three group customers with a ratio of 48.58%, 29.46%% and 18.27% respectively, so there was a risk of consolidation of sales."
— Wiwynn, 2025 Annual Report
Assessment: The American line is the thesis and the risk in one number. Revenue up 64.9% with three customers carrying almost all of it means the stock trades on those customers' data-center budgets and on Wiwynn's share of their rack builds. We treat concentration as permanent, not as something diversification will fix inside our horizon.
Europe and Asia: the fastest growth, off small bases
Europe more than doubled year over year to NT$40.7bn (+164.7%) and rose 80.5% on the quarter, the fastest line in the table; Asia doubled to NT$29.1bn. Together they were 16.1% of revenue against 11.1% a year ago. "Other" collapsed from NT$17.2bn to NT$2.1bn. Destination is where racks ship, not who buys them, so a European data-center build for a US customer reads as Europe. The board summary does not explain the regional mix.
Assessment: Read this as the same customers building in more places, not as new customers, until the 2026 annual report says otherwise. It is still useful: it shows the ramp is not a single-site program that can stop in one decision.
The monthly run-rate
| Month | 2026 (NT$bn) | 2025 (NT$bn) | YoY | MoM |
|---|---|---|---|---|
| April | 82.7 | 63.8 | +29.7% | (16.1)% |
| May | 84.1 | 71.1 | +18.2% | +1.6% |
| June | 111.4 | 85.8 | +29.8% | +32.5% |
| July | 117.7 | 84.5 | +39.2% | +5.7% |
| August | 144.3 | 96.0 | +50.4% | +22.6% |
| September | 172.6 | 86.3 | +100.0% | +19.6% |
| October | 165.1 | 91.3 | +80.9% | (4.3)% |
October revenue of NT$165.1bn, published after the close on November 6, about two and a half hours before the results, fell 4.3% from September's record but was still 14.0% above the third quarter's monthly average of NT$144.9bn and 80.9% above October 2025. Three months at October's level would be a fourth quarter of about NT$495bn, 14% above the third. Ten months of 2026 revenue, NT$1,154.4bn, already exceed full-year 2025 by 21.4%.
Assessment: The run-rate entering the fourth quarter is higher than the third quarter's average, which is the single most important fact for near-term earnings. The board's warning that material shortages "may cause short-term monthly revenue fluctuations" is already visible in October's dip, and we expect at least one more down month in the quarter. A dip caused by supply is a timing issue; a dip caused by a customer pausing would be a thesis issue. Nothing in the October print distinguishes the two yet.
Working capital and funding (NT$bn)
| Balance | Sep 30, 2026 | Jun 30, 2026 | Dec 31, 2025 | Sep 30, 2025 |
|---|---|---|---|---|
| Inventories | 308.4 | 207.4 | 136.3 | 156.4 |
| Raw materials | 169.4 | 120.1 | 52.9 | 78.0 |
| Finished goods | 125.3 | 77.9 | 77.8 | 74.0 |
| Accounts receivable | 114.4 | 84.3 | 76.9 | 74.0 |
| Other receivables (materials bought for others) | 110.3 | 130.9 | 0.0 | 0.0 |
| Accounts payable | 241.4 | 156.2 | 79.4 | 105.8 |
| Cash | 102.6 | 85.3 | 89.9 | 52.5 |
| Short-term bank borrowings | 124.3 | 76.6 | 68.5 | 36.0 |
| Bonds, mostly convertible | 45.4 | 75.4 | 20.8 | 23.2 |
| Long-term bank borrowings | 9.6 | 9.6 | 10.9 | 10.6 |
| Net debt | 76.6 | 76.3 | 10.2 | 17.4 |
| Total equity | 188.0 | 137.3 | 124.7 | 108.0 |
| Net debt / equity | 41% | 56% | 8% | 16% |
| Inventories / total assets | 45% | 37% | 40% | 50% |
Inventory rose NT$101.0bn in the quarter to NT$308.4bn (~$9.3bn), 45% of total assets, with raw materials up NT$49.3bn and finished goods up NT$47.5bn. On the quarter's cost of sales that is about 71 days of inventory, against about 59 days a year ago. Payables rose NT$85.2bn and carried most of it. Operating cash flow was negative NT$56.8bn for nine months, though the third quarter on its own was positive by about NT$4.7bn (nine months less the first half). Net debt was flat on the quarter at NT$76.6bn because NT$31.2bn of convertible bonds turned into equity during the nine months; the debt underneath shifted from bonds to short-term bank loans, which is why bank-loan interest rose 67% on the second quarter, to NT$2.60bn.
Assessment: This is the part of the print the market should be pricing, and the part a revenue-led reading misses. Wiwynn is carrying its customers' build-out on its own balance sheet: in inventory, in NT$110.3bn of materials it bought on others' behalf, and in US$4.2bn of receivables sold to banks without recourse. It is profitable work and, at 7.5% operating margin, well-paid work. It is also capital-hungry in a way the ODM model was not three years ago, and the November 6 equity decision is the proof.
Key Topics & Management Commentary
Overall Management Tone: With no call, the tone is whatever the board chose to put in one paragraph, and it chose caution on the near term and confidence on the long term. The summary leads with records, then attaches two conditions to the near term: tight raw-material supply and the policy implications of the US midterm elections. The same board meeting approved new equity, which says more about management's view of the next two quarters' cash needs than any adjective in the summary.
1. No call, one paragraph, and an equity raise at the same meeting
Wiwynn's board approved the quarter at a meeting whose filings reached the exchange between 17:41 and 17:56 Taiwan time on Friday, November 6. The results filing came first; within fifteen minutes it was followed by capital injections into the US and Malaysian subsidiaries, a resolution to issue up to 18 million new shares for a global depositary receipt offering, and the record date for a further 23.8 million shares from overseas convertible-bond conversions. There was no call to put any of it in context. The company's whole written commentary is one paragraph on its investor page, and the quarterly deck that followed is figures only.
"The Board also approved the issuance of up to 18 million common shares to participate in a GDR offering to replenish working capital, plus capital increases of US$250M in its U.S. subsidiary Wiwynn Technology Corporation and US$200M in its Malaysian subsidiary Wiwynn Technology Service Malaysia SDN. BHD. to support business operations."
— Wiwynn, 3Q26 board summary (Investor Updates)
Wiwynn does not usually hold calls; its register shows one earnings call since the start of 2024, for 4Q25 in February. The absence is not a signal on its own. The combination is: a record quarter, a supply warning and a new share issue, with the explanation left to a broker conference on November 17.
Assessment: The board told investors two things at once: earnings are at a record, and the business needs more equity to keep growing at this rate. Both are true. The second is the one we weight more heavily for the next two quarters, because it sets the share count every earnings estimate is divided by.
2. Revenue was public in October; the print-day news was the operating margin
Monthly filings had already told the market that the quarter's revenue would be about NT$434.6bn. What the market did not know was whether Wiwynn could keep its margin as the revenue base jumped 56% in one quarter. It could. Operating income of NT$32.45bn was a 7.47% margin, up 20bp on the second quarter and above both earlier 2026 quarters (6.31% and 7.27%) and the year-ago 7.34%. Part of the percentage gain since the first quarter is presentation rather than economics (see the next topic), but the sequential comparison is like for like and the absolute number is not affected: operating income rose NT$12.2bn in one quarter.
"Wiwynn’s Board approved 3Q26 financials on Nov 6: consolidated revenue NT$434.618B (+62.9% YoY), PAT NT$23.364B (+51.6% YoY), gross margin 8.9%, operating margin 7.5%, PAT margin 5.4%, EPS NT$41.46 (vs NT$27.80 in 3Q25, retroactively adjusted for the ex-rights issuance), a single-quarter record high for both revenue and profit"
— Wiwynn, 3Q26 board summary (Investor Updates)
Assessment: This is the evidence that matters for an ODM: volume grew much faster than overhead, and pricing did not have to give to win it. A 7.5% operating margin on NT$434.6bn of quarterly revenue is a scale few contract manufacturers reach. We take it as confirmation of the operating thesis, not as a beat to be traded.
3. Gross margin gave back 39bp; operating expenses gave back more
Gross margin fell from 9.26% to 8.87%. The board does not explain the change, and with no call there was nobody to ask. Two forces move this ratio at once. The first is presentation. Management said in August that from April some customers' memory moved to a buy-on-behalf arrangement, and in September that the arrangement is used mainly for general-purpose server projects. Memory bought that way stays out of both revenue and cost of sales, which raises the percentage margin without changing the profit, and it is the likeliest reason the second quarter's gross margin jumped to 9.26% from 7.55% in the first. The second force is mix: sell-side commentary attributes September's acceleration to ASIC-based AI servers, which carry more high-value content per rack. Operating expenses then rose only NT$0.56bn on NT$156.5bn of extra revenue. That lowered the expense ratio by 59bp, more than the gross margin lost, and lifted the operating margin.
"Guided by a prudent and pragmatic approach, Wiwynn will continue to invest in and advance computing, thermal and power solutions, and related technologies and products."
— Wiwynn, 3Q26 results release (November 9)
Assessment: Year-over-year margin percentages now flatter the business. On the pre-April presentation, the third quarter's gross and operating margins would have been lower than reported, so we do not read the 5bp and 13bp year-over-year gains as improvement. The sequential comparison is like for like, and it says operating leverage, not pricing, carried the quarter. With operating expenses at 1.4% of revenue there is little leverage left, so from here the operating margin follows the gross margin. Sell-side notes after the print put the fourth quarter's gross margin near 8.5%; that is the number to watch.
4. The non-operating line is now a financing cost
Taiwan hardware companies' non-operating lines often move with the currency: foreign-exchange gains and losses on dollar balances. Wiwynn's did not this quarter. The NT$2.03bn loss is NT$3.05bn of finance costs, of which NT$2.60bn is interest on bank loans, partly offset by NT$0.34bn of interest income, NT$0.29bn of fair-value gains on the convertible-bond derivatives, NT$0.23bn of other gains and only NT$0.15bn of currency gains.
| Non-operating line (NT$m) | 3Q26 | 2Q26 | 3Q25 |
|---|---|---|---|
| Interest income | 342 | 521 | 319 |
| Foreign-exchange gains (losses), net | 153 | (575) | 1,171 |
| Fair-value gains on financial instruments | 290 | 808 | 1 |
| Other gains (losses) | 225 | 58 | (37) |
| Interest expense, bank loans | (2,597) | (1,554) | (531) |
| Interest expense, other | (448) | (503) | (133) |
| Share of associates | 2 | 0 | (11) |
| Total non-operating income (loss) | (2,034) | (1,244) | 779 |
Short-term bank borrowings were NT$124.3bn at September 30, all US-dollar loans at 4.25% to 5.6%, against NT$76.6bn three months earlier and NT$36.0bn a year earlier. Bank interest of NT$2.60bn is 4.9 times last year's quarter.
"Moreover, the Group expects to comply with the quarterly covenants for at least 12 months after the reporting date."
— Wiwynn, 3Q26 consolidated financial report, note 6(l)
Assessment: At 4.5x to 5x last year's level, interest is no longer a rounding item: it took NT$2.77 from this quarter's EPS. We model it as a recurring cost that scales with the inventory the business carries, not as a one-off. The GDR proceeds will reduce it at the margin; growth will add it back.
5. NT$308bn of inventory and NT$110bn of materials bought for others
The balance sheet more than doubled in a year, from NT$311.6bn to NT$689.6bn of total assets. Two lines explain most of it. Inventory reached NT$308.4bn, of which NT$169.4bn is raw materials. And a line that did not exist a year ago, other receivables, stood at NT$110.3bn (it was NT$130.9bn at June 30). The report gives one sentence on what it is.
"The Group purchases raw materials on behalf of other parties and related parties, and provides human outsourcing services."
— Wiwynn, 3Q26 consolidated financial report, note 6(e)
Management has given the rest of the explanation outside the filings: since April, memory for some customers has been bought on a buy-on-behalf basis, mainly for general-purpose servers. On that reading the receivable is owed by customers for components Wiwynn paid for in advance, and the components never pass through revenue. Receivables are also understated by factoring. At September 30 Wiwynn had sold US$4.2bn of trade receivables to banks without recourse, against US$1.8bn at year-end and less than US$0.1bn a year ago.
"Under the agreements, the Group does not have the responsibility to assume the default risk of the transferred trade receivables but is liable for the losses incurred on any business dispute."
— Wiwynn, 3Q26 consolidated financial report, note 6(d)
Assessment: In plain terms, Wiwynn is financing a large share of its customers' rack builds before the racks ship: buying the components, carrying them as raw materials or as receivables from the parties it bought them for, and selling the finished receivables to banks. That is how a white-box supplier wins rack-scale programs, and with hyperscale customers the credit risk is low. The risk is inventory and timing. If a program slips, Wiwynn holds NT$169.4bn of components and the interest on the loans that bought them.
6. The funding stack, and what it does to the share count
Wiwynn has used every funding channel available to a Taiwan issuer in 2026: a US$2bn zero-coupon overseas convertible bond issued in April, conversions of its overseas bonds into equity, a US$1.52bn three-year syndicated bank loan signed on October 14, a domestic zero-coupon convertible bond of up to NT$15bn now being placed (collection banks named November 11), short-term dollar bank loans, and now a GDR.
| Share-count step | Shares (mn) | Cumulative vs. Sep 30 |
|---|---|---|
| Outstanding at September 30, 2026 | 584.3 | n/a |
| Overseas convertible-bond conversion, record date November 6 | +23.8 | +4.1% |
| GDR offering, up to | +18.0 | +7.2% |
| Overseas convertibles still outstanding (our estimate: 3Q26 diluted count less the November conversion) | about +11.8 | about +9.2% |
| Domestic convertible, up to NT$15bn (our estimate at prices near NT$1,600; conversion price not yet set) | about +9.5 | about +10.8% |
"Except for early redemption, repurchases and cancellations, exercise of conversion rights by the bondholders and the cessation of conversion period, from the day following the three months after the issuance of the bonds to (1) ten days before the maturity date or (2) the fifth business day prior to the repurchase date if the bondholders exercise their put option or the date of early redemption of the bonds (excluding the maturity date), the bondholders may request the issuing company to convert the bonds into shares of common stocks in accordance with the provisions of the relevant laws and the Trust Deed."
— Wiwynn, 3Q26 consolidated financial report, note 6(m)
The adjusted conversion prices are NT$1,017.75 for the 2024 bond and NT$1,396.46 for the 2026 bond, both below the November 12 close of NT$1,570.00, so the remaining overseas bonds are in the money and conversion is the likely path.
Assessment: Committed dilution is about 7% from the September 30 count, and about 11% if every instrument converts. Against earnings growing 50%, that is affordable. It is not free: it explains much of the gap between operating income growth and per-share growth in our estimates, and it is why we value the company on diluted EPS throughout.
7. "Tight raw material supply" and October's dip
The board's caution is specific in effect and vague in content. It does not name the material. In September management said CPUs were the tightest item but not in severe shortage; industry data since then point to server memory staying short through the fourth quarter. October revenue, published after the close on the afternoon of the board meeting, fell 4.3% from September to NT$165.1bn, the first monthly decline since April, and sell-side notes tied the dip to component shortages in general-purpose servers rather than to demand. The raw-materials line on the balance sheet rose NT$49.3bn in the quarter, which is consistent with Wiwynn buying ahead wherever it can. The English release posted on November 9 words the caution more firmly than the investor-page summary: supply "will cause" monthly fluctuations, not "may".
"Although tight raw material supply will cause fluctuations in monthly revenue in the short term, Wiwynn remains optimistic about the long-term growth of the AI and data center markets."
— Wiwynn, 3Q26 results release (November 9)
Assessment: A supply-limited quarter is a better problem than a demand-limited one, and October's level still implies a fourth quarter about 14% above the third. The risk is that "monthly fluctuations" becomes the explanation for a soft quarter that has other causes. The November and December revenue filings (due by December 10 and January 10) will tell us which it is before the fourth-quarter print does.
8. US policy after the midterms, and the US plant
The summary's commitment to "monitor the policy implications of the U.S. midterm election results" came three days after the November 3 vote. The Taiwan market fell 3.2% on November 5 and 4.5% on November 6 on concerns that the election outcome would bring tighter US regulation of AI data centers, and Wiwynn fell 7.6% and 5.8% on those two days, before either the October revenue or the results were out. Democrats won both chambers of Congress and several governorships, Texas among them, and that matters to Wiwynn more than to most. Its first US plant is in Socorro, Texas: a US$1.6bn investment that opened in September and, in the company's own announcement, builds server systems and racks for Amazon. A second building is due to start production in the second quarter of 2027, and management targets about a fifth of group capacity in the US next year, against roughly 70% in Mexico, 20% in Malaysia and 10% in Europe before Texas opened. Almost all of Wiwynn's revenue ships to the Americas, so slower US data-center permitting would reach it directly. The board's same-day decision to put US$250M more into the US subsidiary says the plan continues.
"Wiwynn supports Amazon’s data center infrastructure by manufacturing and integrating server systems and racks, helping to provide the hardware and capacity needed to support Amazon’s growing cloud infrastructure across the U.S."
— Wiwynn, Texas facility release (September 2026)
Assessment: We do not think a change in the US Congress changes the hyperscale build-out plans of Wiwynn's customers in the next twelve months; those budgets were set on their October earnings calls. We do think it raises the multiple investors demand for US data-center exposure, and that is a valuation issue, not an earnings issue.
9. Capacity outside Taiwan: US$450M into the US and Malaysian units
The board put US$250M into Wiwynn Technology Corporation, the US subsidiary, taking the parent's cumulative investment there to US$800M, and US$200M into the Malaysian unit, taking that to US$427.5M. Property, plant and equipment rose to NT$31.3bn from NT$20.8bn at year-end; contracted but unrecognized equipment purchases were NT$13.6bn at September 30 against NT$6.5bn a year earlier; depreciation was NT$1.15bn in the quarter, up 79%. In August the board set a second-half capital budget of US$942M.
"The Company participated in the cash capital increase of its subsidiary, Wiwynn Technology Corporation and Wiwynn Technology Service Malaysia SDN. BHD., at an amount of USD 250,000 thousand and USD 200,000 thousand respectively, based on a resolution approved during its board meeting held on November 6, 2026."
— Wiwynn, 3Q26 consolidated financial report, note 11
Assessment: Capacity spending is still small next to working capital: nine-month equipment spending and prepayments were about NT$17.1bn, against a NT$172.2bn rise in inventory over the same period. The plants are not what strains the balance sheet. The inventory that runs through them is.
10. Wistron: parent, supplier and a shrinking shareholder
Wistron is Wiwynn's controlling shareholder and, increasingly, its contract manufacturer. Purchases from Wistron were NT$64.0bn in the third quarter, 3.4 times a year earlier and equal to 16.2% of the quarter's cost of sales; in 2025 Wistron supplied 8.09% of total procurement. Sales to Wistron-group affiliates were NT$12.9bn, 3.0% of revenue, against NT$0.4bn a year ago. Wistron's stake fell to 33.71% from 35.45% as convertible-bond conversions added shares.
"Trading terms of purchase transactions with related parties can’t be compared with third-party vendors due to product specifications."
— Wiwynn, 3Q26 consolidated financial report, note 7
Assessment: Using the parent's factories lets Wiwynn add capacity without building it, which helps explain how revenue grew 63% while fixed assets stayed small. It also means a growing share of cost of sales is set between related parties on terms the report says cannot be benchmarked. We do not see evidence of value moving between the two companies, and we will watch the purchase share each quarter.
Guidance & Outlook
Wiwynn does not give numeric revenue, margin or earnings guidance, and it did not this quarter. We record the outlook exactly as the board wrote it and do not convert it into numbers.
| Item | What the company said (Nov 6) | Our reading |
|---|---|---|
| Demand | "Data center demand remains strong" | Consistent with October revenue +80.9% YoY |
| Supply | "tight raw material supply may cause short-term monthly revenue fluctuations" | A near-term caution; October already fell 4.3% MoM |
| Long term | "optimistic on long-term AI and data center growth" | Unchanged posture |
| Policy | "will monitor the policy implications of the U.S. midterm election results" | No specific exposure named |
| Investment | "will keep investing in computing, thermal and power solutions" | R&D +45.9% YoY; 2H26 capex budget US$942M |
| Equity | Up to 18 million new shares for a GDR "to replenish working capital" | About 3.1% of September 30 shares |
| Subsidiaries | US$250M into the US unit, US$200M into the Malaysian unit | Capacity and working capital outside Taiwan |
| Revenue, margin, EPS | Not guided | n/a |
"Data center demand remains strong, though tight raw material supply may cause short-term monthly revenue fluctuations; Wiwynn stays optimistic on long-term AI and data center growth, will monitor the policy implications of the U.S. midterm election results, and will keep investing in computing, thermal and power solutions."
— Wiwynn, 3Q26 board summary (Investor Updates)
Implied fourth quarter: October at NT$165.1bn puts the fourth quarter on a NT$495bn pace. We use that pace as our working estimate, 14% above the third quarter. The supply warning is the main risk to it.
Street at: Two aggregators carry fourth-quarter consensus of NT$465–474bn of revenue and NT$39.3–41.2 of EPS, figures that look stale against October's run-rate; sell-side notes published after the print moved to roughly 20% sequential growth. Full-year 2026 EPS consensus is NT$133–136. Our NT$495bn fourth quarter sits above those aggregator figures and below the roughly 20% growth in the post-print notes. Our NT$136 full year sits at the top of the consensus band.
Guidance style: Qualitative and brief, as is the company's practice. This quarter the board attached two named conditions to the near term, supply and US policy, and paired them with an equity raise.
What They're NOT Saying
- Which material is short, and for how long. "Tight raw material supply" is the most consequential phrase in the summary and it is not specified. Management named CPUs in September; industry data now point to memory. Processors, memory, power parts and substrates have very different lead times, and the answer decides whether October's dip is a month or a quarter.
- What the NT$110bn of "other receivables" is, and who carries the risk. The report says only that the group "purchases raw materials on behalf of other parties and related parties." The line did not exist a year ago. Management has described a buy-on-behalf arrangement for memory at investor forums, but the filing names no counterparties, no payment terms, and nothing on who holds price or obsolescence risk on those materials.
- Why equity, and why now. The GDR resolution names the use of proceeds as foreign-currency material purchases. It does not say why a company with NT$102.6bn of cash, a fresh US$1.52bn syndicated loan and a NT$15bn domestic convertible in progress also needs new shares, or why at a share price 40% below the September high.
- What the midterm "policy implications" are. The summary names the risk and gives no exposure: not tariffs, not data-center permitting, not the US plant's role in either.
- Customer and program mix. Nothing on rack versus board mix, on accelerator platforms, or on which customer drove the Europe line. The only customer data is in the annual report, which next arrives in spring 2027.
- Why no call. A record quarter, a new equity raise and a new supply warning would normally be the reasons to hold a call. The company will present the same deck at a broker conference eleven days after the board meeting.
Market Reaction
All levels are TWSE closing prices in New Taiwan dollars; the benchmark is the TAIEX. Results reached the exchange at 17:41 Taiwan time on Friday, November 6, after the 13:30 close and about two and a half hours after the October revenue release, so Monday, November 9 was the first session to price both. Prices before September 2 are adjusted for the stock dividend.
- Pre-print setup: NT$1,555.00 (~$47.13) at the November 6 close, up 3.5% year to date against a 37.1% gain for the TAIEX, down 30.6% over the prior 30 days and up 7.4% over twelve months. The 52-week closing range was NT$1,105.90 to NT$2,613.94, and the stock sat 40.5% below the September 1 high. It had fallen 9.9% on October 15 (TAIEX down 3.4%) after a hot US inflation report and an overnight slide in semiconductor shares, and 7.6% and 5.8% on November 5 and 6 (TAIEX down 3.2% and 4.5%) as investors priced the risk of tighter US data-center regulation after the midterm elections.
- Reaction session (November 9): opened at NT$1,570.00 (+1.0%), traded between NT$1,570.00 and NT$1,700.00 (+1.0% to +9.3%), and closed at NT$1,645.00 (~$49.86), up 5.8% or NT$90.00. Volume was 4.3 million shares, in line with the 30-day average. The TAIEX rose 4.9% that day, its second-largest point gain on record, after the government authorized its market-stabilization fund to intervene and US technology shares rebounded. Hon Hai rose 6.5%, Wistron 5.3% and Quanta 2.1%.
- Since: NT$1,680.00 on November 10 (+2.1%), NT$1,680.00 on November 11 (unchanged), and NT$1,570.00 on November 12 (-6.5%, against -2.6% for the TAIEX in a broad sell-off in AI-related shares). From the pre-print close, the stock is up 1.0% while the TAIEX is up 8.5%.
A beta session, not a verdict
Wiwynn beat the index by 0.9 points on the reaction day, gave back an intraday gain of 9.3%, and traded average volume. Every large Taiwan hardware name rose that morning. A record quarter, a 4.3% monthly revenue dip and an equity raise all arrived on the same evening, and the market netted them to roughly zero relative to the index.
The setup was macro, and the estimates went the other way
The 40% fall from the September high was not built on Wiwynn news. The largest down days were index days: a US inflation shock in mid-October and the US election result in early November. Over the same weeks, the published survey median for 2026 EPS rose from about NT$122 in mid-October to about NT$136 after the print. Price and estimates moving in opposite directions is the core of our initiation case.
Since the print, the stock has trailed by 7.5 points
From November 6 to November 12 the TAIEX recovered 8.5% and Wiwynn 1.0%. No report we found ties a move in those days to company news. Our reading is supply: the market now knows that up to 18 million new shares, a domestic convertible of up to NT$15bn and 23.8 million conversion shares are coming, and the domestic convertible's collection banks were named on November 11, the day before the 6.5% drop. When Wistron priced its own GDR in September at a 5.5% discount, its shares fell 5.3% on the day. We expect the overhang to last until Wiwynn's GDR is priced.
Street Perspective
No broker changed its rating after the print. One cut its target on a valuation roll and margin concerns, two reiterated, and estimates moved up: one large house raised its 2027 EPS forecast by 29%. The debate is about margins and money, not about demand.
Debate: is the margin peak behind it?
Bull view: A 7.47% operating margin at NT$434.6bn of quarterly revenue proves the model scales. ASIC-based AI servers for the largest customer carry better economics than standard GPU racks, and operating expenses keep falling as a share of revenue.
Bear view: Sell-side notes put fourth-quarter gross margin at about 8.5% as GPU racks and higher component costs enter the mix, and at least one model has gross margin falling to about 6.7% in 2027. With operating expenses already at 1.4% of revenue, every point of gross margin lost is a point of operating margin lost.
Our take: The bears are right about the direction of the percentage and less right about what it means. Rack-scale GPU shipments lower the margin percentage and raise the profit per rack. We model a 6.6% operating margin for 2027, below this quarter and below most of the Street, and still get operating income up by a third.
Debate: is the balance sheet a moat or a liability?
Bull view: Financing customers' builds (buying the components, carrying the inventory, selling the receivables) is how a white-box supplier wins rack-scale programs. The counterparties are the largest cloud platforms in the world, the receivables are sold without recourse, and the extra equity funds growth, not losses.
Bear view: Inventory is 45% of assets, nine-month operating cash flow is negative NT$56.8bn, bank interest has risen almost fivefold, and the company is issuing equity at a price 40% below its September high. If one program slips, Wiwynn holds the components and the loans.
Our take: It is a moat for as long as programs ship on schedule, and the third quarter's positive operating cash flow of about NT$4.7bn suggests the cycle turns cash-positive once racks ship. The GDR tells us the company is close to the limit of what it wants to fund with debt. Dilution of about 7% committed, and about 11% at most, is affordable against earnings growing faster than 50%; it belongs in the per-share estimates, not in a discount on top of them.
Debate: was the sell-off about Wiwynn?
Bull view: The drawdown came on macro and policy days while estimates rose. The largest US cloud platforms used their October results to hold or raise 2026 capital spending, and one said 2027 spending would increase significantly.
Bear view: US policy is a direct risk for a company that ships 83.5% of revenue to the Americas and whose first US plant is in Texas, where data-center opponents gained ground in the election. The multiple the market pays for US AI hardware has fallen and may keep falling.
Our take: The earnings risk from US politics is small inside twelve months; the multiple risk is not. That is why we value the shares at 12–13x our 2027 earnings rather than the 16x behind some of the higher targets on the Street.
Model Update & Valuation Framework
This is our first report on Wiwynn and we have no prior model to revise, so we set out the working estimates the rating rests on. They are deliberately below the Street on 2027 margins and fully diluted for every instrument now outstanding or approved.
| Item (NT$bn unless stated) | 4Q26E | FY26E | FY27E | Basis |
|---|---|---|---|---|
| Revenue | 495 | 1,484 | 2,150 | 4Q at October's pace; 2027 +45% |
| Gross margin | 8.5% | 8.6% | 7.8% | GPU racks and component costs |
| Operating margin | 7.2% | 7.1% | 6.6% | OpEx about 1.2–1.3% of revenue |
| Non-operating income (loss) | (2.8) | (5.6) | (9.0) | Interest scales with inventory |
| Tax rate | 23% | 22% | 23% | 3Q26 rate |
| Net income | 25.2 | 77.6 | 102.3 | FY26 +52%; FY27 +32% |
| EPS (NT$) | about 42 basic | about 136 basic | about 158 diluted | FY27 on about 647mn shares |
| Consensus EPS (NT$) | 39.3–41.2 | 133–136 | n/a | Two-source rule; 2027 not confirmed |
The fourth quarter assumes three months at October's NT$165.1bn, a gross margin of 8.5%, in line with where sell-side notes put it after the print, and roughly NT$2.8bn of net interest. Our 2026 estimate lands inside the NT$133–136 consensus range. For 2027 we assume revenue growth of 45%, below the 60% at least one broker models, and a gross margin of 7.8%, between this year's 8.6% and the 6.7% in the most bearish published model. Published 2027 EPS figures that we have seen sit at NT$200 or above; ours is about NT$158 because of lower margins, higher interest and a share count of about 647 million that assumes the GDR is sold in full and every convertible converts.
Valuation. At NT$1,570.00, the November 12 close, and about 608 million shares after the November 6 conversion, the market value is about NT$955bn (~$28.9bn), and net debt was NT$76.6bn at September 30. The shares trade at about 11.5x our 2026 EPS, 14.4x trailing net income and 9.9x our 2027 diluted EPS.
| Scenario | 12-month value (NT$) | vs. NT$1,570 | Basis | What has to happen |
|---|---|---|---|---|
| Bull | 2,600 (~$79) | +66% | 13x a NT$200 2027 EPS | Street-level 2027 margins; ASIC and GPU racks both ramp; the GDR is the last raise |
| Base | 1,850–2,050 (~$56–62) | +18% to +31% | 11.7x–13x our NT$158 | Revenue +45% in 2027, gross margin drifts to 7.8%, full dilution |
| Bear | 1,000 (~$30) | -36% | 8.5x a NT$118 EPS | A customer pause leaves 2027 revenue flat on the 4Q26 run-rate; operating margin 5.5%; inventory write-downs |
Our base multiple of 12–13x is below the 16x that underpins some of the higher sell-side targets, and the discount is deliberate: three customers buy almost everything, the balance sheet carries the build, and the market's multiple for US AI hardware is under political pressure. Even so, the base case is a total return of roughly 20% to 30% over twelve months, ahead of what we expect from the S&P 500 and the TAIEX. The bear case is real and roughly a third down, and it requires a customer to pause, not just a slower month.
Bottom line: initiating at Outperform
Rating decision: We initiate coverage of Wiwynn (TWSE: 6669) at Outperform, with a 12-month value range of NT$1,850–2,050 (~$56–62). The third quarter showed what the business does well: it added NT$156.5bn of revenue in a quarter and NT$12.2bn of operating income with it, and its run-rate entering the fourth quarter is higher still. The shares have fallen 40% from their September high on days driven by US inflation, US politics and AI sentiment, while 2026 estimates rose about 11%. The costs of growth, interest and dilution, are visible and we have put them in the per-share numbers.
What would move us to Hold: a fourth-quarter gross margin below 8.0% without a matching rise in revenue; another large inventory build that does not convert to cash; a GDR that is upsized or priced at a deep discount; or November and December revenue back below the third quarter's NT$144.9bn monthly average.
What would move us to Underperform: evidence that one of the three main customers is pausing or moving share to another supplier; US state or federal action that delays the customers' data-center projects; or a write-down on the NT$169.4bn of raw materials.
Signposts for the fourth-quarter results (early 2027):
| Signpost | What to watch | Bullish if | Bearish if |
|---|---|---|---|
| Monthly revenue | November and December filings | NT$160bn a month or more | Below NT$145bn a month |
| Gross margin | 4Q26 print | 8.5% or better | Below 8.0% |
| Working capital | Inventory and other receivables | Flat or lower on higher revenue | Another NT$100bn build |
| Cash flow | 4Q26 operating cash flow | Positive for a second quarter | Negative again |
| Funding | GDR pricing and size | Up to 18M shares at a modest discount | Upsized, or a deep discount |
| Interest | Finance costs | NT$3bn a quarter or less | Rising faster than operating income |
| Concentration | 2026 annual report (spring 2027) | Top three below 96% of revenue | A larger single-customer share |
| Disclosure | Whether a 4Q26 earnings call is held | A call, as for 4Q25 | Another quarter with no call |
Initiating Coverage: Bull / Bear Matrix
As our first report on Wiwynn, we set out the framework we will track each quarter.
| Thesis point | Status at initiation | Verdict / what we are watching |
|---|---|---|
| Bull #1: Hyperscale AI demand keeps revenue compounding | Confirmed | 3Q26 revenue +62.9%; October +80.9% YoY; ten months already 21% above all of 2025 |
| Bull #2: Operating margin holds near 7% at rack scale | Confirmed | 7.47% in 3Q26, +20bp QoQ on a like-for-like basis; gross margin 8.87% is the line to watch |
| Bull #3: The price has fallen faster than estimates | Confirmed | Shares 40% below the September high; 2026 consensus EPS up from about NT$122 to about NT$136 since mid-October |
| Bear #1: Three customers buy almost everything | Active | 96.3% of 2025 revenue; Americas 83.5% of 3Q26; structural, not fixable within our horizon |
| Bear #2: Growth is funded by the balance sheet and new shares | Active | Inventory NT$308.4bn; nine-month operating cash flow -NT$56.8bn; about 7% committed dilution, about 11% at most |
| Bear #3: Gross margin compresses as GPU racks and component costs rise | Emerging | -39bp QoQ; Street sees about 8.5% in 4Q26; we model 7.8% for 2027 |
| Bear #4: US policy and AI sentiment set the multiple | Active | TAIEX -20% from October 6 to November 6; Texas plant in a state where data-center opponents gained ground |
Overall: Three bull points confirmed against two active structural risks (concentration and funding), one emerging margin risk and one valuation risk from US policy. The operating thesis is intact. The balance-sheet thesis is the one to monitor each quarter.
Action: Initiate at Outperform. We would use the GDR overhang as the entry point, and we would reassess if the fourth quarter brings another inventory build without positive operating cash flow.