A Write-Down Hid a Steady Margin and Suppliers Paid for the Doubling: Initiating Quanta at Outperform, 34% Below the High
Key Takeaways
- Revenue was already public; the margin was the news, and the headline disappointed. Monthly filings had put third-quarter revenue of NT$1,115.5bn (~$33.9bn, +125.2%) in view by early October. What the board added on November 11 was a gross margin of 4.75%, down 27bp on the quarter and 210bp on the year, and operating income down 6.0% sequentially to NT$32.2bn. Diluted EPS of NT$6.82 (~$0.21) missed the NT$7.12–7.22 consensus by 4–6%.
- A NT$4.9bn inventory write-down explains the whole sequential decline. Before it, gross margin was 5.19%, up 10bp, and operating income rose about 6%. Write-downs have reached NT$10.5bn in nine months, so we treat them as a recurring cost. Research and development up 32% and a share count 4.1% larger after the GDR then took basic EPS down 4.7% to NT$7.08.
- Suppliers, not the balance sheet, fund the build. Payables rose NT$234.0bn in the quarter, more than receivables and inventory together, so a business growing 125% generated about NT$22.1bn of operating cash and NT$11.5bn of free cash flow. The GDR then cut net debt to NT$100.7bn, 29% of equity.
- The call defended the margin and conceded a slower rhythm. Management guided fourth-quarter gross margin slightly lower and operating margin flat, called it a transition quarter between GPU generations with larger volumes from the first quarter of 2027, and expects AI-server growth in 2027 at a high double-digit rate, with servers reaching 85–90% of revenue. The script also conceded that tighter permitting will make cloud customers more careful with capital spending.
- Rating: Initiating at Outperform, with a 12-month value range of NT$300–330. At NT$276.50 the shares are 34% below their June high and trade at 10.7x trailing EPS and 9.3x our 2027 diluted EPS of about NT$29.8, with a 5.6% dividend yield. They entered the print up 9.6% for the year against 52.8% for the TAIEX and fell 6.5% on it. Thin margins and dependence on one GPU platform cadence are the real risks, and they are why our multiple is half the one the shares carried in June.
Results vs. Consensus
Two conventions govern this section. First, Quanta'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$1,115.545bn to the thousand. That total was public about a month before the board met on November 11. A revenue "beat" against a consensus set before the monthly filings is stale arithmetic, and we do not score it as one. What the board released was the cost structure: gross margin, operating expenses, the non-operating line, tax and therefore earnings per share. Second, Quanta pays its dividends in cash, so none of its prior-period per-share figures has been restated; the 3Q25 EPS of NT$4.26 is as originally reported. The comparison is still not like for like on shares, because the August GDR added 245 million new shares and lifted the quarter's weighted share count by 4.1%.
The quarter against expectations
| Metric | 3Q26 actual | Consensus | Beat/Miss | Magnitude |
|---|---|---|---|---|
| Revenue | NT$1,115.5bn (~$33.9bn) | NT$1,103–1,114bn | Pre-disclosed | Public from monthly filings by Oct 7; consensus stale |
| Gross margin | 4.75% | No reliable consensus | n/a | (27)bp QoQ; (210)bp YoY |
| Operating income | NT$32.2bn (~$0.98bn) | No reliable consensus | n/a | Margin 2.89%; (6.0)% QoQ |
| Net income to parent | NT$28.4bn (~$0.86bn) | No reliable consensus | n/a | +73.0% YoY; (0.8)% QoQ |
| EPS, diluted (NT$) | 6.82 (~$0.21) | 7.12–7.22 | Miss | (4.2)% to (5.5)% |
| EPS, basic (NT$) | 7.08 (~$0.22) | n/a (consensus is set on diluted) | n/a | +66.2% vs NT$4.26 |
| Inventory write-down in cost of sales | (NT$4.9bn) | n/a | Drag | 44bp of revenue; about NT$1.01 a share after tax |
On EPS the result is a miss on every consensus we can confirm. The two independent aggregator figures, NT$7.21 and NT$7.22, are scored against diluted EPS, and NT$6.82 fell 5.4% to 5.5% short of them; a third aggregator carried NT$7.12, a 4.2% miss. Against basic EPS of NT$7.08 the gap would be 1–2%, but the aggregators score on the diluted figure, and so do we. The operating miss is larger than the EPS miss: the one aggregator that publishes an operating-income forecast carried NT$36.1bn, and net income landed on that aggregator's NT$28.5bn only because non-operating income and a lower tax rate filled the gap. One database put consensus gross margin at 5.1%. Because no second source confirms the gross-margin, operating-income or net-income figures, we show none of them rather than borrow one. After the print, full-year 2026 consensus EPS sits at NT$25.9–26.6 across three sources.
Quarter-over-quarter and year-over-year (NT$bn, Taiwan-IFRS)
| Line | 3Q26 | 2Q26 | QoQ | 3Q25 | YoY |
|---|---|---|---|---|---|
| Net revenue | 1,115.5 | 1,036.6 | +7.6% | 495.3 | +125.2% |
| Gross profit | 53.0 | 52.1 | +1.8% | 33.9 | +56.3% |
| Gross margin | 4.75% | 5.02% | (27)bp | 6.85% | (210)bp |
| Operating expenses | 20.8 | 17.8 | +16.7% | 15.5 | +33.9% |
| of which research and development | 12.8 | 9.7 | +31.8% | 9.7 | +31.9% |
| Operating expenses / revenue | 1.87% | 1.72% | +15bp | 3.14% | (127)bp |
| Operating income | 32.2 | 34.2 | (6.0)% | 18.4 | +75.3% |
| Operating margin | 2.89% | 3.30% | (42)bp | 3.71% | (82)bp |
| Non-operating income | 2.7 | 2.3 | +15.7% | 2.5 | +5.8% |
| Income before tax | 34.9 | 36.6 | (4.6)% | 20.9 | +66.8% |
| Effective tax rate | 17.9% | 21.3% | (3.3)ppt | 20.5% | (2.5)ppt |
| Net income to parent | 28.4 | 28.7 | (0.8)% | 16.4 | +73.0% |
| Net margin | 2.55% | 2.76% | (22)bp | 3.32% | (77)bp |
| EPS, basic (NT$) | 7.08 | 7.43 | (4.7)% | 4.26 | +66.2% |
| EPS, diluted (NT$) | 6.82 | 7.08 | (3.7)% | 4.15 | +64.3% |
| Weighted average shares, basic (mn) | 4,011.6 | 3,854.5 | +4.1% | 3,854.5 | +4.1% |
In US dollars the quarter is revenue of about $33.9bn, operating income of about $0.98bn and net income of about $0.86bn. Nine-month net income of NT$78.3bn and basic EPS of NT$20.03 already exceed all of 2025 (NT$75.0bn and NT$19.45).
Five quarters (NT$bn)
| Line | 3Q25 | 4Q25 | 1Q26 | 2Q26 | 3Q26 |
|---|---|---|---|---|---|
| Net revenue | 495.3 | 638.6 | 809.2 | 1,036.6 | 1,115.5 |
| Gross margin, reported | 6.85% | 6.33% | 4.78% | 5.02% | 4.75% |
| Inventory write-down charged to cost of sales | 1.31 | 1.71 | 4.87 | 0.72 | 4.92 |
| Gross margin before the write-down | 7.11% | 6.59% | 5.38% | 5.09% | 5.19% |
| Operating income | 18.4 | 24.0 | 23.0 | 34.2 | 32.2 |
| Operating margin | 3.71% | 3.76% | 2.85% | 3.30% | 2.89% |
| Net income to parent | 16.4 | 22.2 | 21.2 | 28.7 | 28.4 |
| EPS, basic (NT$) | 4.26 | 5.75 | 5.50 | 7.43 | 7.08 |
4Q25 is the full year less the first nine months. The write-down line is the "loss of inventory valuation and obsolescence" that the company charges to cost of sales; it excludes a separate NT$2.2bn of scrapping and disposal losses in 3Q26 (NT$2.1bn in 2Q26).
Revenue: known, and lumpy inside the quarter
Revenue of NT$1,115.5bn (~$33.9bn) rose 125.2% year over year and 7.6% sequentially, a record by a wide margin. The monthly path was uneven: July NT$366.3bn, August a record NT$424.0bn (+177.4% year over year), and September NT$325.3bn, down 23.3% from August. The second quarter had grown 28.1% on the first, so the third quarter is a slowdown in sequential growth from a much larger base. Quanta's financial report has a single material segment, "electronics," which carried 99.9% of revenue; it does not split servers from notebooks. The product mix reaches investors only through what management says on its calls (Key Topics, item 1).
Margins: a write-down quarter, again
Inventory write-downs now move Quanta's gross margin by more than its mix does. They were NT$4.87bn in the first quarter, NT$0.72bn in the second and NT$4.92bn in the third, NT$10.5bn for nine months against NT$1.4bn a year earlier. Before the write-down, gross margin has drifted down from 7.11% a year ago to 5.38%, 5.09% and 5.19% this year: a 190bp structural step as full AI racks, with their high pass-through content, became most of revenue, then a plateau. The plateau is the encouraging part. The year-over-year comparison is not: incremental gross margin on the NT$620.3bn of year-over-year revenue growth was 3.1%, and on the NT$79.0bn of sequential growth only 1.2%. Operating expenses rose NT$3.0bn in the quarter, more than reported gross profit rose (NT$0.9bn).
EPS and below the line
Basic EPS of NT$7.08 (~$0.22) rose 66.2% from NT$4.26 and fell 4.7% from NT$7.43; diluted EPS of NT$6.82 rose 64.3% and fell 3.7%. The diluted count adds 179.1 million potential shares from Quanta's overseas convertible bonds and 23.0 million from employee compensation, a 3.7% gap between basic and diluted EPS. The non-operating line of NT$2.7bn is mostly currency: NT$3.9bn of foreign-exchange gains against net interest expense of NT$2.3bn, which rose from NT$1.7bn in the second quarter. The tax rate of 17.9% was below the second quarter's 21.3% and the year-ago 20.5% (the first quarter's was 13.2%); the report does not explain the swings. At the second quarter's 21.3%, net income would have been about NT$1.2bn lower.
Nine months (NT$bn)
| Line | 9M26 | 9M25 | Change |
|---|---|---|---|
| Net revenue | 2,961.4 | 1,485.1 | +99.4% |
| Gross margin | 4.86% | 7.27% | (241)bp |
| Operating income | 89.5 | 63.4 | +41.2% |
| Operating margin | 3.02% | 4.27% | (125)bp |
| Non-operating income | 6.5 | 4.4 | +47.4% |
| Net income to parent | 78.3 | 52.8 | +48.3% |
| EPS, basic (NT$) | 20.03 | 13.70 | +46.2% |
| EPS, diluted (NT$) | 19.12 | 13.35 | +43.2% |
Segment Performance
Quanta reports one material operating segment, electronics, which carried NT$1,114.7bn of the quarter's NT$1,115.5bn; everything else is NT$0.8bn. The financial report gives no product split and names no customers. The split between AI servers, general-purpose servers and notebooks exists only in what management says on the call, where it gives shares rather than figures (Guidance & Outlook). We do not estimate a product split. What the documents give is revenue by destination, the monthly run-rate, the annual customer-concentration note and the balance sheet. We use all four.
"Electronic products: 1,114,736,528; Other products: 807,993"
— Quanta, 3Q26 consolidated financial report, note 6(23), revenue by major product (NT$ thousands, our translation of the line labels)
Revenue by destination (NT$bn)
| Destination | 3Q26 | Share | QoQ | YoY | 2Q26 | 3Q25 |
|---|---|---|---|---|---|---|
| United States | 845.9 | 75.8% | +3.8% | +135.9% | 815.0 | 358.5 |
| Mainland China | 20.7 | 1.9% | (4.5)% | (12.4)% | 21.7 | 23.6 |
| Netherlands | 18.3 | 1.6% | +27.4% | +33.1% | 14.4 | 13.7 |
| Japan | 11.6 | 1.0% | (14.2)% | (24.5)% | 13.5 | 15.4 |
| Other countries | 219.1 | 19.6% | +27.4% | +160.8% | 172.0 | 84.0 |
| Total | 1,115.5 | 100.0% | +7.6% | +125.2% | 1,036.6 | 495.3 |
United States: the engine, and a slowing one
Shipments to the United States were NT$845.9bn, 75.8% of revenue, against 72.4% a year ago and 78.6% in the second quarter. They more than doubled year over year, which is the AI rack business: Quanta's server racks are assembled for, and largely delivered to, US cloud data centers. Sequentially, though, US revenue rose only 3.8%, and the quarter's growth came from "other countries," up 27.4% to NT$219.1bn. Destination is where product ships, not who buys it, so a rack built for a US platform and installed outside the US reads as "other countries." The report does not explain the shift.
Assessment: The US line slowing to low single digits sequentially is the number to watch in the fourth quarter. A rack business that grew US shipments from NT$358.5bn a year ago to NT$845.9bn cannot double again from here without the next GPU platform. That is why the timing of that platform, not the quarter's revenue, drives the shares.
Everywhere else: notebooks and smaller markets shrinking
Mainland China fell 12.4% year over year and Japan 24.5%, and both fell sequentially. Those are notebook and consumer-electronics destinations more than server destinations, which fits management's description of lower notebook shipments (Key Topics, item 4). The Netherlands, a European logistics hub, rose 27.4% on the quarter.
Assessment: The non-US, non-"other" lines are now 4.5% of revenue, so their decline matters for mix more than for growth. Notebooks are a shrinking share of a business whose margin is now set by racks, and management's own explanation of the quarter puts notebook material costs among the reasons gross margin fell.
Customer concentration
Quanta names its customers nowhere. The 2025 annual financial statements list three unnamed customers that each took more than 10% of revenue.
| Customer | 2025 revenue (NT$bn) | Share of 2025 | 2024 revenue (NT$bn) | Share of 2024 |
|---|---|---|---|---|
| A | 624.3 | 29.4% | 441.1 | 31.3% |
| B | 433.4 | 20.4% | 210.9 | 14.9% |
| C | 315.6 | 14.9% | 292.1 | 20.7% |
| Three customers | 1,373.3 | 64.7% | 944.1 | 66.9% |
"Major customers (More than 10% of revenue)"
— Quanta, 2025 consolidated financial statements, note 14(d), which lists customers A, B and C without names
Assessment: Concentration is high but less extreme than at some AI-server peers: two-thirds of revenue from three customers, with a long tail of notebook and server customers behind them. Customer B more than doubled in 2025. The 2026 figures, due with the annual financial statements next spring, will show how much further the rack ramp concentrated the book.
The monthly run-rate (NT$bn)
| Month | 2026 | 2025 | YoY | MoM |
|---|---|---|---|---|
| April | 339.9 | 154.0 | +120.7% | (6.3)% |
| May | 311.5 | 160.2 | +94.4% | (8.4)% |
| June | 385.2 | 189.9 | +102.9% | +23.7% |
| July | 366.3 | 158.3 | +131.3% | (4.9)% |
| August | 424.0 | 152.8 | +177.4% | +15.8% |
| September | 325.3 | 184.1 | +76.7% | (23.3)% |
| October | 353.3 | 173.2 | +104.0% | +8.6% |
October revenue of NT$353.3bn rose 8.6% from September and 104.0% from October 2025, but it was 5.0% below the third quarter's monthly average of NT$371.8bn. Three months at October's level would be a fourth quarter of about NT$1,060bn, 5% below the third. Ten months of 2026 revenue, NT$3,314.6bn, already exceed all of 2025 by 56%.
Assessment: The run-rate entering the fourth quarter is a little below the third quarter's, which fits the platform transition management described: fewer racks, at higher prices, while customers move from one GPU generation to the next. Quanta's months are lumpy (August was 30% above September), so one month settles little. The November and December filings, due by December 10 and January 10, will show whether the quarter ends flat, as we assume, or follows October down.
Working capital and funding (NT$bn)
| Balance | Sep 30, 2026 | Jun 30, 2026 | Dec 31, 2025 | Sep 30, 2025 |
|---|---|---|---|---|
| Inventories | 856.1 | 790.2 | 512.9 | 334.6 |
| Finished goods | 396.7 | 329.4 | 127.5 | 96.8 |
| Raw materials | 393.6 | 394.9 | 376.4 | 218.9 |
| Accounts receivable | 931.8 | 765.8 | 495.0 | 429.8 |
| Accounts payable | 1,264.9 | 1,030.9 | 535.2 | 423.2 |
| Receivables + inventory - payables | 523.0 | 525.1 | 472.7 | 341.2 |
| Days of receivables / inventory / payables | 77 / 74 / 110 | 67 / 73 / 95 | 71 / 79 / 82 | 80 / 67 / 84 |
| Cash and equivalents | 280.8 | 269.0 | 178.3 | 152.4 |
| Short-term borrowings | 303.9 | 339.9 | 241.7 | 166.4 |
| Bonds payable, mostly convertible | 57.6 | 57.2 | 56.6 | 29.3 |
| Long-term borrowings | 20.1 | 19.1 | 16.6 | 2.6 |
| Net debt | 100.7 | 147.3 | 136.6 | 45.8 |
| Equity attributable to parent | 344.2 | 243.6 | 243.6 | 213.5 |
| Net debt / equity | 29% | 60% | 56% | 21% |
Receivables rose NT$166.0bn in the quarter and inventory NT$65.9bn, almost all of it finished goods, which rose 20.5% to NT$396.7bn, four times the level of a year ago. Payables rose NT$234.0bn and carried all of it: receivables plus inventory less payables was NT$523.0bn, flat on the quarter while revenue rose 7.6%. Operating cash flow in the third quarter alone was about NT$22.1bn (nine months NT$68.0bn less the first half's NT$45.9bn), capital spending about NT$10.6bn, so free cash flow was positive by about NT$11.5bn. The August GDR then added NT$69.7bn of equity, and net debt fell from NT$147.3bn to NT$100.7bn. Quanta also sells receivables to banks without recourse; NT$64.5bn of sold receivables were outstanding at September 30, against NT$24.4bn a year earlier.
Assessment: This is the part of Quanta's print that compares best with other rack builders. Its suppliers, not its balance sheet, carry the build: payables run at 110 days against 77 days of receivables and 74 of inventory. That gives positive free cash flow at a revenue growth rate of 125%. The cost is a dependence on supplier terms. If suppliers shortened payment terms by ten days, about NT$115bn of working capital would move onto Quanta's balance sheet. The NT$396.7bn of finished goods is the line we watch most closely: finished racks are waiting for shipment or customer acceptance, and a platform transition is the moment such inventory is most exposed.
Key Topics & Management Commentary
Overall Management Tone: Confident on demand and on 2027, defensive on margin, and more cautious in the script than in the answers. The prepared remarks spent several minutes arguing that falling percentages reflect the pass-through value of the racks rather than weaker pricing, and they closed with a new and unusually candid line about tighter permitting buying the industry "a healthier digestion period." In the Q&A the chief financial officer was steady and specific on orders, inventory protection and consignment, and deflected only on long-range margin and on quantifying consignment.
1. A Chinese-only call, and a long defense of the margin
Quanta's board approved the quarter in a filing that reached the exchange at 14:42 Taiwan time on Wednesday, November 11, after the 13:30 close. Forty-eight minutes later, at 15:30, management held its results teleconference, hosted as in every recent quarter by a broker, this time Macquarie. It ran just over an hour: about twenty minutes of prepared remarks read by investor relations, forty minutes of questions answered by the chief financial officer, and a short close from the vice chairman and president. The record is a Chinese audio file and a deck that is figures only, closing with a slide that reads "Q&A". A large part of the prepared remarks was an argument about how to read the margin.
"However, this is not a weakening of pricing power or a deterioration in core profitability. It is caused by the continuing jump in the unit price of the multi-million-dollar AI racks in the product mix; as customer-specified chips and other key components become a larger share of the value, the structural dilution created by gross-basis accounting inevitably occurs."
— Carol, Investor Relations, prepared remarks (our translation)
The remarks went on to ask investors to judge the company on the absolute amount of operating profit, which grew 41% in nine months, and on a return on invested capital they put above 20%, and they warned that margin guidance for the next few quarters would keep converging. The chief financial officer repeated the point in the first exchange of the Q&A.
"That is why we hope that when investors look at us, or at any ODM, they look as far as possible at the growth in the absolute amount of operating profit and its compound growth rate; that is, operating profit should grow faster than expenses, which is still fairly positive."
— 楊俊烈 (Yang), Senior Vice President and CFO (our translation)
Assessment: The argument is correct as accounting and incomplete as investing. Gross-basis revenue on racks that carry millions of dollars of customer-specified GPUs does dilute the percentage without changing the profit per rack. And the test management proposes, operating profit growing faster than expenses, passes only on the year-over-year basis it chose: operating income +75.3% against expenses +33.9%. Sequentially it failed. Expenses rose 16.7% while operating income fell 6.0%, and even before the write-down operating income rose only 6.2%. We adopt management's test, on both bases, and will hold the company to it each quarter.
2. Revenue was public in October; the print-day news was the margin
Monthly filings had told the market by early October that third-quarter revenue would be about NT$1,115.5bn, up 125% year over year. Every one of those filings gives the same one-phrase explanation for the change. What the market did not know was the margin, and it came in at 4.75% gross, the lowest of the past five quarters, and 2.89% operating, the second-lowest, with operating income down 6.0% sequentially on revenue up 7.6%.
"Remarks / explanation of the change in revenue: Change in product mix"
— Quanta, monthly revenue filings for each month from January through October 2026 (our translation)
Assessment: For a company whose revenue doubles, a sequential fall in operating income is the headline, and the stock traded on it. The detail (Topic 3) is more forgiving than the headline, but the market cannot see the detail without reading a Chinese financial report the evening of the print.
3. NT$10.5bn of inventory write-downs in nine months
The single largest swing in the quarter is a cost line most readers will not see. Quanta charged NT$4.92bn of "loss of inventory valuation and obsolescence" to cost of sales, against NT$0.72bn in the second quarter and NT$1.31bn a year ago, plus NT$2.18bn of scrapping and disposal losses. Nine-month write-downs reached NT$10.5bn, against NT$1.4bn for the same period of 2025 and NT$3.1bn for all of 2025. The write-down alone was 44bp of revenue this quarter. The report gives the standard explanation and nothing specific.
"The write-down of inventory valuation was due to obsolescence or being outdated resulted in the net realizable value of inventory to be lower than the cost; hence, the write-down was recognized as operation costs."
— Quanta, 2Q26 consolidated financial statements, note 6(g); the 3Q26 report's note 6(7) carries the same wording in Chinese
The pattern, a large charge in the first and third quarters and a small one in the second, does not match revenue, which rose every quarter. It does match the timing of platform transitions and of the sharp rise in memory and other component prices through 2026, both of which can push the realizable value of components below cost. Finished goods rose to NT$396.7bn at September 30, four times a year earlier. Management's explanation of the inventory build points forward rather than back.
"The inventory increase is mainly because high-priced key components are being stocked in advance, as AI server racks on the new-generation GPU platform gradually enter small-batch mass-production shipments in the fourth quarter."
— Carol, Investor Relations, prepared remarks (our translation)
Assessment: We treat write-downs as a recurring cost of running a NT$856bn inventory through a fast product cycle, not as a one-off. Our estimates carry about NT$2.5bn a quarter. On that basis the third quarter's gross margin would have been about 5.0%, slightly above the second quarter's on the same basis. The issue is not that margins collapsed but that they stopped rising once the AI mix stabilized, and management has guided them slightly lower again for the fourth quarter as the next generation of racks, at close to twice the price, begins to ship.
4. Notebooks: higher material costs, fewer units and a customer dispute denied
Notebooks were the other half of the margin explanation. Shipments were 9.6 million units, down 17% on the quarter and 24% on the year, and management expects another double-digit decline in the fourth quarter, narrower than the third's, and a fall of 15–20% for the full year.
"Although notebook shipments were slightly better than the sequential decline of more than 20% we expected last quarter, the third quarter is usually the peak of the year for notebook shipments, and market demand in the second half of this year has been weak."
— Carol, Investor Relations, prepared remarks (our translation)
Management also said that in 2027 notebooks will, for the first time, be a single-digit percentage of revenue, and the vice chairman closed the call by tying next year to memory prices, which he said "next year depends a great deal on," and which, "from what we hear at present," may stay about where they are now. On October 7 Quanta filed a statement on the exchange denying a television programme's claim that it had threatened a customer with a production halt. The filing does not name the customer or the dispute, but it calls coordination of supply-chain materials normal practice in the industry, and it volunteers a shipment data point.
"The 'threat of a production halt against a customer' referred to by the programme is pure fabrication. In fact, coordination of supply-chain materials and flexible scheduling are normal practice in the industry; with close collaboration between both parties, that customer's September notebook shipments were better than expected."
— Quanta, media clarification filed October 7, 2026 (our translation)
Assessment: The clarification and the margin explanation describe the same problem from two sides: component costs, memory above all, are rising faster than notebook prices, and the cost has to be shared between the assembler and the brand. With servers now more than four-fifths of revenue, Quanta's notebook business is under a fifth, so the volume matters less than the margin, and the margin is where the dispute is.
5. Operating expenses: research and development up 32% in a quarter
Operating expenses rose 16.7% sequentially to NT$20.8bn, the first quarter of 2026 in which they rose faster than revenue. Research and development drove it: NT$12.8bn, up 31.8% on the second quarter and 31.9% on the year, after NT$9.5bn and NT$9.7bn in the first two quarters. General and administrative costs rose 11.3%; selling expenses fell 20.5%. Part of the rise is profit-linked: the employee remuneration accrual, set by the articles of incorporation at no less than 2% of pre-tax profit before the accrual, was NT$2.97bn in the quarter against NT$1.63bn a year ago, spread across cost of sales and operating expenses.
"Research and development expenses rose NT$3.08 billion in the quarter, mainly development costs ahead of the mass production of new AI server products, while selling and administrative expenses were stable."
— Carol, Investor Relations, prepared remarks (our translation)
Assessment: A one-quarter jump of NT$3.1bn in research spending is, on management's own account, the cost of developing new AI server products before they reach mass production and earn revenue. We model research and development near this quarter's level rather than reverting. That is about 28bp of operating margin that does not come back unless revenue grows into it.
6. The non-operating line is a currency gain financing an interest bill
Non-operating income of NT$2.7bn looks steady against NT$2.3bn and NT$2.5bn, but its pieces are moving apart. Foreign-exchange gains rose to NT$3.9bn, 11.2% of pre-tax income, while net interest expense widened to NT$2.3bn from NT$1.7bn in the second quarter and NT$0.7bn a year ago, as finance costs doubled year over year to NT$4.0bn.
| Non-operating line (NT$m) | 3Q26 | 2Q26 | 3Q25 |
|---|---|---|---|
| Interest income | 1,680 | 2,418 | 1,361 |
| Finance costs (interest expense) | (3,975) | (4,111) | (2,035) |
| Foreign-exchange gains, net | 3,903 | 3,273 | 2,271 |
| Fair-value gains on financial instruments | 229 | 663 | 400 |
| Fees on receivables sold | (281) | (452) | (299) |
| Dividend income | 389 | 97 | 298 |
| Other gains and losses | 738 | 434 | 545 |
| Share of associates | (2) | (4) | (5) |
| Total non-operating income | 2,681 | 2,317 | 2,535 |
"The proceeds will be used to fund foreign-currency purchases of materials."
— Quanta, material-information filing of August 3, 2026, on the GDR (our translation)
Quanta buys components and sells racks in US dollars and reports in New Taiwan dollars, so it carries large dollar balances on both sides of its balance sheet. The report does not split the currency line between translation and hedging. What it does show is that the gains have recurred: NT$3.2bn in the first quarter, NT$3.3bn in the second and NT$3.9bn in the third, NT$10.4bn for nine months against NT$5.0bn a year earlier, while net interest expense rose almost as fast.
Assessment: Strip out the currency gain and pre-tax income would have been about NT$31.0bn, down 6.9% sequentially, much closer to the operating trend. Because the gains have recurred we do not model them at zero, but we carry them at about NT$2bn a quarter, below this year's run-rate, and we treat this quarter's NT$0.80 a share of after-tax currency gain as the main low-quality support to EPS.
7. Suppliers fund the build, and banks buy the receivables
Payables of NT$1,264.9bn are now 55% of total assets. They rose NT$234.0bn in the quarter, more than receivables (NT$166.0bn) and inventory (NT$65.9bn) together, and that is why a company growing revenue 125% generated NT$22.1bn of operating cash in the quarter. On top of that, Quanta sells receivables to banks without recourse: NT$64.5bn of sold receivables were outstanding at September 30, with US$2.0bn advanced, against NT$24.4bn a year earlier.
"According to the agreements, the Group does not have the responsibility to assume the default risk of the transferred trade receivables, thus is eligible for derecognizing the financial assets."
— Quanta, 2Q26 consolidated financial statements, note 6(e); the 3Q26 report's note 6(5) carries the same terms
Management told the call that the overall cash conversion cycle was more than two weeks shorter than a year earlier, which matches our arithmetic (about 41 days against 62).
Assessment: Quanta's working-capital model is the strongest part of this print. Its 110 days of payables cover most of its 151 days of receivables and inventory, leaving a cash cycle of about 41 days against 62 a year ago, so growth largely funds itself as long as the main suppliers keep their terms. The risk is the mirror image: the model depends on a small number of component suppliers, one of which supplies most of the value in an AI rack.
8. The GDR: US$2.2bn at NT$287, and a share price now below it
Quanta priced a global depositary receipt offering on July 29: 49 million GDRs at US$44.17 each, five shares per GDR, 245 million new shares at about NT$287.06 a share, for US$2.16bn. The shares were issued on August 3 and listed in Luxembourg. The price was 7.3% below the July 29 close, and the shares fell 9.85% the next day against a flat market. The issue added 6.3% to the share count (5.97% of the enlarged total, as the company states it) and NT$69.7bn of equity.
"This cash capital increase issues 245,000,000 common shares to participate in the GDR offering, with an equity dilution ratio of about 5.97%, which will not cause material dilution to shareholders' equity."
— Quanta, material-information filing of August 3, 2026 (our translation)
Assessment: The raise was not forced: Quanta's free cash flow was positive before it. It lowered net debt to 29% of equity and gave the company room for a larger rack ramp in 2027, and its cost is visible in this quarter's EPS, down 4.7% while net income was flat. With the shares at NT$276.50, the GDR buyers are under water, which is a modest overhang rather than a thesis issue.
9. Building in the US, funded from Taipei
The same board meeting routed another US$50M of capital to Quanta Manufacturing Nashville (QMN) through the Quanta Manufacturing Inc. chain, taking that chain's cumulative investment to US$1,478M. It also lent another NT$54.1bn to QCH Inc., a 100%-owned Nevada subsidiary, for working capital, bringing the loan balance to NT$223.0bn, 64.8% of parent equity. Capital spending for nine months was NT$32.2bn, three times last year's NT$10.9bn, and property, plant and equipment rose to NT$96.7bn from NT$73.0bn at year-end; right-of-use assets nearly doubled to NT$14.6bn.
On the call, management named the California and Thailand sites as the focus of expansion, said the goal of doubling AI capacity between the end of 2025 and the end of 2026 is on track, and explained the jump in right-of-use assets: overseas it mostly leases existing industrial buildings rather than building plants, so the real capital spending is in fit-out, power and cooling. Third-quarter capital spending was about NT$12.8bn on the company's own measure, and next year's will be a little higher.
"We expect 2027 capital expenditure to be slightly higher than this year's, between NT$40 billion and NT$45 billion, with the growth in total investment narrowing."
— Carol, Investor Relations, prepared remarks (our translation)
One capacity decision went the other way. On October 29 Quanta cancelled the purchase of land and buildings from AU Optronics that its board had approved on July 28, because the two sides "did not reach agreement on the plant's specifications and delivery schedule." The call said what replaces it.
"The new capacity originally planned for that plant is expected to join production in 2028."
— Carol, Investor Relations, prepared remarks (our translation)
Assessment: The direction is clear: US assembly capacity, financed by the parent, with working capital lent to the US operating companies rather than raised locally. A loan balance equal to two-thirds of parent equity is large, but it is intercompany and eliminated on consolidation; what matters is that the US build is funded. The AU Optronics cancellation moves a Taiwan expansion to 2028 and abroad, and management said its 2028 capacity plan is unchanged. Capital intensity of about 1% of revenue keeps the plant side light; the inventory that runs through the plants is the heavy part.
10. Selling the equity portfolio: NT$27bn of assets "activated"
The board also authorized management to sell, in batches and at its discretion, the equity portfolio Quanta carries at fair value through other comprehensive income, about NT$27.2bn at the latest book value. Under IFRS 9 any gain goes straight to retained earnings, so it will not show in EPS.
"This portfolio is classified as FVOCI; under IFRS 9 the disposal gains or losses are transferred to retained earnings and do not affect profit or EPS for the period."
— Quanta, material-information filing of November 11, 2026 (our translation)
Assessment: This is another source of cash, worth about two-fifths of the GDR. Quanta's total dividend income for nine months was NT$0.5bn. Together with the GDR and the intercompany loan it says the company is assembling liquidity for a bigger 2027, not shoring up a weak 2026. It also suggests that management prefers cash to market exposure at today's prices.
11. A new caution in the script: "a healthier digestion period"
The prepared remarks ended with a passage no press account we saw picked up, and it is the most cautious thing management said. It concedes that tighter permitting for power and data centers in the main markets will make the largest cloud customers more careful with their capital spending, and it describes the result as a slower rhythm rather than a change of direction.
"As reviews of infrastructure construction, infrastructure regulation and utility provision tighten in the main markets, large cloud service providers will also become more cautious and more focused on efficiency in deploying capital expenditure."
— Carol, Investor Relations, prepared remarks (our translation)
"What changes is the rhythm of AI's advance, buying a healthier digestion period for the whole infrastructure build-out and its commercial deployment; it does not change the long-term direction in which AI will profoundly reshape modern society."
— Carol, Investor Relations, prepared remarks (our translation)
In the Q&A the chief financial officer was firmer: state restrictions on construction predate the election, and no customer has changed an order (Analyst Q&A, first entry). The vice chairman's closing view of 2027 was "cautiously optimistic."
Assessment: A company does not volunteer the phrase "digestion period" in its script by accident. We read it as management preparing investors for slower growth in customers' capital spending after 2027's ramp, not as a warning about the next two quarters, which rest on orders already placed. It is the main reason our 2027 growth assumption sits at the bottom of the range management gave.
Guidance & Outlook
Quanta gives no numeric revenue, margin or earnings guidance in writing, and its documents for this quarter contain no outlook. The guidance below is what management said on the call, from the machine transcript in our translation, checked against press accounts; where the transcript is garbled we say so. We do not turn qualitative language into numbers.
| Item | What management said on the call | Our reading |
|---|---|---|
| 4Q26 gross margin | Slightly lower than the third quarter | We model 4.65%, from 4.75% |
| 4Q26 operating margin | Target is to hold it flat with the third quarter | We model 2.79%, slightly below 2.89% |
| 4Q26 AI servers | Revenue up sequentially on higher-priced racks; total rack count slightly lower during the generation change | A transition quarter |
| 4Q26 general-purpose servers | Shipments up sequentially | Driven, management said, by AI-agent demand |
| Notebooks | 4Q shipments down double digits sequentially, less than in 3Q; 2026 down 15–20%; 2027 down roughly with the industry (high single digits); below 10% of revenue in 2027 | A shrinking, cost-squeezed line |
| 2026 mix | Servers above 80% of revenue since January; AI servers 75–80% of servers and about 60–65% of total revenue for the year | Consistent with our 83% server share |
| 2027 servers | AI and general-purpose servers both grow at double-digit rates; AI servers at a "high double-digit" rate; servers 85–90% of revenue, AI more than 80% of that | We model AI +60%, the bottom of that range |
| 2027 margin | Margin percentages to keep converging for several quarters as rack prices rise; one press account adds an aim to hold operating margin near the second-half 2026 level | We model 2.9% operating margin |
| Consignment | More high-priced GPU projects moving to consignment (GPUs, sometimes with memory; never yet a whole rack); buy-and-sell "well above half" of new racks in 1H27 | Higher margin percentage, lower revenue |
| Capacity and capital spending | AI capacity doubling by end-2026 on track; 2027 capital spending NT$40–45bn, slightly above 2026; mostly power and testing | About 1% of revenue |
| Orders and policy | No customer has changed orders since the US midterms; tighter permitting will make cloud customers more careful with capital spending | Firm on 2027, cautious beyond it |
| Revenue, EPS | Not guided | n/a |
"Gross margin will be slightly lower than this quarter's, and our target is to hold the operating margin flat with this quarter."
— Carol, Investor Relations, prepared remarks (our translation)
"AI servers and general-purpose servers will continue to grow at double-digit rates year over year; the server business's share of total revenue will keep rising to 85% to 90%, and AI servers are expected to account for more than 80% of it."
— Carol, Investor Relations, prepared remarks (our translation)
Implied fourth quarter: October at NT$353.3bn puts the fourth quarter on a NT$1,060bn pace, and management's description (more AI-server and general-server revenue, much less notebook revenue) points to a quarter close to flat. We use NT$1,110bn.
Street at: Fourth-quarter consensus after the print is NT$1,124–1,147bn of revenue and NT$6.59–6.70 of EPS. Our NT$1,110bn and about NT$5.9 of diluted EPS sit below both, mainly on smaller assumed currency gains. Full-year 2026 consensus EPS is NT$25.9–26.6, and the sources do not state its basis. Our NT$26.2 basic sits inside that range, and our NT$25.0 diluted sits just below it.
Guidance style: Verbal, qualitative and margin-focused, as is the company's practice. In August management said full-year margins would hold at the first-half level (a 4.92% gross margin and a 3.10% operating margin); the third quarter and the fourth-quarter guide make that unlikely, and the call did not revisit the promise. This quarter's message was that margin percentages will keep falling while the operating margin is held, and that the only way to lift the percentage is to report less revenue.
Analyst Q&A Highlights
Orders after the US midterms, and what binds customers
The host opened with the question investors arrived with: whether cash-flow worries at the cloud platforms and tighter data-center policy after the US elections had changed next year's orders, and whether the capacity reservations management cites are binding. The answer was unequivocal on orders and candid on the contracts: there is no written agreement, but build-to-order racks with long lead times carry, in management's description, more mutual commitment than the consumer business ever did.
Q: "As a major ODM in direct contact with the CSPs, has Quanta recently seen any change in next year's orders?"
— Kalin, Macquarie (host)
A: "The restrictions some states place on this construction already existed before the election, so this is not a new event. Second, so far, even with that event, we have not seen customers adjust their orders. And in the recent CSP earnings calls we also saw that they have not cut their capital spending."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: The right answer, and also the only one available a quarter before orders would show anything. October revenue up 104% supports it. The prepared remarks' "digestion period" line (Key Topics, item 11) is the more informative statement about 2028.
Who carries the inventory if a customer delays
The follow-up went to the balance sheet: if a customer defers pull-ins, how does Quanta protect a NT$856bn inventory? Management said custom racks are bought by the customer at cost, and that a customer who delays covers the carrying costs. The protection, it added, holds as long as AI demand itself does not disappear.
Q: "If customers' orders change, or their pull-ins are deferred, how do we handle inventory control?"
— Kalin, Macquarie (host)
A: "In principle customers take what they originally ordered; because these are custom products, the customer buys them at cost, and that is not a problem. If there is a delay, because of the build cost, the customer will cover the related interest, warehouse rent and other costs that arise."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: This is a stronger protection than the inventory line suggests, and it bears on the write-downs. If finished custom racks are effectively pre-sold at cost, the NT$10.5bn of write-downs this year more likely sit in components and notebook inventory than in racks. That is our inference; management did not say where they sit.
Why AI rack shipments fell in a record quarter
Two analysts pressed on the third quarter's slight decline in AI rack shipments. Management corrected one premise (AI-server revenue rose; only the number of racks fell) and attributed the dip to customers moving between GPU generations on different schedules, some pulling early and some late. The larger volumes of the new generation arrive in the first quarter of 2027.
Q: "So what was the reason in the third quarter? And when do we expect to see a larger surge in AI shipments? Can we expect it in the first quarter?"
— Albert, analyst (firm not legible in the machine transcript)
A: "It is a gradual transition process, and the larger volumes are still in the first quarter of next year, that is right."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: Consistent with the destination data, where US shipments rose only 3.8% sequentially, and with October's revenue running 5% below the third-quarter average. Two transition quarters in a row is the near-term risk; a first quarter without the volume step-up would be the first real test of the thesis.
How fast AI servers can grow in 2027
An analyst asked whether 2027 could repeat 2026's triple-digit AI growth, given that new-generation GPUs carry much higher prices. Management lowered the bar explicitly, citing the base.
A: "I think the base is very high, so triple-digit growth may be a little more difficult; it is still high double-digit growth."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: "High double-digit" growth on a 2026 base of about NT$2.5tn of AI-server revenue (60–65% of our NT$4.07tn) would add at least NT$1.5tn of revenue at the bottom of the range. Whether consignment deals not yet signed are inside that estimate is the one point the transcript leaves ambiguous, and it matters: each consignment deal lowers reported revenue.
What consignment covers, and how much of 2027 is still buy-and-sell
Two analysts probed the consignment shift. Management confirmed that buy-and-sell, where Quanta buys the GPUs and books them in revenue, will remain well over half of new racks in the first half of 2027. Consignment so far covers the costliest components, GPUs and sometimes memory, and has not yet been applied to a whole rack. It takes customer-by-customer negotiation and is not, in management's phrase, one size fits all.
Q: "Can I take it that, at least for the new racks in the first half of next year, buy-and-sell is still well above half?"
— Alex Wang, analyst (firm not stated in the machine transcript)
A: "That is certain. (…) So we hope that more customers will accept the consignment model next year."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: Consignment builds through 2027 rather than arriving at once, which means first-half 2027 revenue estimates built on buy-and-sell are not yet at risk from it, and first-half margin percentages will not get much help from it either.
Can margin and revenue both improve?
The longest exchange asked management for a framework to forecast revenue and margin when every customer's terms, component mix and consignment status differ. Management conceded that forecasting is hard and that simple valuation shortcuts mislead, and then stated the trade-off more bluntly than any analyst had.
Q: "Could the CFO give us a little more guidance on how to anticipate this momentum, and what it looks like next year and even the year after?"
— Angela, analyst (firm not stated in the machine transcript)
A: "The market's expectations for AI stocks are somewhat contradictory: it wants your margin to improve, and it wants your revenue not to fall, and those two things basically cannot happen at the same time."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: The most useful sentence on the call for valuing the stock. It tells investors to value Quanta on operating profit, not on revenue growth or margin percentage, which is what we do. It also amounts to a refusal to give a margin framework, which leaves the multiple exposed to every quarter's headline percentage.
Component shortages: memory, again
Asked which parts are short as new platforms ship slowly across the supply chain, management named memory and the circuit boards and CPUs that depend on it, said the shortages are not new, and said the effect on its cloud customers has been much smaller. The vice chairman returned to memory in his closing remarks as the main swing factor for 2027.
Q: "So I would like to ask whether Quanta sees the same thing, and if so, which parts are in shortest supply?"
— Alex Wang, analyst (firm not stated in the machine transcript)
A: "It is memory and some related PCBs and CPUs. (…) But these have been there all along."
— 楊俊烈 (Yang), Senior Vice President and CFO
Assessment: Memory is a cost problem for notebooks and a timing problem for racks. Management's claim that rack shipments were not delayed by shortages this quarter is consistent with revenue up 7.6%, but one Street report describes component and yield problems on the new platforms in the fourth quarter, so we would not rule out a timing slip.
What They're NOT Saying
- What the NT$4.9bn write-down was for. The largest swing in the quarter's gross margin gets one boilerplate sentence in the report. Which components, which platform, and whether the first-quarter and third-quarter charges have the same cause would tell investors whether this is a cost of the cycle or a cost of one program.
- The product split, in writing. The share of servers, the AI-server growth rate and the notebook trend reach investors only through the call, and this quarter the company published the call only as Chinese audio, with no transcript. A NT$1.1tn-revenue company reporting a single segment leaves its most important numbers to a recording most foreign investors cannot use.
- What the research spending buys, and whether it recurs. NT$3.1bn of extra research and development in one quarter is about 28bp of revenue. The call called it pre-production development for new AI products; it did not say whether the step-up is one platform's cost or the new run-rate.
- The August margin promise. In August management said full-year margins would hold at the first-half level. The third quarter's 4.75% gross and 2.89% operating margins, and a fourth-quarter guide that is lower again on gross margin, make that unlikely, and the call did not revisit it.
- What the notebook dispute was about. The October 7 filing denies a threat to halt production and calls the issue material coordination. It does not say whether memory and other cost increases are being passed through to notebook customers, which is the question that decides that business's margin in 2027.
- Why raise more cash when free cash flow is positive. In five months the company sold US$2.2bn of new shares, authorized the sale of NT$27bn of investments and lent another NT$54bn to its US operating company. The documents give the uses in a phrase each ("foreign-currency purchases of materials," "working-capital needs," "activating assets") and no total funding plan.
- The tax rate. A rate of 17.9% against 21.3% in the second quarter added about NT$1.2bn to net income, and the report gives no reason.
- How much consignment is in the 2027 numbers. Management would not size the consignment share, and the transcript leaves ambiguous whether deals already agreed are inside the high double-digit growth estimate. Each consignment deal lowers reported revenue, so the answer decides how 2027 growth will look.
- Customer names. Three customers took 64.7% of 2025 revenue, and the documents name none of them.
Market Reaction
All levels are TWSE closing prices in New Taiwan dollars; the benchmark is the TAIEX. The board filing reached the exchange at 14:42 Taiwan time on Wednesday, November 11, after the 13:30 close, so Thursday, November 12 was the first session to price the results.
- Pre-print setup: NT$298.00 (~$9.06) at the November 11 close, up 9.6% year to date against a 52.8% gain for the TAIEX, down 7.5% over the prior 30 days and up 2.2% over twelve months. The 52-week closing range was NT$260.00 to NT$417.00, and the stock sat 28.5% below the June 3 high. It fell 12.9% from October 6, the TAIEX's closing high, to November 6, against a 20.3% fall for the index, then rose only 2.6% over the next three sessions while the TAIEX rebounded 11.5%. The October and November falls came on index days: a hotter-than-expected US inflation report on October 15 (TAIEX -3.4%, Quanta -2.0%), and, after the November 3 US midterm elections in which Democrats won both chambers of Congress, falls of 3.2% and 4.5% on November 5 and 6 on concern that approval of new AI data centers would slow. The government's market-stabilization fund intervened on November 6.
- Reaction session (November 12): opened at NT$294.00 (-1.3%), traded between NT$278.50 and NT$296.00, and closed at the low, NT$278.50 (~$8.46), down 6.5% or NT$19.50. Volume was 32.8 million shares, 2.6 times the 30-day average. The TAIEX fell 2.6%, and Wiwynn fell 6.5%. The index fell on reports that NVIDIA's board was weighing the removal of its chief executive, Jensen Huang, who resigned later that day, US time; Taiwan's NVIDIA suppliers led the decline.
- Since: NT$276.50 on November 13 (-0.7%), on a day the TAIEX rose 1.35% and Wiwynn 4.1%. From the pre-print close, the stock is down 7.2% while the TAIEX is down 1.3%.
A margin sell-off with high volume
Quanta fell 3.9 points more than the index on the reaction day, closed at its low, and traded 2.6 times its normal volume. About two-fifths of the fall was the market: AI hardware sold off across Taiwan that day. The rest was the print. Revenue was known; what the board added was a lower gross margin and lower operating income, and the write-down detail that softens both sat in a Chinese-language report. The next session, when the index and Wiwynn recovered, Quanta did not.
The underperformance began long before the print
The more striking number is the setup. Quanta entered the print up 9.6% for the year while the TAIEX was up 52.8%, and 28.5% below its June high, while its own nine-month earnings had already passed all of 2025. The shares fell 9.85% the day after the GDR priced at a 7.3% discount in late July, recovered to NT$347.50 by September 7, and have since given all of that back and more. In the month before the print they fell less than the index, but they missed almost all of the November rebound. Part of the gap is NVIDIA exposure: on November 13, when custom-chip design names rallied and the index rose 1.35%, Quanta fell again. Part is the company's own: a year of falling margin percentages, a July share issue whose buyers are now under water, a notebook business in decline, and disclosure thin enough that the headline percentage is most of what foreign investors see.
Street Perspective
No broker downgraded to a sell after the print, but almost every one cut its target, and one local house moved to neutral. Post-print targets we could see run from about NT$324 to NT$450, and the published median fell from about NT$432 to the NT$400–410 range. The cuts were about multiples more than estimates: several houses lowered the multiple they apply to 2027 earnings to 11–12 times, and at least one raised its 2027–28 EPS forecasts at the same time. The debate is about operating margin and what consignment does to the numbers, not about demand.
Debate: where is the operating-margin floor?
Bull view: Gross-margin dilution from high-priced racks was expected and does not mean weaker pricing; management targets a flat operating margin into the fourth quarter and, in one press account, a 2027 operating margin near the second-half level, while revenue keeps growing faster than costs.
Bear view: The third quarter showed no operating leverage at all. The expense ratio rose, research spending jumped for the next platform, and Vera Rubin racks at nearly twice the price will dilute the gross margin further from the first quarter. A target is not a floor.
Our take: The bears are right that the third quarter had no leverage, but they are reading a write-down quarter as a run-rate. Before the NT$4.9bn write-down, operating margin was 3.33% against 3.37%. We model 2.9% for 2027, near the second-half level and below this year, and still get net income up 23%.
Debate: does consignment fix the margin or just shrink the revenue?
Bull view: Moving the costliest components to customer ownership lifts the margin percentage, cuts the working capital and inventory risk Quanta carries, and leaves gross profit per rack roughly where it was. Some customers have already agreed to it for high-priced GPU projects.
Bear view: Management itself says higher margins and higher revenue are unlikely to arrive together. Revenue estimates built on buy-and-sell will have to come down as consignment spreads, and a lower revenue line will look like slower growth even if profit is unchanged.
Our take: Consignment changes the presentation more than the economics, and on balance it lowers risk: less inventory to write down and less dependence on supplier payment terms. We value Quanta on operating income and EPS for that reason, not on revenue, and we would welcome a lower revenue line with a higher margin.
Debate: is this a Quanta problem or a sector multiple?
Bull view: The shares have halved their trailing multiple since June while earnings rose, and the cuts after the print were to multiples, not to demand. Customers have not changed orders since the US election, and 2027 server growth is guided in the high double digits.
Bear view: Some on the Street now question how durable AI demand is after the US midterms, and others cite data-center approval risk. Quanta has among the thinnest margins in the chain and little written disclosure, so it de-rates first when sentiment turns, and its share count has just grown 6%.
Our take: The policy risk is real for the multiple and small for twelve-month earnings, which rest on orders already placed. That is why we value the shares at 10–11 times 2027 earnings, at and below the bottom of the range brokers now use, rather than the 14 times behind the highest targets.
Model Update & Valuation Framework
This is our first report on Quanta and we have no prior model to revise, so we set out the working estimates the rating rests on. They follow management's guidance where it exists (fourth-quarter gross margin slightly below the third quarter's, operating margin held flat and, in one account, a 2027 operating margin near the second-half 2026 level), take the stated 2027 mix at face value, carry inventory write-downs as a recurring cost, and assume currency gains below their 2026 run-rate.
| Item (NT$bn unless stated) | 4Q26E | FY26E | FY27E | Basis |
|---|---|---|---|---|
| Revenue | 1,110 | 4,071 | 5,534 | 4Q flat on 3Q (fewer racks at higher prices, notebooks down double digits); 2027 +36% |
| Gross margin | 4.65% | 4.80% | 4.46% | Guided slightly lower in 4Q; next-platform racks dilute further in 2027 |
| Operating expenses | 20.6 | 75.0 | 86.2 | Research and development near the 3Q run-rate; 2027 +15% |
| Operating margin | 2.79% | 2.96% | 2.90% | Guided flat in 4Q; we sit slightly below; 2027 at the 2H26 level |
| Non-operating income | 0.8 | 7.3 | 0.0 | Currency gains NT$2bn a quarter in 4Q, then offset by interest |
| Tax rate | 20% | 18.5% | 20% | Between the 2Q and 3Q rates |
| Net income to parent | 25.2 | 103.5 | 127.4 | FY26 +38%; FY27 +23% |
| EPS (NT$) | 6.13 basic / 5.92 diluted | 26.2 basic / 25.0 diluted | 29.8 diluted | On about 4,310mn diluted shares, the full post-GDR count |
| Consensus EPS (NT$) | 6.59–6.70 | 25.9–26.6 | about 31.8 (one feed) | Two-source rule; 2027 not confirmed |
The fourth quarter assumes revenue flat on the third, as management described fewer racks shipped at higher prices during the change of GPU generation, more general-purpose servers and notebook shipments down by double digits. Gross margin is 4.65%, slightly below the third quarter as guided, which on our NT$2.5bn write-down assumption is about 4.9% before write-downs. Our diluted EPS of about NT$5.9 is below the NT$6.59–6.70 consensus, mainly because we assume smaller currency gains. For 2027 we assume AI servers grow 60%, the bottom of the "high double-digit" range management gave, general-purpose servers 10%, and notebooks and other products fall 20%, which reproduces management's stated mix of servers at 85–90% of revenue. Our 2027 revenue of NT$5.53tn sits between the two published consensus figures we can see (about NT$5.50tn in the last pre-print FactSet survey and NT$5.94tn on the S&P Global feed). Our diluted EPS of about NT$29.8 is about 6% below the roughly NT$31.8 on that feed. Post-print broker estimates for 2027 that we could see range from about NT$28.8 to NT$35.9.
Valuation. At NT$276.50, the November 13 close, and 4,107.6 million shares, the market value is about NT$1,136bn (~$34.5bn), and net debt was NT$100.7bn at September 30. The shares trade at 10.7x trailing basic EPS of NT$25.76, 10.6x our 2026 EPS and 9.3x our 2027 diluted EPS. Quanta paid NT$15.60 a share in cash this year for 2025, 80% of that year's EPS, a 5.6% yield at today's price. If the payout holds near 80%, the dividend paid in 2027 for 2026 would be about NT$21, a yield near 7.6%; that is our assumption, not a company statement. In June, at NT$417, the shares traded at about 21 times trailing earnings.
| Scenario | 12-month value (NT$) | vs. NT$276.50 | Basis | What has to happen |
|---|---|---|---|---|
| Bull | 405 (~$12.3) | +46% | 11.5x a NT$35.1 2027 diluted EPS | AI servers grow about 90% in 2027 and operating margin holds 3.0% as Vera Rubin ramps on time |
| Base | 300–330 (~$9.1–10.0) | +8% to +19%, plus about 7–8% in dividends | 10x–11x our NT$29.8 | Revenue +36% in 2027, operating margin about 2.9%, write-downs about NT$2.5bn a quarter |
| Bear | 190 (~$5.8) | -31% | 9.5x a NT$20.0 EPS | A platform delay or customer pause leaves 2027 revenue flat on the 4Q26 run-rate at a 2.5% operating margin |
Our base multiple of 10–11 times 2027 earnings is about half the trailing multiple the shares carried in June and at and below the bottom of the 11–14 times that brokers applied after the print. The discount is deliberate. A rack assembler's margin is thin and getting thinner as rack prices roughly double with the next platform, write-downs are now a recurring cost, and the product disclosure is too thin for investors to pay up for growth they can only hear about in Chinese on a conference line. Even so, the base case plus the dividend is a total return of roughly 16% to 27% over twelve months, ahead of what we expect from the S&P 500 and the TAIEX. The bear case is nearly a third down and it needs a platform or customer problem, not a margin quarter.
Bottom line: initiating at Outperform
Rating decision: We initiate coverage of Quanta Computer (TWSE: 2382) at Outperform, with a 12-month value range of NT$300–330 (~$9.1–10.0). The third quarter's headline was poor: gross margin of 4.75%, operating income down 6.0% on the quarter and EPS down 4.7% after the GDR. Underneath it, gross profit before the write-down rose faster than revenue, operating cash flow was NT$22.1bn, and net debt fell by NT$46.6bn. Nine-month earnings already exceed all of 2025. The shares have fallen 34% from their June high to about 9 times our 2027 earnings with a dividend yield above 5%, and the market has priced the margin squeeze more fully than the growth.
What would move us to Hold: a fourth-quarter gross margin before write-downs below 4.7%, or an operating margin below 2.6%, which would mean the flat-margin target failed; another write-down above NT$5bn; days of payables falling below 90, which would mean suppliers are tightening terms; or November and December revenue averaging below NT$330bn a month without an explanation tied to platform timing.
What would move us to Underperform: evidence that the next GPU rack platform slips into the second half of 2027; one of the three main customers cutting rack orders; or operating cash flow turning negative as payables unwind.
Signposts for the fourth-quarter results (early 2027):
| Signpost | What to watch | Bullish if | Bearish if |
|---|---|---|---|
| Monthly revenue | November and December filings | NT$380bn a month or more | Below NT$330bn a month |
| Gross margin | 4Q26 print, before write-downs | 5.0% or better | Below 4.7% |
| Write-downs | Inventory valuation loss in cost of sales | Below NT$2.5bn | Above NT$5bn |
| Working capital | Days of payables; finished goods | Payables above 100 days; finished goods below NT$400bn | Payables below 90 days; finished goods still rising |
| Cash flow | 4Q26 operating cash flow | Positive for another quarter | Negative |
| Operating expenses | Research and development | Flat near NT$12.8bn on higher revenue | Another step up |
| Concentration | 2026 annual financial statements | Top three below two-thirds of revenue | A single customer above a third |
| Disclosure | 4Q26 call materials | An English recording or transcript | Chinese audio only again |
Initiating Coverage: Bull / Bear Matrix
As our first report on Quanta, we set out the framework we will track each quarter.
| Thesis point | Status at initiation | Verdict / what we are watching |
|---|---|---|
| Bull #1: AI racks keep doubling the business | Confirmed | 3Q26 revenue +125.2%; October +104.0% YoY; ten months already 56% above all of 2025; servers above four-fifths of revenue |
| Bull #2: Suppliers fund the growth, so it generates cash | Confirmed | 3Q26 operating cash flow about NT$22.1bn, free cash flow about NT$11.5bn; net working capital flat at NT$523bn; net debt down to NT$100.7bn |
| Bull #3: The price has fallen faster than earnings | Confirmed | Shares 34% below the June high; 9M26 EPS of NT$20.03 already above FY25's NT$19.45; 10.7x trailing EPS, 5.6% yield |
| Bear #1: Gross margin is thin and write-downs are now recurring | Active | 4.75% reported, 5.19% before NT$4.92bn of write-downs; NT$10.5bn of write-downs in nine months |
| Bear #2: Platform and customer concentration | Active | Three customers 64.7% of 2025 revenue; the rack business depends on one GPU supplier's platform cadence |
| Bear #3: Notebook margins squeezed by component costs | Emerging | Management cited notebook material costs and lower shipments (9.6mn units, -24% YoY); notebooks below 10% of revenue in 2027; memory prices the 2027 swing factor |
| Bear #4: Per-share leakage from new shares and thin disclosure | Emerging | GDR added 6.3% to the share count; 179mn potential shares from convertibles; no written product split; the company published the call only as Chinese audio |
Overall: Three bull points confirmed against two active risks (thin, write-down-prone margins and concentration) and two emerging ones (notebook costs and per-share leakage). The cash thesis is the strongest; the margin thesis is the one to monitor each quarter.
Action: Initiate at Outperform. We would add on further margin-driven weakness as long as payables terms hold, and we would reassess if the fourth quarter brings a write-down above NT$5bn or a fall in days of payables.