1815 / 1815

Fulltech Fiber Glass Corp (富喬工業, 1815.TWO, Taiwan TPEx)

WIKI tickertaiwanmaterialsfiberglassglass-clothai-infraccl updated 2026-07-01

#stock #deepdive #fiberglass #glass-cloth #ai-ccl #taiwan #investment #analysis

🔴 Live NT$118 (yf:1815.TWO) · 2026-09-05 · research written 2026-07-01 your re-engage <NT$40–NT$55 (own-DD, cycle-roll) — +115% away

The cleanest listed way to rent Taiwan's second-source position in the AI-server glass-cloth bottleneck. The cycle is real and the margins prove it, but the moat is leased from Nittobo's shortfall on a mid-2027 clock, and the stock has already tripled. Own the dislocation, do not underwrite it as a franchise.

Sector context: industry-wide material (the glass grade ladder, the CCL chain, the full player map) lives on the sector page advanced-packaging. This page is Fulltech-specific and links up rather than re-hosting the primer.

PART I. THE BUSINESS

1. Executive summary

Fulltech Fiber Glass Corp (富喬工業, 1815.TWO, Taiwan TPEx) is a vertically integrated glass-yarn and electronic-grade glass-cloth maker, and the cleanest listed expression of one theme: the AI-server copper-clad-laminate (CCL) bottleneck. Its qualified, volume low-dielectric (low-Dk) product sits at the NE class feeding the M7 CCL grade, and it is also a qualified low-thermal-expansion (T-glass) second source. Those are exactly the two grades that incumbent Nittobo (3110.T) cannot supply enough of before its capacity expansion lands around mid-2027. The thesis is a timing trade on that gap. A scarce, slow-to-qualify capability, an incumbent that raised list prices twice and still cannot meet demand, and a Taiwan pure-play already printing the best gross margin of its five domestic peers.

Price and size (yfinance, 30 Jun 2026; reconcile live in IBKR). Price about NT$101, down roughly 20% from the NT$127 high of 17 April 2026 and flat-to-down since the Q1 print, so much of the good news is already in the tape. Market cap NT$59.0B (about US$1.85B) on roughly 584.3M shares. Enterprise value is close to market cap, around NT$61.6B on a rough net-debt estimate, though net debt is not independently verified here and the Thailand build is partly bank-funded, so treat EV as approximate (flagged as a gap). Trailing PE about 61.6x on FY2025 EPS of NT$1.64. Forward PE about 31x on 2026E EPS of NT$3.27 (CMoney 法人 consensus; 2026E revenue NT$9.591B, up about 58%). Mega Securities (兆豐) is the standout bull with a NT$206 target on 2026E EPS of NT$4.55 and revenue NT$10.519B, one aggressive broker rather than a consensus. Price-to-book about 4.48x.

Conviction is constructive on the franchise, neutral on the stock at NT$101, Medium overall. The operating story is real and largely proven in the numbers. Three things cap the risk-reward and are treated in full below. Valuation is full at roughly 31x forward on what is likely peak-cycle EPS. The stock trades on TPEx with a plus-or-minus 10% daily price limit and thin liquidity, a genuine execution risk for any real position. And the second-source rent is contestable by design, because the same mid-2027 that ends Nittobo's constraint also opens a capacity wave (Nittobo tripling, Nan Ya taking 20% of Nittobo's weaving, Taiwan Glass, Asahi Kasei and Shin-Etsu certifying). A better entry, or hard proof that Fulltech is climbing the grade ladder toward NER and M8, would move this to a buy.

One honesty note on provenance. The Fulltech story is a Collyer Bridge / @illyquid (Substack) idea sourced from Taiwanese market chatter, and STF Research holds Fulltech as its low-Dk demand-story pick. The originating analyst is markedly more cautious than the excitement around the name suggests, framing it as "less compelling than Macronix" and as a filler of the lower-end gap the market leaders vacate. SemiAnalysis has no direct Fulltech coverage, so there is no independent Western check on this specific name (flagged, see Sources). The glass-cloth shortage is well corroborated. Fulltech-specifically is the lower-conviction, lower-end leg of the trade, and the roughly 3x re-rate has already closed most of the cheap-entry window.

2. Corporate overview

Fulltech is a fiberglass yarn-and-cloth pure-play, founded January 1999 and headquartered in Douliu, Yunlin County, Taiwan. It does one thing and does it whole. It melts electronic-grade and industrial-grade glass into filament yarn, then weaves the electronic-grade yarn into thin glass cloth. The cloth is sold to CCL makers, who laminate it with resin and copper foil to build the dielectric core of printed circuit boards. The industrial-grade yarn goes into construction and general reinforcement. That the same company both melts the yarn and weaves the cloth is the whole point of the business, and the source of its margin edge.

Two product segments carry the revenue. In FY2024 fiberglass cloth was NT$2.02B, or 47.47% of revenue, with fiberglass yarn making up the balance of roughly 52.5%. Total FY2024 revenue was about NT$4.26B at a blended gross margin of 14.45%. The mix is shifting hard toward cloth, and toward the high end within cloth. FY2025 delivered net income of NT$857M and EPS of NT$1.64, and 2026E consensus revenue of NT$9.591B implies the mix and pricing shift roughly doubling the top line inside two years.

The business model is vertical integration from yarn to cloth. When low-Dk yarn is scarce and its price volatile, an integrated maker captures margin at both the melt and the weave stage and can flex allocation between them. A weave-only converter is hostage to yarn supply. This is why Fulltech's Q1 2026 gross margin of 35.97% led all five listed Taiwan fiberglass makers, ahead of Glotech (5475) at 28.32%, Chien Jung (5340) at 21.19%, Taiwan Glass (1802) at 19.23%, and the conglomerate-wide Nan Ya Plastics (1303) at 11.94%.

Geographically, yarn and cloth sell mainly into Taiwan, China and Korea, with further exports to Europe, the US and Japan. Manufacturing sits across three countries. In Taiwan, Fulltech runs its yarn and cloth plants around Douliu, Yunlin (yarn and cloth lines dating to 2007) plus a second yarn facility of about 45,000 tons per year that began production in 2010. In China it operates a weaving plant in Dongguan. The new leg is Thailand, where through wholly owned subsidiary Fulltech Fiber Glass (Thailand) Co., Ltd. it is building a cloth plant with a first phase of about NT$3.1B (roughly US$97M), funded by internal cash and bank borrowing, targeting mass production in Q3 2027. The Thai plant is designed around advanced-process cloth for low-earth-orbit satellites, AI servers and IC substrates, and doubles as geographic de-risking for Western customers pursuing de-China sourcing. There are no material joint ventures. Fulltech stands alone as an independent second source, which cuts both ways. Note that the Nittobo-Nan Ya weaving arrangement and the Nittobo-Chien Jung T-glass tie-up are competitors' partnerships, not Fulltech's.

3. First principles: the technology

A high-speed PCB is copper foil bonded to a dielectric made of woven glass cloth impregnated with resin. The glass is not inert filler. At 112G and 224G per-lane signaling, the electrical properties of the glass become part of the signal design, and two physical properties dominate.

The dielectric constant (Dk) sets how fast a signal travels and how much it leaks sideways. Propagation delay scales with the square root of the effective Dk of the composite, and the composite Dk is a blend of the resin (roughly 2.5 to 3.0) and the glass. Ordinary E-glass has a Dk of about 6.8 to 7.1, which drags the blend up, slows the signal and, worse, creates the fiber-weave effect. Because the glass bundles have a much higher Dk than the resin-filled gaps between them, a trace running over glass sees a different delay than one running over resin, producing timing skew that corrupts high-speed differential pairs. Lowering the glass Dk shrinks that contrast, cuts skew and speeds propagation. The dissipation factor (Df) sets how much signal energy the dielectric burns as heat. Insertion loss rises with frequency times Df, so at switch speeds moving from 800G to 1.6T even a small Df improvement buys meaningful reach. As a working threshold, GHz-class PCB glass targets a Dk at or below 4.9, a Df at or below 0.004, a softening point at or below 860 degrees C, and a coefficient of thermal expansion (CTE) at or below 5.0 ppm per degree C.

CTE is the second axis, and it is a mechanical-reliability property, not an electrical one. Large GPU and ASIC substrates and high-layer-count boards stack glass, resin and copper with mismatched expansion rates, so every reflow and thermal cycle warps the package. As substrate area grows, warpage scales with it and destroys yield. Low-CTE T-glass runs about 2.8 ppm per degree C against roughly 5.5 for E-glass, which is why it is the substrate material of choice for GPU and ASIC packages, a distinct requirement from the low-Dk needed on motherboards and switch backplanes.

Both improvements come from the same chemistry lever and carry the same penalty. Dk, Df and CTE all fall as you raise silica (SiO2) and boron oxide (B2O3) content and strip out the alkaline-earth and alkali oxides such as calcium oxide. NE-class low-Dk glass runs roughly 45 to 55% silica and 15 to 20% boron oxide. The more aggressive NEZ composition pushes to 50 to 55% silica and 25 to 30% boron oxide. The penalty is that a higher-silica, higher-boron, low-flux glass is far harder to melt and draw, demanding higher furnace temperatures, chewing through platinum-alloy bushings faster and running at lower throughput. That physics, not weaving, is the true bottleneck, and it is why each rung up the ladder costs a multiple of the last.

The grade ladder, with the metrics and economics that matter:

Grade Property Dk / Df / CTE anchor Application ASP anchor
E-glass General electronic Dk 6.8-7.1, Df ~0.006, CTE ~5.5 ppm/C Standard PCB 1x baseline
T-glass Low CTE CTE ~2.8 ppm/C GPU / ASIC substrates premium to E-glass
NE-glass Low Dk ~45-55% SiO2, 15-20% B2O3 AI-server motherboards, 400G switches ~6x E-glass
NER-glass Low Dk2 further reduced Dk/Df 800G switches ~2.5x NE-glass
NEZ-glass Low Dk3 50-55% SiO2, 25-30% B2O3 1.6T switches (2027) above NER
Q-glass Quartz ~99.9% SiO2 M9 substrates highest; brutal to drill and laminate

On absolute price, verified sources give multiples rather than clean per-kilogram figures. NE-glass at roughly 6x E-glass and NER at roughly 2.5x NE. Nittobo's list-price moves are the cleanest read on the direction of travel, up about 20% in August 2025 and a further 20 to 30% in April 2026. Absolute dollar-per-kilogram per grade is a gap, flagged rather than guessed.

4. Product and segment deep-dive

Fulltech's franchise straddles two product families and two grades, and precision here matters more than anywhere else in the write-up.

The low-Dk line. The qualified, volume product is FLD1, Fulltech's first-generation low-Dk cloth, which sits at the NE class feeding the M6/M7 CCL grades (M7 is the top of its current shipping range). This is the correct label. Fulltech is an NE / M7 low-Dk maker, not an M7-plus or M8 maker. Independent grade mapping lists it among the Low-Dk1 (NE) second sources alongside Asahi Kasei, Taiwan Glass, Taishan Fiberglass and Hong Ho, and specifically not among the Low-Dk3 / Q-glass names. Above it, Fulltech is ramping FLD2, a next-generation product the market reads as its move toward Low-Dk2 (NER class), which carries a materially higher ASP than FLD1 and whose shipments management expects to keep rising through the second half of 2026, lifting mix and margin. Two honesty flags on FLD2. First, company disclosure on FLD2 volume and share is thin, so it should be characterized as climbing toward NER, not as an established M8-plus incumbent. NER remains Nittobo's stronghold at roughly 60 to 70% share. Second, at the top of the ladder Fulltech only began M9 quartz-cloth certification around its 23 June 2026 shareholder meeting. The 24 June disclosure says the M9 quartz cloth is R&D-complete (研製成功) and that certification has just launched (已正式啟動), so M9 is optionality, not base case, and revenue from it is years out. It should not be described as M9-qualified. Glotech's (5475) Q-glass sits a full step above at M9, roughly two CCL grades higher than Fulltech's core book.

The low-CTE line. FLE is Fulltech's T-glass-class, low-CTE cloth for IC substrates. It passed customer certification and began small-volume shipment in Q4 2025 (通過客戶認證並進行小量出貨), and management guides sequential shipment growth through 2026 as order overflow from the constrained incumbent spills to qualified alternatives. Fulltech appears on the short list of qualified T-glass second sources alongside Taiwan Glass and Hong Ho.

ASPs. The company does not publish per-style prices, but the direction is clear and confirmed. FLD2 commands a significantly higher unit price than FLD1, the industry ladder puts NER at about 2.5x NE and NE at about 6x E-glass, and Fulltech announced a fiberglass-cloth price increase effective July 2026 of up to 30%, on top of the sector-wide hikes led by Nittobo. Customers are Taiwan's high-end CCL trio, detailed in section 5b. The single cleanest evidence that the mix and pricing are landing is the gross margin, 35.97% in Q1 2026 against 14.45% for full-year 2024, a swing of about 21 points driven by the shift into low-Dk high-speed cloth. On a like-for-like quarterly basis the improvement is real but narrower, roughly 6.24 points year on year and 6.96 points quarter on quarter, so the 21-point figure is a mix-and-cycle story against a depressed base, not a pure quarter-on-quarter step change.

One reconciliation the audit demands. Two similar-looking mix statistics circulate and measure different things. The company's own disclosure is a production-allocation figure. Low-Dk first-generation plus next-generation advanced-process products accounted for 45% of Q4 2025 capacity share (產能比重), with a guide above 60% in 2026 (verified; earlier chatter of "up to 70%" is unsourced and dropped). Separately, STF Research frames a shipment-output mix of roughly 40 to 60% of electronic-grade cloth output, up from about 30% in 2024. The trajectory is consistent, but they are not the same metric, and the STF figure is that firm's output-mix framing rather than a company-stated number. Fulltech's disclosed figure is a 產能比重, itself ambiguous between installed-capacity share and produced-output share. Treat the 45%-to-60%-plus as the company capacity-allocation guide and the 30%-to-40/60% as STF's derived shipment-mix framing. The precise reconciliation is not verifiable from public filings and is flagged.

5. Value chain position

The chain runs from batch minerals to hyperscaler, and the rent concentrates at one node. Upstream sit the raw inputs, silica sand and high-purity silica, a boron source, alumina and clay, platinum-rhodium bushings, and the energy to run melt furnaces. For low-Dk and low-CTE grades, high-purity silica and boron are the pinch points, because the low-flux, high-boron chemistry is what makes the glass electrically good and physically hard to melt. Fulltech melts its own yarn, so it owns that step rather than buying it, which is the crux of the moat.

The scarce capability in this chain is melting low-Dk and low-CTE glass yarn, not weaving it. Weaving is comparatively commoditized. The proof is that Nan Ya can weave 20% of Nittobo's specialty fabric using Nittobo's own NER yarn. The rent therefore accrues to whoever controls qualified low-Dk and low-CTE yarn. Fulltech is one of the few makers outside Japan that both melts the yarn and weaves the cloth, so it captures value twice, at the melt (the scarce step) and at the weave (the integration margin). That integrated capture is visible in the sector-leading 35.97% gross margin.

Midstream, Fulltech is the yarn-plus-cloth maker. Downstream, its cloth flows to CCL makers who add resin and copper foil to produce the laminate, then to PCB and substrate fabricators, then to server ODMs and the AI chip vendors pulling the whole chain. Value is captured by selling qualified low-Dk cloth at multiples of E-glass ASP and by integration margin, and Q1 2026 shows the capture working. The caution, developed in Part III, is that the rent is structurally contestable. A qualified second source exists precisely to relieve the incumbent, so the moment new capacity floods in from mid-2027, the same rent can be second-sourced away from the second source.

5b. Key customers and partners

Fulltech is qualified into Taiwan's high-end CCL trio, the makers that supply the Nvidia AI-server chain. Named correctly with tickers, these are Elite Material (台光電, EMC, 2383), ITEQ (聯茂, 6213) and Taiwan Union Technology (台燿, TUC, 6274). Coverage confirms Fulltech has earned certification from the major local CCL producers, and these three are the buyers of M7-and-above cloth for AI motherboards, ASIC and GPU substrates, and 800G-to-1.6T switch material. Elite Material in particular supplies the GB300, TPU v7 and Trainium3 chains with Fulltech as an upstream glass-cloth source. Note the disambiguation, because Chinese coverage repeatedly mistranslates it. 台光電 is Elite Material, a CCL maker and Fulltech customer, not Taiwan Glass (台玻, 1802), which is a fiberglass peer and competitor.

A customer-concentration caveat belongs here and is developed in Part V. Taiwanese forum chatter suggests Elite Material (2383) may reduce its Fulltech cloth allocation in 2027, which, if true, speaks directly to the by-design fragility of a second-source position. Treat it as unverified for now.

On partners, the important point is what Fulltech is not. It is a standalone second source to Nittobo, competing for the same qualifications, without an incumbent's supply agreement behind it. The visible partnerships in this space are competitors'. Nan Ya Plastics (1303) is Nittobo's weaving partner, contracted to weave 20% of Nittobo's specialty fabric by 2027 using Nittobo-supplied NER yarn, and Chien Jung (5340) has an actual Nittobo T-glass relationship whose low-CTE product began mass production in Q1 2026. Fulltech's independence is an asset when the incumbent is constrained and a liability when the incumbent's ecosystem expands.

6. Why it matters and TAM

The name expresses one theme, the AI-server copper-clad-laminate bottleneck. A PCB is copper foil on a woven-glass-cloth-plus-resin dielectric, and AI servers force a climb up the glass grade ladder from E-glass to T-glass (low-CTE) to NE and NER (low-Dk) to Q-glass (quartz, M9). Each step improves signal integrity and package reliability, and each step is harder to melt and slower to qualify, so the material that used to be background becomes a performance-critical, supply-constrained design variable.

The demand curve behind this is physical and compounding. GPU substrate area has grown from about 3,190 mm2 in the Hopper generation to roughly 4,780 mm2 in Blackwell and toward 8,000 mm2 in Rubin, about 2.5x, and every extra square millimeter of large substrate raises the low-CTE requirement. Motherboard layer counts are rising from roughly 20 to 28 in 2024 and 2025 toward 24 to 40 in 2026 and 2027, and per-lane signaling at 224G with switches moving from 800G to 1.6T pushes the low-Dk requirement up a grade at a time. More area, more layers and faster lanes all pull more high-grade glass cloth per server.

On formal TAM, the available market-report figure looks conservative and narrowly scoped against that narrative. One report sizes the low-Dk/Df electronic glass cloth market at about US$470M in 2025 rising to US$707M by 2032, a 6.1% CAGR. That growth rate is hard to reconcile with an incumbent raising prices roughly 20% and then 20 to 30% inside eight months, and a second source lifting cloth prices up to 30% in a single move, so the near-term pricing-and-volume story likely runs well ahead of the reported CAGR. The report is presented as a data point, not the governing estimate. Fulltech's own slice is the cleaner anchor, 2026E revenue of NT$9.591B (up about 58%) on consensus, or NT$10.519B on Mega's more bullish model, with high-end cloth above half of revenue.

6b. Sector inflection: why now

The catalyst is a dated supply wall. Nittobo holds roughly 90% of T-glass and 60 to 70% of NER, and it is capacity-constrained until its expansion lands. It responded to the shortage with price rather than volume, raising list prices about 20% in August 2025 and a further 20 to 30% in April 2026, and it is only now tripling T-glass capacity with a roughly 15 billion yen (about US$96M) investment at its Fukushima base, with new lines coming online in 2027. Until that capacity arrives around mid-2027, demand runs ahead of what the incumbent can ship, and the overflow, in both volume and price, routes to qualified second sources. That is the window Fulltech is monetizing right now, visible in its record Q1 2026, its sector-leading 35.97% gross margin, and its July 2026 price increase of up to 30%.

One caveat on the mechanism. The stronger "Nittobo is structurally exiting the lower end" reading that circulates in the Taiwanese chatter is an interpretation, not a Nittobo-stated exit, and it is unverified. What is independently confirmed is the mix-shift up-market plus the roughly 20% and 20-to-30% list-price hikes (TrendForce, DigiTimes). Fulltech's opportunity rests on the confirmed shortage and pricing, not on the softer exit narrative.

The supply-demand set-up is unusually clean for a materials name. The incumbent is capped for roughly a year, qualification is slow (typically six to twelve months or more, so the qualified set cannot expand overnight), and demand is compounding through substrate area, layer count and switch speed. Fulltech is qualified on the two grades that are actually short, NE-class low-Dk at M7 (FLD1) and T-glass low-CTE (FLE, certified Q4 2025).

The catalyst path from here is sequential and mostly near-term. FLD2 low-Dk2 shipments ramp through the second half of 2026, lifting mix and margin. The July cloth price increase of up to 30% flows into Q3 2026 revenue. FLE low-CTE volume grows sequentially into IC-substrate demand. M9 quartz certification, begun in June 2026, is genuine optionality but not the base case. And the Thailand plant reaches mass production in Q3 2027, adding advanced-process capacity and Western-friendly geography just as the incumbent's expansion arrives.

That last point is the tension, and it is where Part III picks up. The same mid-2027 date that ends Nittobo's constraint also opens the capacity wave, Nittobo tripling, Nan Ya weaving 20% of Nittobo's fabric, and Taiwan Glass, Asahi Kasei and Shin-Etsu pushing through certification. A second source that exists to relieve a shortage is, by construction, the first to feel the relief when the shortage ends. The window is real and Fulltech is executing inside it, but the clock that opened the window is the same clock that can close it.

PART II. MANAGEMENT AND GOVERNANCE

Governance is the weakest leg of the Fulltech thesis. The fiberglass story is clean, the stewardship is not. This is a family-controlled TPEx small cap whose chairman runs two listed companies at once, one of which is insolvency-adjacent, and whose most recent capital-return and capital-raising decisions favor tax efficiency and hand-picked outside subscribers over a straightforward return on the record profits the business is now printing. None of it breaks the thesis. All of it means you own Fulltech for the glass-cloth cycle, and you underwrite the governance rather than lean on it.

The two-hat chairman

Chang Yuan-pin (張元賓) chairs Fulltech and has since March 2010, when he took the seat from his father, founder and patriarch Chang Ping-chao (張平沼), the man behind the Jin Ding Group (金鼎集團) and Fulltech's former chairman. Chang holds an MBA in Finance from Washington University. He returned to Taiwan in 2007 to run Fulltech as general manager, moved to the affiliated bicycle maker IDEAL BIKE (愛地雅, 8933) as special assistant in 2008 and its general manager in 2009, then stepped up to the Fulltech chair in 2010. The board re-elected him unanimously in June 2023 for the ninth term.

The distinguishing fact is that Chang is, by his own account, the only person in Taiwan's listed universe serving as the full-time chief executive of two separate public companies simultaneously. He is chairman and de facto CEO of Fulltech and full-time general manager of IDEAL BIKE at the same time. Fulltech also lists co-general managers Chen Pi-cheng (陳壁程), Chang Chin-shuo (張晉碩) and Lin Sheng-tsun (林聖尊), a deputy general manager Chang Chih-yuan (張致源), finance head Lu Chi-fang (呂季芳) and accounting head Liu An-tsang (劉安倉). The bench exists, but ultimate direction sits with one person whose attention is split across a fiberglass boom and a bicycle-maker rescue.

Competence is not the issue. Chang inherited a marginal commodity yarn business and has steered it into the highest-margin position among Taiwan's five fiberglass makers. The issue is bandwidth and conflict, both of which route through IDEAL BIKE.

Ownership and skin in the game

Insider alignment is thin and getting thinner. Directors and supervisors held roughly 12.4% of Fulltech as of June 2026, down from about 13.8% earlier in the cycle. Foreign investors hold about 10.6%. The controlling family exercises influence not through direct personal stakes but through investment vehicles, chiefly Dah Tai Investment (達泰投資), which holds about 10.4% (roughly 60.7 million shares) and is represented on the board by the founder generation. Smaller family-aligned vehicles, Shang Ling Investment (商領投資) and Derlong Warehouse and Handling (德隆倉儲裝卸), hold sub-1% board stakes. The affiliated IDEAL BIKE (via its Aidia industrial entity) held about 2.6% of Fulltech at the last clear reading, a direct cross-holding.

The chairman's personal stake is small and heavily encumbered. Chang holds roughly 4.95 million shares, about 0.85% of the company, and has pledged 68.8% of that personal position. On an absolute basis the pledge is minor, but a two-thirds personal pledge ratio is a revealed-preference signal that the chairman is drawing personal liquidity against his Fulltech stock, and it introduces forced-selling risk on any sharp drawdown, on a stock already 20% off its high. Aggregate director and supervisor pledging is more moderate at about 8.8%.

Insider trading into the tripling from NT$34.6 to a NT$127 high was, to their credit, restrained rather than a distribution. The largest observed move was the chairman trimming about 8% of his personal lot count (roughly 5,392 to 4,951 lots) in January 2026, alongside a batch of thirteen managers moving shares into a Taishin bank restricted-stock trust the same month, and small odd-lot reductions by the general manager and Derlong. No wholesale insider dump. The real dilution to insider ownership did not come from selling. It came from the family declining to defend its stake in the December 2025 share issue, which is where the capital-allocation story turns.

The IDEAL BIKE cross-holding, the central red flag

Fulltech is the largest institutional shareholder of IDEAL BIKE (8933), a bicycle assembler chaired by the same Chang Yuan-pin. The last clearly sourced ownership figure is 21.52%, above the 20% threshold that forces equity-method accounting, meaning Fulltech books its share of IDEAL BIKE's results in its own income statement. That figure dates to 2019 and has almost certainly moved after IDEAL BIKE's serial restructurings, but Fulltech remains a major holder.

IDEAL BIKE is a chronically distressed business, and the distress is current, not historical. It posted a full-year loss of NT$1.52 billion in 2018, saw book value fall below NT$5 per share, and was placed on full-cash-settlement status in 2019, after which it ran a 30% capital reduction and a rescue rights issue. It raised capital again in 2024. Then in the first quarter of 2026 it repeated the pattern, a net loss of NT$98 million, revenue down 31% year on year, book value back below NT$5 at NT$4.94, and a return to full-cash-settlement status effective 20 May 2026, with another capital reduction and rights issue planned for later in the year.

This is a live conflict, not a legacy footnote, and it cuts three ways. First, an equity-method stake in a loss-making affiliate depresses Fulltech's reported net income, which means the fiberglass business is modestly better than the headline EPS implies, though the carrying value of the stake may already be written down and the offset capped. Second, and more seriously, IDEAL BIKE's 2026 rescue creates a direct channel for Fulltech's AI-boom cash to leak into a dying bicycle business. A 21%-plus shareholder chaired by the same person is the natural backstop for a rights issue, and Fulltech's participation or non-participation in the 2026 rescue is the single most important governance question minority holders should be asking. It is not disclosed in the material reviewed here. Third, the arrangement consumes the scarce commodity, the chairman's attention, at the exact moment the fiberglass ramp needs it most.

Fulltech's charter (Article 18) does require board approval for related-party transactions, for cross-investments and share disposals, and for endorsements and guarantees, so a formal control exists. Its value is limited by the fact that the same chairman sits on both sides of the table.

Capital allocation, the placement and the capital-reserve dividend

Two decisions across 2025 and 2026 define the capital-allocation record, and they connect.

In November and December 2025 Fulltech issued 60 million new common shares at NT$66, a raise reported around NT$600 million in par-capital terms and roughly 10% dilution, priced well below the prevailing market. Both family-controlled directors, Dah Tai Investment and Derlong, waived 100% of their subscription rights, with the stated reason recorded only as "financial planning." The chairman then placed the waived shares with hand-picked "specific persons," including a Hong Kong custody account held through Capital Securities (群益金鼎), the investment vehicles Jing Tai (景泰投資) and Lian Sheng (聯昇投資), and several named individuals. The stated use of proceeds was to repay bank borrowings and top up working capital, not to fund the low-Dk capacity expansion. Raising discounted equity from an undisclosed-relationship group of insiders' choosing, while the controlling family passes on cheap stock, is the textbook Taiwan minority-shareholder concern. It diluted alignment and handed allied parties cheap paper just ahead of the run.

Then in May 2026 the board proposed the 2025 distribution, NT$0.5 in cash plus NT$0.5 in stock. The stock dividend was capitalized from earnings, a genuine reinvestment of record profits. The cash dividend, about NT$292 million, was paid entirely from capital reserve rather than retained earnings. This is the quality-of-earnings flag, and it deserves to be read precisely. Across the seventeen years Fulltech has paid cash dividends, this is the first sourced from capital reserve rather than earnings, in a company with a spotty payout history and nine zero-dividend years. The timing invites the read that the freshly raised NT$66 placement premium, which flows to capital reserve, is being partially recycled back to shareholders as a "dividend," a return of capital dressed as a distribution rather than a return on operating cash.

The honest verdict is that this is a yellow flag, not a red one, and the framing matters. The most probable primary driver is tax efficiency. Capital-reserve distributions from share premium are treated in Taiwan as a return of capital that reduces cost basis rather than as taxable dividend income, and they sidestep the supplementary health-insurance levy, which benefits the controlling family most. That is inference, since the filing gives no rationale. Set against it, the company did reinvest its actual record earnings through the stock dividend and retention for the ramp, which is the correct call at this point in the cycle. What lowers the quality of the signal is the combination, a first-ever capital-reserve payout layered on a discounted insider placement, with no disclosed reasoning for either, in a company generating more than enough real cash to fund a conventional dividend.

Compensation and alignment

Hard compensation figures are where Taiwan small-cap disclosure limits visibility. Fulltech discloses aggregate director and supervisor remuneration and top-manager brackets, plus a remuneration-to-net-income ratio, only in the full annual report on MOPS, and those NT$ figures are not surfaced in the secondary sources reviewed here. What can be assessed is alignment by revealed preference rather than by pay slip, and there the read is cautious. The clearest alignment signal of the cycle was negative, the family declining to commit fresh cash to the discounted December 2025 placement, choosing dilution over investment even at NT$66. Equity alignment is modest, the chairman's personal stake is under 1% and mostly pledged, and the family's economic interest sits in holding vehicles rather than personal shares. The mild offset is that insiders did not distribute stock into the rally.

Board and governance

The ninth-term board seated 7 June 2023, with a three-year term expiring 6 June 2026, so a full re-election fell due at the 2026 annual meeting and the current roster may have just refreshed, which is a disclosure gap here. As documented, the board runs nine seats, six non-independent and three independent. The non-independent side is family and insider weighted, the chairman, the Dah Tai and Derlong corporate vehicles, and three professional directors (two lawyers and a computer-science professor), one of whom represents a family investment entity.

The three independent directors are a credible academic slate rather than industry operators. Hsueh Fu-ching (薛富井) is an accounting PhD, professor at National Taipei University and its former president, and convener of both the audit and remuneration committees. Nieh Chien-chung (聶建中) is a finance PhD and professor at Tamkang University. Chen Hou-ming (陳厚銘) is an international-business PhD and professor at National Taiwan University. Audit and remuneration committees exist and are staffed as required. The limitation is the familiar Taiwan template, credentialed academic independents whose oversight bandwidth is diluted across multiple boards, seated on a body the founding family controls, policing a chairman who sits on both sides of the group's most material related-party relationship. Formal governance is present, substantive independence is thinner than the box-ticking suggests.

One market-structure point compounds the governance risk at the execution layer. Between the family vehicles, the cross-holding, the placement to friendly specific persons, and roughly 10.6% foreign ownership, the genuine free float is constrained, and on the TPEx the shares carry a daily price limit of plus or minus 10% with thin small-cap liquidity. Concentrated ownership plus a hard daily limit means both entry and exit can gap, and a governance shock would be hard to trade out of.

Management DD verdict

Dimension Rating Basis
Leadership competence and experience Yellow Credentialed, long-tenured chairman (Washington University MBA, chair since 2010) who built the margin lead, but attention is split full-time across two listed companies, one in crisis.
Insider ownership and skin in the game Yellow to Red Directors and supervisors only about 12.4% and falling; chairman's personal stake under 1% with 68.8% of it pledged; family declined to defend its stake in the placement.
Related-party and cross-holding Red Roughly 21.5% equity-method stake in IDEAL BIKE (8933), a full-cash-settlement affiliate under the same chairman, with a 2026 rescue capital raise pending and Fulltech's participation undisclosed.
Capital allocation Yellow First-ever capital-reserve cash dividend layered on a discounted insider placement raises recycling-of-capital optics, though record earnings were genuinely reinvested; spotty dividend history (nine zero years).
Board and governance Yellow Three of nine credentialed academic independents, audit and remuneration committees and a related-party approval charter in place, but a family-controlled board with thin substantive independence.
Compensation and alignment Yellow NT$ figures not visible in secondary disclosure (annual report and MOPS required); revealed alignment weak, the family passed on cheap equity.
Insider trading behavior Green to Yellow No distribution into the tripling; largest move was the chairman trimming about 8% of a tiny personal stake and manager transfers into a bank trust.
Disclosure and transparency Yellow TPEx small-cap filings are terse, no rationale given for the capital-reserve dividend or the specific-persons placement, and current IDEAL BIKE exposure is not readily visible.

Overall management DD verdict, Yellow with a Red on the IDEAL BIKE cross-holding. The chairman is capable and the fiberglass execution has been excellent, but the governance package, a two-company chairman, a distressed related-party affiliate that could pull capital, a discounted placement to hand-picked insiders, a first-ever capital-reserve dividend, and a family that dilutes itself rather than invest, means the stewardship is something to underwrite rather than trust. Own the name for the low-Dk cycle, size it for the governance.

PART III. COMPETITIVE DYNAMICS

The glass-grade ladder, and where each maker sits

Every claim about who wins in AI-server glass cloth reduces to one question. Which rung of the glass ladder can you actually make at yield, and at which CCL grade does it qualify. The ladder runs E-glass, then T-glass (low-CTE), then the low-Dk family of NE and NER, then quartz (Q-glass). Each rung maps to a CCL grade and a cost multiple over commodity FR-4. NE-glass anchors M6 and M7 at roughly 3x to 9x FR-4 cost. NER-glass anchors M8 at roughly 10x to 15x. Q-glass anchors M9 at roughly 15x to 20x, because M9 requires a dissipation factor at or below 0.0007 and only quartz (Df near 0.0005) clears it, while the interim NEZ-glass (Df near 0.001) does not.

Fulltech (1815.TWO) sits on the NE rung. Its qualified flagship is NE-class low-Dk cloth at the M6 and M7 CCL grades, and it is a second source on T-glass (low-CTE), where Nittobo (3110.T) is otherwise close to a monopoly. It is not an M8-plus supplier. That tier is NER, and NER is Nittobo's stronghold. This distinction matters because Fulltech markets its qualified product as "low Dk2," which is its own shorthand for second-generation low-Dk and maps to NE. It does not correspond to TrendForce's taxonomy, in which "Low Dk2" means NER-glass. The two labels collide, and conflating them would overstate Fulltech's grade by a full rung. Read against the standard ladder, Fulltech is an NE and T-glass house that has only just begun M9 quartz certification (announced at its June 2026 annual meeting), which places any quartz revenue years out and behind the incumbents on that rung.

Glotech (5475.TWO) sits above, on quartz. Glotech is Taiwan's quartz-cloth pure-play, shipping quartz-fiber yarn samples to Japanese customers ahead of Nvidia's Rubin platform and targeting the M9 CCL grade directly. Fulltech and Glotech are therefore adjacent on the glass ladder but do not compete at the same CCL node. Fulltech is fighting for NE and T-glass share at M6, M7 and low-CTE. Glotech is fighting for Q-glass share at M9. In CCL-grade terms Glotech's target sits roughly two grades higher than Fulltech's core book, M9 against M7. Any framing that treats the two as head-to-head at the same spec is wrong.

Competitive landscape

Taiwan has five listed fiberglass makers of scale. Only two are electronic-grade pure-plays. The gross-margin spread across the five, all drawn from Q1 2026, is the single cleanest read on how concentrated each is in the high-value low-Dk mix versus commodity yarn and industrial cloth.

Maker Ticker Role in the AI-cloth chain Grade reach Position / share Moat source Pure-play? Q1'26 gross margin
Nittobo (Nitto Boseki) 3110.T Incumbent, sets the price T-glass, NER; NEZ due 2027, next-gen T-glass due 2028 ~90% T-glass; ~60-70% NER (TrendForce) 30-year formula IP, patent web spanning material to loom, 2-year kiln lead time No, glass-fiber-led but diversified n.a. (Japan filer)
Fulltech 1815.TWO Qualified second source NE (M6/M7) plus T-glass second source; M9 quartz cert only just started Taiwan's low-Dk pure-play, Nittobo second source Vertical yarn-to-cloth, qualification, richest mix Yes 35.97%
Glotech 5475.TWO Quartz leader among Taiwan makers Q-glass (M9), quartz yarn and cloth; low-CTE certified Taiwan's third-largest cloth maker; the local Q-glass name Quartz weaving know-how, Japanese-customer sampling for Rubin Yes 28.32%
Chien Jung (Baotek) 5340.TWO Cloth for CCL, climbing toward low-Dk Fiberglass fabric for CCL; low-Dk in progress; Nittobo T-glass tie-up Cloth-focused mid-tier Cloth weaving Yes (industrial materials) 21.19%
Taiwan Glass 1802.TW Diversified giant certifying into low-Dk Low-CTE, low-Dk, low-Dk2; entered Nvidia's chain Cited as the third company worldwide certified for low-Dk; NT$2.25B capex to take low-Dk lines from four to twelve Scale, capital, float-glass base No, container and float-glass conglomerate 19.23%
Nan Ya 1303.TW Formosa-group scale, Nittobo's partner Industrial and PCB cloth; NER via Nittobo raw material Among the world's largest cloth capacities; to handle 20% of Nittobo's specialty-cloth supply by 2027 Formosa integration, Nittobo alliance No, plastics and petrochemical conglomerate 11.94% (group-wide)

The margin gradient is the story. Fulltech's 35.97% is the highest of the five and roughly 8 points above Glotech, despite Glotech sitting on the higher quartz rung. The reason is mix concentration, not grade. Fulltech has poured its book into low-Dk high-speed cloth while carrying little commodity ballast, so cloth-side pricing flows through undiluted. The two conglomerates, Taiwan Glass at roughly 19% and Nan Ya at under 12% group-wide, show what happens when advanced cloth is a thin slice of a commodity-glass or petrochemical base. Their headline entry into Nvidia's chain is real, but the earnings signal is buried. Fulltech is the cleanest single-name expression of the low-Dk cloth trade among the five, with Glotech the cleaner expression of the quartz sub-trade one step up.

Moat analysis, and the business-quality three-test

Fulltech's moat is real but rented. It exists because Nittobo cannot physically supply the market and adds no meaningful capacity before mid-2027, and because CCL makers, having qualified Fulltech, will not casually re-qualify away from it inside a build cycle. Qualification is sticky and the incumbent is capacity-walled. That is a genuine, if narrow, position. It is not a durable franchise, and the three business-quality tests make the distinction concrete.

First, the five-year lock-up. Could an owner hold this business locked away for five years and be confident the economics survive intact. No. The rent is a function of one incumbent's capacity wall, and that wall is dated to mid-2027. The moat's half-life is the 18-to-24-month window between now and the 2027 capacity wave, not five years. Fulltech fails a clean lock-up test.

Second, the unique economic engine. Is the earnings engine proprietary and hard to copy. Only partially. Fulltech's engine is the highest-margin pure-play mix in Taiwan, and the swing from a 14.45% full-year 2024 gross margin to 35.97% in Q1 2026 shows real operating leverage on the mix shift into low-Dk high-speed cloth. But the engine is a qualified NE product two CCL grades below quartz, not a formula only Fulltech owns. The unique engine on this ladder belongs to Nittobo, whose NE and T-glass formulas took thirty years to commercialize. Fulltech has a good, replicable engine, which is a different and lesser thing.

Third, the blank-check disruptor. If a well-capitalized rival set out to take this position, could it. Yes, and several already are. The barrier is capital and time, not secret sauce. A glass-melting kiln runs 500M to 1.5B yuan and a production cycle exceeds two years, which is why the wall has held this long. But Asahi Kasei, Shin-Etsu, Taishan Fiberglass, Hong Ho and Feilihua are each mid-build on exactly the rungs Fulltech occupies or aspires to. A business that a blank check can enter within one kiln cycle does not pass this test.

The verdict follows. Fulltech is a high-quality cyclical beneficiary of a supply-demand dislocation, not a durable-moat compounder. The position is being rented from Nittobo's shortfall, and the lease expires on the 2027 capacity schedule.

Industry structure and cycle position

The CCL chain is mid-transition from commodity grades toward M8, M9 and M10, and the market has bifurcated. High-end low-Dk and quartz cloth is structurally short while commodity FR-4 faces structural oversupply and price deflation. On the high-end side, 2026 is the tight, allocation-driven phase of the cycle. Glass-cloth weaving equipment is booked through Q1 2028, with lead times stretched from roughly eight months to twenty-four, so no marginal supply can physically arrive before then. Apple, Nvidia, AMD, Microsoft, Amazon and Qualcomm executives have all visited Nittobo inside the past year, and Tier-1 CCL makers are bidding against each other for allocation. That auction dynamic is what underwrites second-source pricing power today, and it is why Fulltech's margin and Q1 2026 record earnings have arrived now. The Rubin-specific CCL market alone is modeled to expand from roughly US$275M in 2026 to US$2B in 2027, and AI-server boards have moved from about 15% of the total PCB market in 2025 to over 25% in 2026.

This is peak-rent, not early-cycle. The demand-side scarcity is genuine, but the phase, characterized by equipment booked out, lead times at their widest and customers touring the incumbent, is the one that historically precedes a supply response rather than the one that follows it.

The 2027 capacity wave is that response, and it is unusually well telegraphed. Nittobo is tripling T-glass capacity, backed by roughly 15 billion yen into Fukushima and more than 50 billion yen across Japan and Taiwan over 2026 and 2027, with new supply reaching the market from mid-2027 and a six-month yield ramp after that. Nan Ya will handle 20% of Nittobo's specialty-cloth supply by 2027 under their raw-material alliance, converting a rival into incremental incumbent capacity. Taiwan Glass is taking its low-Dk lines from four to twelve on NT$2.25B of capex. Nittobo's improved NEZ-glass is slated for 2027 and a next-generation T-glass for 2028. The wall does not crumble gradually. It is scheduled to break in the second half of 2027, and much of that supply lands on the exact NE and low-Dk rungs where Fulltech lives.

Emerging threats

The threat to Fulltech is not that its rung disappears. It is that its rung is the most crowded on the ladder, and it commoditizes first. The lower a grade sits, the more entrants can reach it, and NE at M6 and M7 is the entry point for every credible newcomer.

The named entrants targeting NE-class low-Dk (Low Dk1) are Asahi Kasei, Taiwan Glass, Taishan Fiberglass and Hong Ho, alongside Fulltech itself. The named entrants targeting quartz (Low Dk3) are Asahi Kasei, Shin-Etsu, Glotech, Feilihua, Taishan Fiberglass and Hong Ho. Asahi Kasei announced its formal entry into the AI-chip glass-cloth chain in April 2026, explicitly aimed at Nittobo's roughly 90% share. Shin-Etsu is pushing on quartz. Taishan Fiberglass shipped second-generation low-dielectric cloth in 2024 and is on both NE and quartz. Feilihua reached M9-grade quartz with Nvidia certification in 2025, with an integrated chain from quartz sand to finished cloth. Three Japanese makers, Nittobo, Asahi Kasei and AGC, already hold close to 70% of the global high-end electronic-cloth market between them, and the two challengers are now expanding rather than standing still.

Two threats cut most directly at Fulltech's book. First, the Nittobo and Nan Ya alliance gives Nan Ya NER raw material and 20% of the incumbent's specialty supply by 2027, which pressures Fulltech precisely on the low-Dk grades just above its NE core and consolidates volume behind the incumbent rather than dispersing it. Second, because Fulltech's core rung is NE rather than quartz, it is more exposed to the capacity wave than Glotech, whose quartz position sits on a less-crowded, harder-to-reach rung. Commoditization runs up the ladder from the bottom, and Fulltech is nearer the bottom of the AI-relevant grades than Glotech is.

That frames the bear case. The rent Fulltech collects exists because it is a sanctioned, qualified second source, and a second source is by construction second-sourceable. The same logic that qualified Fulltech will qualify Asahi Kasei, Taishan and Hong Ho on the same rung, and the 2027 capacity wave, Nittobo tripling, Nan Ya at 20%, Taiwan Glass moving to twelve lines and multiple challengers certifying, is a scheduled compression of exactly the scarcity that produced the record Q1 2026 margin. The competitive setup rewards owning the dislocation while it lasts, not underwriting it as a franchise.

PART IV. THE NUMBERS

All figures NT$ (TWD) unless noted; currency confirmed via yfinance. FY = calendar year (December). LTM = Q2'25 through Q1'26. Price basis NT$101 (yfinance last close, delayed; reconcile live in IBKR). yfinance TPEx coverage was better than expected here, with full FY2021-25 statements and five clean quarters; gaps (Q1'26 actuals, FY2026E, monthly revenue, 1H'24 quarterly EPS) were filled from Taiwan financial press.

Income statement (FY-2 / FY-1 / FY0 / LTM / FY+1E)

NT$ millions unless noted.

Line FY2023 FY2024 FY2025 LTM FY2026E (consensus)
Revenue 3,199.9 4,246.7 5,974.8 6,473.2 ~9,591
Gross profit (12.2) 613.8 1,646.9 1,912.6 n.a.
Gross margin % (0.4)% 14.5% 27.6% 29.5% mgmt guiding higher
Operating income (522.7) 90.2 932.5 1,062.3 n.a.
Operating margin % (16.3)% 2.1% 15.6% 16.4% n.a.
Net income (651.2) 62.9 856.5 1,143.2 ~1,911 (implied)
EPS (NT$) (1.42) 0.13 1.64 1.96 3.27
Operating cash flow (130.5) 264.2 694.7 n.a. n.a.
Capex (421.7) (1,276.1) (2,403.7) n.a. n.a.
Free cash flow (552.3) (1,011.9) (1,709.0) n.a. n.a.
Capex / revenue 13% 30% 40% n.a. n.a.

FY2026E base is the CMoney 法人 consensus, revenue NT$9,591M (up about 58% per the consensus; that implies a FY2025 base near NT$6.05B, so on the yfinance FY2025 actual of NT$5,975M the growth is closer to 60%), EPS NT$3.27, implied net income NT$1,911M on 584.3M shares. Mega Securities (兆豐) bull case: revenue NT$10,519M, EPS NT$4.55, target NT$206. Treat Mega as one aggressive broker, not a consensus mean; there is no multi-analyst consensus for a TPEx micro-cap. Statements are yfinance (Yahoo/TPEx); FY2023 EPS is a calc (net income divided by about 458M shares); reported FY2024 EPS 0.13 and FY2025 EPS 1.64 confirmed via winvest and 財報狗. The share base rose across the period, 462.9M (2023) to 522.0M (2024) to 580.9M (2025) to 584.3M current (stock dividends plus the December 2025 raise), which makes per-share history noisy.

Quarterly trend and incremental margins

NT$ millions. Clean quarterly gross profit and EBIT are only available in yfinance from Q4'24 onward; the 1H'24 quarters (reported EPS -0.13 / -0.01 / -0.03, essentially breakeven-to-loss) lack a reliable quarterly gross-profit split on TPEx, so the incremental series starts Q4'24. Incremental margin = change in gross profit divided by change in revenue (and change in operating income divided by change in revenue), quarter on quarter.

Quarter Revenue Delta Rev Gross profit GM% Incr. GM% Op. income OM% Incr. OM% Net income EPS
Q4'24 1,286.9 n.a. 300.7 23.4% n.a. 160.8 12.5% n.a. 139.5 0.30
Q1'25 1,382.6 +95.7 411.0 29.7% 115% 272.2 19.7% 116% 171.3 0.33
Q2'25 1,422.4 +39.8 349.1 24.5% neg 202.0 14.2% neg (33.6) (0.06)
Q3'25 1,532.7 +110.2 411.8 26.9% 57% 258.9 16.9% 52% 282.9 0.53
Q4'25 1,637.1 +104.4 475.0 29.0% 60% 199.4 12.2% neg 435.9 ~0.83
Q1'26 1,881.0 +243.9 676.7 36.0% 83% 402.0 21.4% 83% 458.0 0.79

Reading the noise. Two distortions matter. First, Q2'25 net income went negative despite positive operating income, a roughly NT$200M non-operating hit (yfinance "Other Non-Operating," almost certainly FX on TWD strength). Q4'25 ran the other way with about NT$361M of non-operating gain, inflating net income above operating income. Non-operating and FX swings are large relative to core profit, so net income is a poor read on the business quarter to quarter. Use gross and operating margin. Second, Q4'25 opex spiked (SG&A plus R&D about NT$276M against a NT$140-150M run-rate), crushing that quarter's incremental operating margin, which looks like a year-end true-up. The clean signal is Q1'26, a 36.0% gross margin, a 21.4% operating margin and 83% incremental margins as advanced-process mix rises. Monthly revenue cross-checks (winvest, 財報狗): 2026 Jan 571, Mar 682, Apr 734, May 708; Jan-May cumulative 3,324, up about 42% year on year, up about 50% in May, tying to the Q1'26 print of 1,881.

Valuation

Price NT$101.0; 584.286M shares.

Metric Value Basis / source
Price NT$101.0 yfinance (last close, delayed)
52-wk range NT$33.85 – 129.5 yfinance
Market cap NT$59.0bn ~US$1.85B
Net debt ~NT$2.2-2.6bn FY2025 BS 2.24bn; MRQ (Q1'26) ~2.6bn (total debt 5.0bn less cash 2.37bn); not independently verified
Enterprise value ~NT$61.6bn market cap plus net debt; approximate, flagged
P/E trailing (FY25 EPS 1.64) 61.6x yfinance / calc
P/E LTM (EPS 1.96) 51.5x calc
P/E FY2026E consensus (EPS 3.27) 30.9x calc on CMoney 法人 consensus
P/E FY2026E Mega bull (EPS 4.55) 22.2x calc on single-broker EPS
EV/EBITDA (trailing) ~29-30x LTM EBITDA ~2.03-2.15bn
EV/Revenue (LTM) 9.5x calc
P/B 4.48x BVPS NT$22.56
Dividend yield ~0.5% 2025 declared NT$1.0 (0.5 cash + 0.5 stock); winvest

The valuation gap is the whole debate, 61.6x trailing against roughly 31x on the consensus 2026E and about 22x on Mega's bull EPS. The NT$206 Mega target embeds FY2026 EPS of NT$4.55, close to 2.8x FY2025, plus continued margin and mix expansion. It prices 2026 as a peak-growth year, and the same note flags an explicit risk of a post-2027 slowdown. On the venue, TPEx illiquidity and the plus-or-minus 10% daily band mean the multiple can gap in either direction and trap a position.

The Core Four read

Organic growth is inflecting hard on a single driver. Revenue ran NT$3.20B (2023, the post-glut trough) to 4.25B (2024) to 5.97B (2025, up 41%) to LTM 6.47B, with 2026 monthly prints at plus 40 to 50% year on year (May plus 50%). The bull case is a near-doubling to about NT$10.5B in 2026, or NT$9.6B on consensus. This is one theme, low-CTE and low-Dk glass fabric for high-layer-count AI-server PCBs, with management targeting above 60% advanced-process capacity mix in 2026. High-quality demand, but concentrated and cyclical, and the same Mega note flags 2026 as the likely peak. All organic, no M&A.

Margins show real operating leverage but are volatile below the line. Gross margin ran -0.4% (2023) to 14.5% to 27.6% to 36.0% in Q1'26; operating margin -16.3% to 2.1% to 15.6% to 21.4%. Incremental gross margin of about 83% in the latest quarter confirms genuine mix and utilization leverage in a fixed-cost furnace business. The caveat, again, is that net income is whipped around by large FX and non-operating swings, so judge the business on gross and operating margin, not reported EPS.

Capital intensity is very high, and it is the key financial risk. This is a furnace and capacity business spending like one, capex NT$422M (2023) to 1,276M (2024) to 2,404M (2025), about 40% of 2025 revenue. Free cash flow has been deeply negative three straight years (-552 / -1,012 / -1,709), because operating cash flow (695M in 2025) covers less than a third of capex. Depreciation, about NT$928M a year and rising, is a growing fixed cost, and margins hold only if the AI-glass volumes and pricing that justify the new furnaces actually show up.

Capital deployment has been equity-funded expansion with deleveraging. Despite the capex blowout, net debt fell from about NT$4.5B to NT$2.24B in 2025 because they funded growth with equity. Shares rose from 462.9M to 580.9M (stock dividends plus the raise) and book equity jumped from NT$7.75B to NT$12.60B. So existing holders are being diluted to build capacity, but the balance sheet came out stronger. Dividends are token (NT$1.0 for 2025, half in stock, about 0.5% cash yield), which is a reinvest-everything story, appropriately so given the demand, but it leaves shareholders exposed if the 2026 peak-cycle bet does not hold. Note the quality flag from Part II: the NT$0.5 cash portion was paid from capital reserve, not retained earnings.

The softest input in this whole section is anything tagged FY2026E or the NT$206 target. It rests on a single brokerage and should be read as one aggressive model, not consensus, the CMoney 法人 figure (9.591B / 3.27) being the base and Mega (10.519B / 4.55) the bull.

PART V. THE DECISION

The thesis is a timing trade dressed as a growth story. The demand curve is real and independently corroborated (TrendForce, DigiTimes, and the industry-wide pricing tape), but Fulltech's edge is a rent, not a moat. It earns outsized margins because Nittobo (3110.T) cannot supply and because a handful of second sources were deliberately qualified to relieve a single-supplier chokepoint. Everything below turns on how long that window stays open and whether NT$101 already pays for it.

Growth drivers and catalysts

Near-term, over zero to twelve months, the engine is mix, not volume. Low-Dk high-speed cloth has climbed toward the high end of Fulltech's electronic-grade output, and each point of that shift lifts the blended gross margin that already reached 35.97% in Q1'26 against 14.45% for FY2024. That reprices as record quarterly EPS while Nittobo stays sold out, letting Fulltech take price and volume at the same time and drop the ASP resets through at high incremental margin. The near-term tape is the monthly revenue line (May 2026 at NT$708M, up 49.9% year over year) building toward the NT$9.591B (up about 58%) that the 法人 consensus models for 2026, with EPS of NT$3.27; Mega Securities carries the bull case at NT$10.519B and NT$4.55. The discrete catalysts are new CCL and PCB customer qualifications on T-glass and NE-cloth, each of which converts spot allocation into design-in volume, plus any capacity-utilization disclosure that confirms the mix trajectory.

Medium-term, over one to three years, the demand driver is the Nittobo capacity wall. The incumbent holds roughly 90% of T-glass and 60 to 70% of NER (TrendForce) and adds no major low-Dk capacity before its Fukushima 3x T-glass line, which comes online in late 2026 and reaches full impact only in 2028. Through 2026 and into 2027 the incumbent structurally cannot meet the AI-server demand curve, so qualified second sources absorb the overflow. Fulltech's own lever is the grade ladder. Its low-Dk product is NE-class cloth at the M6/M7 CCL rung, and it is also a T-glass (low-CTE) second source to Nittobo. It sits below Nittobo's NER (M8+) and roughly two CCL grades below Glotech's (5475) Q-glass at M9, so Fulltech and Glotech are adjacent on the ladder rather than direct competitors. The M7/NE tier is where the bulk of current AI-server CCL volume actually is, which is why the ceiling is real but not yet binding. Fulltech's furnace and loom additions to lift electronic-grade cloth output are the self-help layer on top of the incumbent's shortfall.

Risks

Risk Likelihood Mitigant Can it close?
Second-source-by-design rent. The rent exists because customers and Nittobo qualified a second source on purpose, so the same logic keeps adding third and fourth sources (Taiwan Glass 1802, Nan Ya 1303, Taishan, Hong Ho). High over 2 to 3 years First-mover qualification lead, the highest gross margin of Taiwan's five fiberglass makers (a yield and cost edge), and design-in stickiness once a CCL house locks a laminate recipe No. Structural to the thesis. The rent compresses as sources multiply; this is a cyclical trade, not a durable moat
2027 capacity wave and oversupply. Nittobo tripling T-glass, Nan Ya reaching 20% of Nittobo's specialty-fabric output by 2027, and Taiwan Glass, Asahi Kasei and Shin-Etsu certifying. High, and largely scheduled AI-server unit growth may absorb the new supply; grade migration keeps the top rungs tight even as M7 loosens No. The supply is already committed. The only open question is whether demand outruns it
TPEx liquidity and the plus-or-minus 10% daily price limit. Thin float and a hard daily band make exit in a drawdown difficult, and a bad tape can lock limit-down across sessions. Certain (venue structure) Small position, staged entries and exits, limit orders only No. Structural to the market. Managed only through sizing, not stops
Customer concentration. The CCL and PCB customer base is narrow, so a few laminate houses drive the order book, and forum chatter flags a possible Elite Material (2383) allocation cut in 2027. Medium Broadening qualifications across CCL makers dilutes single-customer dependence Partly. Improves as the qualification base widens, but the current base is concentrated
FX (TWD). Reporting is in TWD against USD-linked AI demand, so TWD strength compresses margin, and for a USD investor TWD depreciation cuts the realized return. Medium and ongoing Partial natural hedge from USD-linked sales against TWD cost, plus current pricing power No (macro), but hedgeable at the portfolio level
Governance and the IDEAL BIKE (8933) cross-holding. A roughly 21.5% equity-method stake in a full-cash-settlement affiliate under the same chairman, with a 2026 rescue raise pending and Fulltech's participation undisclosed; plus a first-ever capital-reserve dividend and a discounted insider placement. Medium Charter requires board approval for related-party transactions; no wholesale insider selling into the rally Partly. Closes only with disclosure that Fulltech is not backstopping the IDEAL BIKE rescue and a return to earnings-funded dividends

Bear case and downside

The bear case is the thesis read backward. The rent was engineered, not earned, so the industrial logic that made Fulltech valuable in 2025 and 2026 is the same logic that dismantles the rent. When Nittobo's Fukushima 3x lands from late 2026 into 2028, and Nan Ya (1303), weaving Nittobo NER yarn, reaches its 20%-of-Nittobo target in 2027, the shortage that let Fulltech take price and volume together inverts into an ASP fight among Taiwan Glass (1802), Taishan, Hong Ho and Fulltech at the same M7/NE rung. Margin normalizes off the 35.97% peak, and the mix gains that carried gross margin from a 14.45% base flatten.

The valuation is the accelerant. The market capitalizes 2026 near-peak EPS at roughly 31x forward, and Mega's NT$206 target implies about 45x on NT$4.55. A growth multiple on a cyclical peak is a double de-rating risk, because both the earnings and the multiple fall together. The stock has already tripled from about NT$34.6 in July 2025 to NT$127 in April 2026 and sits near NT$101, roughly 20% off the high and flat-to-down since the Q1 print, so much of the good news is in the tape. At 4.48x book, there is little asset value to cushion a de-rate.

The downside math is a scenario, not a forecast, and rests on an assumption of a mid-cycle multiple and a normalized EPS below the current peak. A de-rate to 15x-18x, a reasonable mid-cycle multiple for a Taiwan materials cyclical, on a normalized NT$2.50 to NT$3.00 of EPS, implies roughly NT$40 to NT$55, a drawdown of 45% to 60% from NT$101. In a hard oversupply that coincides with any AI-capex pause, the pre-run base near NT$35 to NT$40 is the floor reference. Sizing has to survive that path, because TPEx illiquidity means you cannot count on selling into it.

Ownership and analyst sentiment

The register is foreign and retail momentum, not domestic-institution anchored. Foreign investors (外資) held 11.71% of the shares as of 30 June 2026, domestic investment trusts (投信) just 0.60%, and dealers (自營商) 2.07%, so total institutional ownership sits near 14% and skews heavily foreign. Taiwan's own long-only funds are essentially absent, which is unusual for a domestic pure-play and tells you this is not an institution-sponsored compounder in the local market. The foreign flow is also churny rather than sticky, with violent two-way days (about plus 10,300 lots on 22 June, roughly minus 9,700 on 26 June, plus 6,400 on 29 June). That is fast money, and it amplifies the liquidity risk above.

Coverage is domestic and thin. Mega Securities (兆豐) rates it a buy with a NT$206 target on 2026E revenue of NT$10.519B and EPS of NT$4.55, while the CMoney 法人 consensus sits lower at NT$9.591B and NT$3.27. There is no meaningful Western sell-side coverage, and SemiAnalysis has no direct Fulltech coverage at all.

Read the thin Western coverage as the edge, not the flag, but with eyes open. This is a roughly US$1.85B TPEx name that global funds structurally will not or cannot hold, given the venue, the liquidity, the 10% band and Chinese-language disclosure. The AI-CCL glass-cloth bottleneck is a Western-research obsession, yet the cleanest Taiwan expression of it trades on 法人 notes that most of that audience never reads, and STF Research holds Fulltech as its low-Dk demand-story pick. The information asymmetry is the opportunity. The other side of that coin is that the originating analyst (Collyer Bridge / @illyquid) rates Fulltech below Macronix and as a lower-end-gap filler, no Western coverage means no Western marginal buyer to force a further re-rating, and there is no institutional floor if the name breaks.

Position sizing and risk management

Treat this as high conviction on a low-durability edge. The demand story is verifiable and the near-term earnings power is genuine, but the moat is not, so this belongs in a book as a cyclical timing trade, not a core compounder. That argues for a smaller position than a structural-moat name would earn, given the by-design rent, TPEx illiquidity, full valuation and the governance package stacked together.

On entry, buying about 20% below the high after a triple is neither the bottom nor egregious, but it is not a buy here. Prefer scaling in on weakness to chasing strength, and let the 10% daily band work for you, since a bad tape can hand you a better entry quickly. Stage in tranches, use limit orders, never send a market order into a thin book, and avoid adding into vertical up-days. The bear math points to fair value well below spot on normalized EPS, so a meaningful de-rate toward the NT$40 to NT$55 mid-cycle range, or hard evidence that Fulltech is climbing to NER and M8 (which would re-underwrite a higher multiple), is what turns this from a pass into a buy.

Because you cannot rely on liquidity to exit, the invalidation has to be defined before entry and the position kept small enough that a sequence of limit-down gaps is survivable. Hard stops do not execute in a lock-limit market, so the position size is the stop. The triggers that flip the thesis are the Nittobo Fukushima ramp or Nan Ya's 20% target arriving ahead of schedule, a rollover in low-Dk cloth ASPs or a second consecutive quarter of gross-margin compression, new M7/NE qualifications by Taiwan Glass (1802) or Asahi Kasei that dilute the rent, foreign holding rolling over hard in a thin float, the dividend staying capital-reserve-funded rather than earnings-funded, disclosure that Fulltech is backstopping the IDEAL BIKE rescue, and any re-rating above roughly 40x forward without an accompanying EPS upgrade.

Decision log

2026-07-01, Inaugural canonical page (deep-dive, Section A). Verdict: PASS / NEUTRAL at NT$101, Medium conviction on the franchise, cyclical timing trade only. New page created from the sector work in advanced-packaging; no prior canonical 1815 page existed.

  • The call. Constructive on the business, neutral on the stock at NT$101 (about US$1.85B, roughly 20% off the NT$127 April high). This is a rented moat, not a franchise, so it earns a smaller-than-normal slot and only as a timing trade. Not a buy at spot. Accumulate on a de-rate toward the NT$40-55 mid-cycle range (15x-18x on normalized NT$2.50-3.00 EPS), or on hard proof Fulltech is climbing to NER/M8, which would re-underwrite a higher multiple. Pre-run base NT$35-40 is the deep-value floor reference.
  • Why not a buy now. Valuation is full at ~61.6x trailing / ~31x forward consensus (EPS 3.27) on likely peak-cycle EPS; Mega's (兆豐) NT$206 bull target is ~45x on EPS 4.55 and one aggressive broker, not consensus. Double de-rating risk if the 2027 capacity wave lands as scheduled.
  • Grade precision (load-bearing). Fulltech is NE-class low-Dk at M6/M7 (FLD1), plus a qualified T-glass (low-CTE) second source (FLE, certified Q4'25). It is NOT M8-plus; NER/M8 is Nittobo's stronghold (~60-70% share). Glotech (5475) Q-glass sits at M9, roughly two CCL grades above; the two are adjacent on the ladder, not head-to-head. FLD2 is climbing toward NER (thin disclosure). M9 quartz is R&D-complete with certification only just launched (24 Jun 2026), so it is optionality, years from revenue, not "M9-qualified."
  • What is proven vs interpreted. Confirmed: Q1'26 GM 35.97% (highest of Taiwan's five makers), 2026E consensus 9.591B/3.27, 45% to 60%+ advanced-process capacity guide, second-source qualification on NE and T-glass, Nittobo ~90% T-glass / ~60-70% NER, and the ~20% (Aug 2025) and 20-30% (Apr 2026) Nittobo list-price hikes. Softened: the "Nittobo structural exit" reading is an unverified Taiwanese-chatter interpretation, not a stated exit; carry the mix-shift-plus-price-hikes framing only. Dropped: the unsourced "up to 70%" capacity claim. Flagged as STF framing, not company-stated: the "~40-60% output mix, up from ~30% in 2024."
  • Provenance and sentiment. The idea is a Collyer Bridge / @illyquid Substack call sourced from Taiwanese market chatter; the originating analyst rates it below Macronix and as a lower-end-gap filler, i.e. the lower-conviction leg of the glass-cloth trade. SemiAnalysis has no direct coverage (SA-mirror cross-check: absence flagged; the "Fulltech" text in the SA mirror is misfiled illyquid material, not SA coverage, and should be cleaned up). Ownership skews foreign and churny (外資 11.71%, 投信 0.60% as of 30 Jun 2026); no Western marginal buyer and no institutional floor.
  • Governance is the weakest leg (underwrite, don't trust). Two-company chairman; roughly 21.5% equity-method stake in the distressed full-cash-settlement affiliate IDEAL BIKE (8933) with a 2026 rescue raise pending and Fulltech's participation undisclosed; first-ever capital-reserve cash dividend; discounted (NT$66) December 2025 placement to hand-picked "specific persons" that the family declined to subscribe. DD verdict Yellow with a Red on the cross-holding.
  • Execution note. TPEx plus-or-minus 10% daily limit and thin liquidity mean entry and exit can gap; size is the stop. Price/EV reconciled to NT$101 across the page (an earlier financials pull used NT$111); EV ~NT$61.6B is approximate because net debt is not independently verified and the Thailand build is partly bank-funded. Reconcile live in IBKR before acting.

Net stance: right theme, real cycle, rented moat, full price. Own the dislocation small if you must, but the better trade is patience for a de-rate or proof of a rung up the ladder.

Sources

Sector page (industry-wide context, link up): advanced-packaging, the glass grade ladder, the CCL chain, the Nittobo capacity wall, and the full player map. Fulltech is the Layer-2 glass-cloth pure-play pick there.

Financial data (reconcile live in IBKR):

Estimates and targets (single-broker / 法人, not multi-analyst consensus):

Technology / grade / catalyst:

Thesis provenance (calibrate conviction to the source):

  • Collyer Bridge / @illyquid (Substack) APAC Wraps, KB/raw/substack-archive/illyquid/2026-05-02-apac-wrap-17-april-2026-900.md (introduces Fulltech), 2026-05-09 and 2026-06-24 (downgrades conviction: "less compelling than Macronix," lower-end-gap filler), 2026-06-24-apac-wrap-22-june-2026.md (corroborates the broader glass-cloth squeeze). Sourced from Taiwanese market chatter (@QQ_Timmy, @LinQingV, Macro_Lin).
  • STF Research, holds Fulltech as its low-Dk demand-story pick.

SA-mirror cross-check (absence flagged): SemiAnalysis has NO direct Fulltech coverage. The "1815" hits in KB/wiki/semianalysis/ are false positives (image pixel dimensions), and the one text match, KB/wiki/semianalysis/2024/groq-inference-tokenomics-speed-but.md (lines ~119-143), is misfiled illyquid content sitting inside a 2024 SemiAnalysis Groq article, not genuine SA coverage. Recommend vault cleanup; do not cite it as an SA source.

from the vault · open in obsidian ↗