報告_UBS_大立光3008_20261012

PDF 原檔：報告_UBS_大立光3008_20261012_original.pdf

圖片清單（已驗證 2026-10-13）

原始內容

Largan Precision

Core weakness outweighs progress in AI optics

CPO progress is more tangible but substantial success is already priced in

We maintain Sell and our NT$5, target price as erosion in the core smartphone business outweighs CPO progress. The Q326 call confirmed that Largan's fiber array has passed customer qualification although its customer qualified another vendor's prism micro-lens assembly (PMLA) that Largan would now buy in. If Largan qualifies the full FAU plus carrier assembly for 2H27, content rises but with more pass-through cost and less margin upside. We roll our valuation to 2 28E, the first full year of CPO production, though 25% below the NT$6,66 close on October 8.

Full FAU assembly ould lift content, but ith more pass-through

We believe Largan's FA customer is one of two module suppliers on NVIDIA's 2H27 program, which we estimate could reach 4 % share. The program is still subject to successful yield ramp and could move to metalens in 2 28, which would reopen supplier qualification. Assembling the full FAU could lift Largan's ASP to US$125-15 leveraging its precision alignment and automation but the PMLA, V-groove and carrier parts would be pass through that cuts into margin leverage. We expand our FAU TAM analysis (Figure 9Sum ary o fFAUd ives :1 2 .3 n /5 6 b c 8 -) with FAU market in 2 28/3 expanding to 12.3mn/35.6mn units, with Largan base case set at 25% of the TAM.

Core smartphone business stays soft into 2027

Q326 sales of NT$15.7bn, +15% QoQ/-11% YoY missedUBSe/Street +36%/+33% on weaker flagship smartphone demand. GMs fell 7.2ppts QoQ to 42.2%, lowest in 13 years on renewable energy costs and low variable-aperture yields, dropping OpM to 17 year lows at 3 .3% and EPS to NT$39.6, below UBS/street NT$51.5 /NT$49.69. We cut Q4 to NT$16.7bn +7% QoQ, 11% below consensus as its customer trimmed orders after a pricier iPhone 18 launch and possible lag to Genius which is picking up periscope share and outgrew in Q3 (+1% YoY vs. -11% YoY). We cut 2 27 from NT$77bn to NT $67bn, before a recovery to NT$79bn in 2 28 with the next periscope upgrade. We see GM recovering to 47.5%/48.9% in 2 27-28 as yields improve and energy costs pass through in 2 28 pricing.

Maintain Sell and our NT$5,000 target

We cut 2 26E/27E EPS by 14%/16% to NT$16 /NT$18 and introduce 2 28E EPS of NT$22 for the core smartphone business. We value the lens business at NT$3,3 (15x 2 28E ex-CPO EPS of NT$22 , the middle of Largan's 13-18x historical range) and CPO option at NT$1,7 or US$7bn (49x our base-case 2 28E CPO EPS of NT$35, near the top of optical peers' 14-54x). Our base case gives Largan 25% of a 12.3mn-unit, US $1.85bn FAU market at US$15 per unit and a 5 % gross margin. That credits it with yields and margins it still needs to prove in a quarter when yields and energy costs cut 7.2ppts from its lens margin. At ~NT$6,66 , the shares imply ~NT$69 of 2 28E CPO EPS, about twice our base case.

Source: Company accounts, LSEG Eikon, UBS estimates. Metrics marked as (UBS) have had analyst adjustments applied. Valuations: based on an average share price that year, (E): based on a share price of NT$ 6,66 . on 12-Oct-2 26 6:29:19 CST

Equities

Taiwan

Electric Components & Equipment

12-month rating

12m price target Sell NT$5,000.00

Price (08 Oct 2026)

NT$6,660

RIC: 3 8.TWBBG: 3 8 TT

Trading data and key metrics

52- k range NT$7,81 . -2, 25.

Market cap.

NT$875b/US$27.4b

Shares o/s

131m (ORD)

Free float

7 %

Avg. daily volume ('000)

2,379

Avg. daily value (m)

NT$13,134.2

Common s/h equity (12/26E)

NT$199b

P/BV (12/26E)

4.4x

Net debt to EBITDA (12/26E)

NM

Randy Abrams

Analyst randy.abrams@ubs.com +886-2-8722 7338

Irene Chang

Associate Analyst irene.chang@ubs.com +886-2-8722 728

Core smartphone lens seeing eakness ith declining margins

Q326 sales was reported light at NT$15.7bn, +15% QoQ missing UBSe/Street +36%/ +33% amid weaker flagship smartphone demand. GMs fell 72 bp to 42.2%, lowest in 13 years on renewable energy costs and low variable-aperture yields, dropping OpM to 17 year lows at 3 .3% and EPS to NT$39.6, below UBS/street NT$51.5 /NT$49.69. Renewable-energy costs are structural under Largan's long-term power commitments and rise with Apple's share of the mix; with 2 27 pricing set, pass-through is more likely in 2 28.

We cut Q4 to NT$16.7bn +7% QoQ, 11% below consensus as its customer trimmed orders after a pricier iPhone 18 launch and possible lag to Genius which is picking up periscope share and outgrew in Q3 (flat vs. -11% YoY). October revenue should be flat MoM and November lower as the key customer has shortened its forecast after cutting flagship orders in September. The spring-2 27 model starts shipping in October but will not offset the flagship decline.

Figure 1: Genius outpacing Largan in recent months

Source: Company data

Figure 2: Largan margins dropping in recent years

Source: Company data, UBS estimates

We cut 2 26E/27E revenue 9%/13% to NT$61.6bn/NT$67.4bn and EPS 14%/16% to NT$163/NT$18 , and introduce 2 28E revenue of NT$79.5bn, +18% YoY for the new smartphone lens upgrade, taking EPS to NT$22 . Our estimates exclude CPO, which we value separately under base, bull and bear scenarios. Including our base case, 2 8E EPS would be NT$255, 6% above the NT$24 consensus which likely includes some CPO. We see GM recovering to 47.5%/48.9% in 2 27-28 as yields improve and energy costs pass-through in 2 28 pricing. With a still a range of outcomes on 2 28 CPO, we keep it in our valuation as a scenario for our upside/downside/base cases.

Figure 3: UBS vs. VA Consensus Q126A-Q426E and 2025A-28E

Source: Company data, Visible Alpha, UBS estimates. Note: EPS based on reported and may differ with cover page diluted EPS.

CPO fiber array qualified, but rising pass-through as it upgrades to FAU production

Largan's fiber array (FA) has passed qualification with a volume customer, and management expects mass-production capacity by mid-2 27 in a newly completed fab dedicated to FA/FAU, using in-house equipment. The company's next step is to produce the full detachable FAU mounted on a carrier, a board with three alignment beams that sits on the optical engine's receptor. Largan can build FAUs now and says the customer would prefer it to supply the full assembly, but qualification is pending with the customer preferring FAU and carrier sampled separately.

Even as its potential to supply the full FAU improve, Largan will buy in most of the content. The customer has qualified another PMLA vendor and prefers semiconductormade optics; V-grooves and carrier beams and boards are outsourced and the customer does not want Largan to make MT ferrules. Management described its role as closer to a contractor and said margins hinge on yields, as the FA carries a high ASP but also a large bill of materials. Supplying the complete FAU could raise billed content, but a larger purchased bill of materials also dilutes the relationship between revenue and value added. Our full-FAU ASP assumption requires Largan to capture that invoice scope.

Next year's product is grating-coupled and works for both CPO and NPO while edgecoupled designs are being explored with the customer. The current product does not go directly to end customers, but management expects to work directly with end customers on next-generation designs (likely 2 28) as it progresses toward multi-row FAs.

To ramp production, Largan has bought several land parcels and will build new fabs from scratch, with construction finishing around 2 3 . It may outsource some capacity in the interim and said it cannot yet be sure capacity will meet demand.

Annual FAU TAM reaches US$1.85bn in 2028 and US$4.5bn in 2030 in our base case

We update our FAU (fiber array unit) TAM factoring CPO/NPO penetration into AI networking switches, NVIDIA multi-rack systems, scale-up on compute and switch boards, and non-NVIDIA accelerators. We assume one passive FAU assembly connecting a fiber array to a photonic engine with one unit per optical engine.

We assume one FAU per optical engine and 32 per switch ASIC in our base case, matching NVIDIA's 32x3.2T engines on its 1 2.4T Spectrum-X package. Broadcom's 1 2.4Tb/s Davisson uses 16 engines at 6.4Tb/s, so our bear case assumes and even mix of these two designs (24 per ASIC). We note that higher switch bandwidth does not automatically increase FAU count if engine bandwidth rises at the same rate. Our 2 3 assumption maintains the 32 FAUs per switch based on bandwidth rising to 2 4.8Tb/s ASICs using 6.4Tb/s optical engines.

Figure 4: UBS scenario based on 32 FAUs per s itch although s itches to date carry a range of specifications

Source: Company Data, UBS

Our base case for the FAU market sums up 4 opportunities for CPO/NPO attach:

1. Scale-out and scale-across s itches

We estimate annual switch ASIC shipments x CPO + NPO penetration x optical engines per ASIC x FAUs per engine. In 2 28, 25% CPO/NPO penetration or 1. m 1 2.4T switch ASICs at 32 FAUs each produces 8. m FAUs. In 2 3 , 55% of 1.1mn 2 4.8T switch ASICs at 32 FAUs drives a market size of 19.4mn FAUs.

Figure 5: Scale-out/scale-across could drive 8.0/19.4mn FAUs in 2028/2030

Source: UBS

2. NVIDIA's multi-rack connections for Rubin Ultra and Feynman

NVIDIA at its 2 26 GTC San Jose proposed a multi-rack option to network together 8 Oberon NVL72 racks for the Rubin Ultra generation and 8 Kyber NVL144 racks. Feedback at the time showed possible transceiver or NPO connections for Rubin Ultra and CPO connections for Feynman. Our 2 28 base case uses 11 , racks, 15% multirack adoption and 75% CPO/NPO penetration, giving approximately 12,4 connected racks.

For 2 3 , our rack count falls to 35, racks as we assume Kyber would have 4x the number of GPUs, equivalent to 14 k Oberon equivalents. We assume the 35, racks at 55% multi-rack adoption x 8 % CPO/NPO attach gives 15,4 connected racks. We assume about 216 and 576 inter-rack FAUs per rack, respectively, driving a market opportunity of 2.7mn FAUs in 2 28 and 8.9mn in 2 3 .

Figure 6: NVIDIA rack connected pods could drive 2.7mn/8.9mn FAUs in 2028/2030

Source: UBS. Switch count increases to connect to other racks

3. Initial CPO/NPO optical connections on compute and s itch boards by 2030

We include by 2 3 CPO native optical connections on compute and switch boards in 5% of racks each adding 1,152 FAUs across the GPUs and switches or 2. mn FAUs by 2 3 .

Figure 7: NVIDIA compute board optics could drive 2.0mn FAUs in 2030

Source: UBS.

4. Non NVIDIA scale-up opportunity

Non-NVIDIA scale-up uses accelerator rack shipments x optical attach rate x FAUs per rack, producing about .5m units in 2 28 and 3.7m in 2 3 .

Figure 8: ASIC market could drive additional compute board optics could drive 3.7mn FAUs in 2030

Source: UBS

Overall market opportunity: 12.3mn/35.6mn generates a US$1.85bn/US $4.5bn TAM in 2028/2030

We sum four pools: scale-out/scale-across switches, NVIDIA multi-rack pods, CPO native NVIDIA GPUs and non-NVIDIA scale-up. We add 1 %/5% for FAUs lost at downstream attach and test and for field spares, driving a market size of in our base case of 12.3mn FAUs in 2 28 and 35.6mn in 2 3 . At US$15 /US$125 per FAU, that is a US$1.85bn market in 2 28 and US$4.5bn in 2 3 (bear/bull US$ .5bn/US$3.5bn in 2 28; US $1.4bn/US$9. bn in 2 3 ). The expansion to 2 3 comes from wider optical adoption and greater content in scale-up systems, despite an assumed FAU ASP decline from US $15 to US$125.

Figure 9: Summary of FAU drivers: 12.3mn/35.6mn base case in 2028-2030

Source: UBS

Scenario analysis on Largan's CPO opportunity

Our base case gives Largan the benefit of the doubt for successfully ramping up a full FAU production and capturing 25% of the overall FAU TAM, yielding NT$35 of 2 28E EPS, about 16% of our 2 28E lens EPS. A 25% share means 3.1mn FAUs in its first full year of production: about 29% of NVIDIA-linked units and roughly three-quarters of the allocation our checks suggest its module partner holds. We book these at the full US $15 FAU price and 5 % gross margin, allowing for an early-supply premium even though the optics, V-grooves and carrier parts are bought in. That gives 2 28E CPO sales of US$462mn, rising to US$1.1bn and NT$84 of EPS in 2 3 on a larger market.

Our bear case (15% share, 45% GM) gives NT$5 of CPO EPS in 2 28E and NT$14 in 2 3 E; our bull case (35% share, 55% GM) gives NT$1 3 and NT$267. Each 5ppt of share moves 2 28E CPO EPS by ~NT$6.9 (~NT$34 /share at 49x), and each 5ppt of gross margin by ~NT$4.6 (~NT$225/share). If gross margin were 35% rather than 5 %, holding operating expenses at 12% of sales, EPS would fall to approximately NT$21. A next-generation shift in lens or coupling design could also change content and alignment requirements including a potential metalens design which could invite in different suppliers.

Our 2 3 base case implies ~8.9mn Largan FAUs a year (21mn in the upside case), though with risks around competition in a crowded field, yields and technology. A shift to semiconductor-made optics or metalens would loosen alignment tolerances and erode the precision premium behind Largan's pitch. The upside lies in direct work with cloud providers on multi-row FAs from 2 28, where Largan's multi-layer alignment is most differentiated.

Figure 10: Largan do nside/base/upside cases for 2028 and 2030

Source: UBS

Valuation and investment implications

We roll our valuation forward to 2 28E and value Largan on a sum of the parts. We value the lens business at NT$3,3 (15x 2 28E EPS, the middle of Largan's 13-18x historical range) and CPO at NT$1,7 (49x our base-case 2 28E CPO EPS of NT$35, near the top of optical peers' 14-54x). Our base case gives Largan 25% of a 12.3mnunit, US$1.85bn FAU market at US$15 per unit and a 5 % gross margin. The 49x multiple also assigns a substantial value to a business whose complete assembly still requires qualification and whose margins still need to demonstrate can reach similar levels to its mature smartphone business. At ~NT$6,66 recent share price level, the shares imply ~NT$69 of 2 28E CPO EPS, about twice our base case.

Upside risks include faster full-FAU qualification, firm volume commitments from additional customers, stronger lens demand and better yields. Near-term catalysts are the customer's capacity forecast and its visit to the new fab in the coming month. A large forecast or faster FAU plus carrier qualification would be upside risks to our sell.

Figure 11: UBS SOTP valuation for Largan

Source: Company data, UBS estimates

Figure 12: Optical component peers average 27x 2028 ith a broad range of 12-75x 2028 P/E

Source: UBSe, Eikon

Figure 13: Share price rally has re-rated Largan ell above its historical range

Source: LSEG Workspace

Figure 14: Largan's share price has run ell ahead of for ard EPS as investors price in CPO

Source: LSEG Workspace

Figure 15: UBS P&L for Largan - mild gro th at lo er margins in recent years

Source: Company data, UBS estimates. Note: 2 26E EPS based on reported.

Figure 16: UBS Cash Flo Summary - staying positive to invest in CPO

Source: Company data, UBS estimates

Largan Precision (3008.TW)

Source: Company accounts, UBS estimates. (UBS) metrics use reported figures which have been adjusted by UBS analysts. 1 Cash EPS (UBS, diluted) is calculated using UBS net income adding back depreciation and amortization.

Largan Precision (3008.TW)

Source: Company accounts, UBS estimates. (UBS) metrics use reported figures which have been adjusted by UBS analysts.

Forecast returns

Company Description

Founded in 1987, Largan Precision (Largan) is one of the world's leading manufacturers of lenses used in cameras and IT products. It specialises in plastic/hybrid lenses and, to a lesser extent, glass lenses and modules. In 2 22, 2 MP+ camera lenses were 1 -2 % of Largan's shipment mix, 1 MP lenses were 5 -6 %, 8MP lenses were -1 % and other lenses were 3 -4 %.

Valuation Method and Risk Statement

Our price target is based on SOTP valuation on the CPO opportunity and traditional lens business.

We think Largan's downside risks include: 1) ongoing pricing pressure from smartphone OEM clients and increasingly intense competition among lens suppliers; and 2) a slower-thanexpected smartphone demand recovery in China, and 3) CPO commercial success below the market's elevated expectations.

We believe the key upside risk is a faster-than-expected CPO qualification process at Largan. In addition, if the company demonstrates exceptionally efficient and highly accurate automation capabilities, it could support greater market share gains within the industry than we currently anticipate.

Quantitative Research Revie

UBS Global Research publishes a quantitative assessment of its analysts' responses to certain questions about the likelihood of an occurrence of a number of short term factors in a product known as the 'Quantitative Research Review'. The views for this month can be found below. Views contained in this assessment on a particular stock reflect only the views on those short term factors which are a different timeframe to the 12-month timeframe reflected in any equity rating set out in this note. For previous responses please make reference to (i) previous UBS Global Research reports; and (ii) where no applicable research report was published that month, the Quantitative Research Review which can be found at https://neo.ubs.com/quantitative, or contact your UBS sales representative for access to the report or the Quantitative Research Team on ubs-quant-answers@ubs.com. A consolidated report which contains all responses is also available and again you should contact your UBS sales representative for details and pricing or the Quantitative Research Team on the email above.

Largan Precision