LandMark Optoelectronics (3081) Q3 2026 Earnings: EPS Hits a Record NT$6.12, but the Real Signal Is 30–50% Fewer Lasers per Wafer

17 hours ago

8 min read

1. Executive Takeaway

First, order visibility already exceeds this year's revenue. Major customers have placed orders running more than six months out, and the company says its backlog already exceeds full-year 2026 revenue. The shortage is no longer an adjective; it is written into contracts.

Second, expansion has shifted from chasing demand to building against contracts. In August LandMark signed a four-year LTA with a US data center customer and added roughly NT$3B of CAPEX to fulfill it, bringing the past year's total to more than NT$8B. That buys +150% output in both 2027 and 2028, about 6x over two years. The company stressed that it expands according to customer contracts, not according to whatever forecast customers hand over.

Third, high power is the demand multiplier. CPO-class CW lasers need 250 or even 300mW. Die size grows and yield falls, so each wafer produces 30–50% fewer lasers. That inserts a multiplier between laser unit growth and epitaxy capacity demand; counting only units underestimates the wafers needed.

The laser shortage is not just about volume: every laser now eats more wafer.

2. Revenue and Financials

LandMark again raised all three margins in Q3, and by a wide step:

Metric	3Q26	QoQ	YoY
Revenue	NT$1.562B	+28%	+183%
Gross margin	58.56%	+1.1pp	+15pp
Operating income	NT$771M (49.4% operating margin)	+35%	+644%
Net income to parent	NT$621M	+33%	+528%
EPS	NT$6.12	+33%	—

For the first nine months, revenue was NT$3.688B (+136% YoY), net income NT$1.407B (+491% YoY) and EPS NT$13.88, already three times full-year 2025 EPS. September revenue of NT$540M set another monthly record.

These numbers carry two layers of meaning. The first is operating leverage: revenue grew 28% QoQ and operating income 35%, so almost all incremental revenue dropped to profit. The second is product mix: Datacom is 80–85% of revenue, mostly CW lasers for silicon photonics (SiPh); Telecom is down to 5–10% and others about 5%. That puts roughly NT$1.25–1.33B of quarterly revenue in Datacom. LandMark is now a silicon photonics laser company, not a traditional optical epitaxy house.

As we noted in LandMark Q2 2026 earnings highlights, gross margin passed 57% mainly because multi-wavelength products gained share. This quarter it rose another 1.1pp to 58.56%: same direction, but a clearly flatter slope, which matches what management said about the outlook.

LandMark 3081 Q3 2026 earnings data card

▲ LandMark Q3 2026 key financials: revenue NT$1.562B, 58.56% gross margin, 49.4% operating margin, EPS NT$6.12, and Q4 2026 guidance of about +16% QoQ (Image source: Simple Tech Trend original; data from LandMark's Q3 2026 earnings call and filings)

3. Technology and Business Highlights

1.6T in volume production; 3.2T, NPO and CPO in validation

1.6T silicon photonics products are in volume production. Higher-power SiPh products for 3.2T, NPO (Near-Packaged Optics) and CPO (Co-Packaged Optics) are in joint development and validation with customers. Next-generation FTTH, smartphone LD/PD and LiDAR are extensions of existing product lines and are on schedule.

Asked whether lasers above 100mW must integrate an SOA (semiconductor optical amplifier), the company answered pragmatically: every customer's design differs, and LandMark has solutions with or without an SOA. In other words, high power is not a single technology path but a custom list of specs. Whoever can serve the most variants wins the most programs.

Multi-wavelength is the moat, with ASP 50% above single-wavelength

Eight-laser designs started at 800G, and 1.6T has both 8- and 16-wavelength designs, all in volume production. Multi-wavelength lasers carry an ASP about 50% higher than single-wavelength parts, and very few suppliers worldwide have both the capacity and the experience to make them. LandMark considers itself one of its customers' main suppliers.

This is also why laser demand cannot be counted in units alone. As we broke down in Laser Demand Is Counted in Wavelengths, Not Units, once DWDM architectures win, the number of wavelengths per fiber multiplies, and the beneficiaries are the epitaxy houses that can make multi-wavelength lasers.

On process, multi-wavelength parts can be made with DUV, but semiconductor-grade masks are expensive, so the volume per wavelength decides between DUV and the other two grating processes. Running three processes in parallel is how LandMark turns multi-wavelength from a technical capability into capacity flexibility.

High-power CW lasers: bigger dies, lower yield, 30–50% fewer lasers per wafer

This was the line most worth remembering from the call. CSPs sign LTAs with LandMark to lock output, and CW lasers for CPO need 250 or even 300mW. Even if the number of laser chips stays the same, larger dies and lower yield cut the lasers each wafer produces by 30–50%, which greatly amplifies demand.

The company also gave boundary conditions: die size grows noticeably only above about 300mW per laser. On substrates, LandMark has been moving from 2-inch to 3-inch, but 4-inch supply is scarce and almost nobody produces 6-inch in volume. Each step up in wafer size can cost several times more, so costs do not fall proportionally. Bigger wafers are not the answer; stable supply is.

Higher-power CW lasers cut usable dies per InP wafer by 30-50%

▲ Moving from mainstream CW lasers to 250–300mW parts cuts usable lasers per InP wafer by 30–50%, so the same laser count needs roughly 1.4–2x the epi wafers (Image source: Simple Tech Trend original; data from LandMark's Q3 2026 earnings call)

4. Management Outlook

Q4 2026: about +16% QoQ, with Q1 2027 as the next step up

The company expects Q4 2026 revenue growth to be similar to Q4 2025 (+16.2% QoQ), implying about NT$1.81B for the quarter. On that guidance, full-year 2026 revenue would be about NT$5.5B, roughly 2.5x 2025's NT$2.203B (STT estimate).

The next visible jump comes in Q1 2027. Three equipment technologies are being added in parallel: new e-beam tools entered production in July, which is why Q3 revenue accelerated; the second arrives by year-end; the third in 2027. Front-end tools such as MOCVD add more capacity in Q1 2027. Capacity will ramp in batches over the next eight quarters, not all at once.

Capacity: +150% in both 2027 and 2028, about 6x in two years

Output is expected to rise 150% in 2027 versus 2026 and another 150% in 2028. The upgrade came from new contracts signed in August, with the extra CAPEX showing up mainly in 2028. Capacity is measured in laser units.

Management's wording is worth noting. Asked about buying new fabs, the company said output at the current expansion pace is still below customer demand, but that adding 6x output over the next two years is already the most it can do, and higher targets would bring funding risk. This is the language of confirmation, not aspiration: the 6x is a contract-backed commitment, not a slogan to lift the valuation.

Gross margin: the company says it will not keep rising

This was the quarter's most important cooling signal. With capacity rising sharply over the next two years, hiring and equipment depreciation will gradually flow through, and the company does not expect gross margin to keep growing in coming quarters. Pricing will keep a reasonable profit, but not excessive returns. On price, CW lasers have been in short supply for one to two years, customers have long stopped negotiating, ASP has not fallen, and LandMark will raise prices to reflect costs when appropriate.

A few hard numbers on costs:

Q3 2026 depreciation was NT$76.6M, about 4.9% of revenue. The company aims to keep depreciation at 8–10% of revenue, meaning depreciation as a share of revenue could nearly double (STT estimate).
Convertible bond (CB) interest: if not converted to common shares, about NT$15M of interest expense per quarter on average going forward.
Much of the expense growth comes from bonuses, and headcount keeps rising.

Volume will scale; margins may not. From here, watch LandMark's gross profit dollars rather than its gross margin.

5. Supply Chain and Customer Clues

LTAs lock both ends: output on the customer side, InP on the supply side

Of the LTAs signed in August, the largest is a four-year contract with a US data center customer. LandMark added CAPEX for it, and the contract includes some funding support (details not disclosed). On materials, LandMark signed another four-year LTA with a Japanese InP substrate supplier and now has LTAs with two Japanese and one US InP substrate suppliers, ensuring enough wafers once the CAPEX is in place.

We broke down this customer → capacity → materials chain in LandMark's five board resolutions as a risk disclosure. This call effectively confirms the status of those three pressure points: orders are in, tools are on the way, and substrate supply is contracted with three vendors.

Back-end partners: expanding alongside the front end

LandMark ships in two ways: it hands epi wafers to downstream foundry partners who make the laser chips, which LandMark then ships to customers; or it ships epi wafers directly to customers. Asked whether the back end becomes the new bottleneck as the front end ramps, the company said it has added many back-end vendors in Taiwan and overseas, and its 2027–2028 output plans already account for back-end capacity, outsourcing and equipment lead times.

Customer concentration: staying high for now

The company admitted it now has room to choose customers, but it also weighs their mid- to long-term prospects, so customer concentration will remain high for some time. Asked whether this relates to the LTAs, it said it could not comment. The flip side of locking volume with long-term contracts is that bargaining power and concentration rise together.

NPO is the transition in 2H 2027; CPO is not coming that fast

LandMark is cautious on CPO timing: adoption in general data centers will not be quick, since whether the laser sits inside or outside and the connector specs are still undecided, so volume production in the near term is unlikely. NPO will become the transition solution in 2H 2027, nearly all major customers are preparing for it, and it looks like they all use external light sources. Internal and external light sources differ in thermal design, emission method and power, and internal light sources are progressing more slowly.

This matches the discussion at ECOC. As we covered in External Light Sources Are No Longer a Stopgap: ECOC Draws the ELS Spec Line, the bar for external light sources is not simply chasing 400mW but a combined problem of power, heat and serviceability.

Other product lines: EML and PD

EML: always in production, but mainly for Telecom; Datacom 100G/200G EML is a small share. New capacity is focused on CW lasers.
PD: capacity has favored LDs for the past one to two years. A 200G PD requested by a few customers has passed validation, some epitaxy capacity has been set aside, and Datacom PD will enter production in the coming months, though at a small scale compared with LDs.

For the broader industry context, see A-Share Laser-Chip Stocks Priced the Wrong Year: Laser Capacity Is Booked to 2029: 2027 will be decided by InP capacity, not by the restriction list. This call is that thesis confirmed at the level of a single company.

6. Conclusion

LandMark's Q3 2026 proves one thing: the silicon photonics laser shortage has moved from orders to contracts, and from units to wafer area. High-power CW lasers make every laser consume more substrate, and that multiplier will make demand in 2027–2028 look steeper than unit growth alone.

But management also drew two lines: gross margin will not chase higher, and expansion will not exceed contracts. LandMark's next phase is a volume story, not a margin story.

Three metrics to track next quarter:

Whether Q4 2026 revenue lands near +16% QoQ (about NT$1.81B), and whether the second tool type arrives by year-end as planned.
Whether depreciation moves from 4.9% toward 8–10% of revenue, and whether gross margin holds above 55% as depreciation rises.
How fast new capacity comes online in Q1 2027, which decides whether the +150% for 2027 is hit comfortably or only shows up in the second half.

In one line: LandMark is no longer just selling lasers; it is selling InP wafer area pre-booked by long-term contracts.