Earnings Highlights: Luxnet (4979) | 2026 Q2 — The 3-Point Margin Dip Is Timing; the Real Signal Is InP Moving from 2-Inch to 4-Inch

2 days ago

9 min read

• Q2 2026 revenue hit a record NT$1.289 billion (QoQ +8.0%, YoY +18.5%), but gross margin fell from 25.4% to 22.4% and EPS dropped from NT$1.62 to NT$1.29. The company attributed this mainly to a timing difference in tax/accounting recognition (to be recognized back in Q3) and a 10% surtax on undistributed earnings levied in June — not price cuts.

• What really matters is not this quarter's P&L but the timeline by which Luxnet is rewriting itself from a "module packaging house" into an "Indium Phosphide (InP) chip fab + optical engine packaging house": the entire Hejiang plant converted into an InP wafer fab, the 5,500 m² Bade plant coming online in 2027, cumulative cleanroom space of 13,000 m² by end-2027, and capex of NT$1.8–2.0 billion in 2026 and a conservatively estimated NT$3 billion+ in 2027.

• The hardest line came in the Q&A: for 800G, "orders are not the problem; it's mainly a supply problem." That means the remaining variable for 2026 is not demand but Luxnet's own capacity ramp and yield.

1. Executive Takeaway: Three Core Signals

Signal 1: The quarterly margin dip is accounting rhythm, not a price war. Gross margin was 22.4%, down 2.98 points QoQ. Management cited a tax/accounting recognition timing difference that will be caught up in Q3; meanwhile, the 10% surtax on undistributed earnings in June pushed Q2's effective tax rate above the full-year estimate of 20%. Net income still nearly doubled YoY (+94.7%) — the most direct evidence for judging "one-off vs. structural."

Signal 2: The company's center of gravity is shifting from "modules" to "chips." Revenue today is roughly 85% high-speed modules + optical engine packaging and about 15% InP. Yet capex, buildings and equipment are almost entirely bet on InP and packaging: the whole Hejiang plant (about 5,000 m²) becomes an InP chip fab with 3,000 wafers/month of capacity, and 4-inch/6-inch automated lines are planned for 2028. This is not capacity expansion — it is a change of lanes.

Signal 3: NPO is explicitly ranked ahead of CPO. Management said outright that from 2027 to 2029, Near-Packaged Optics (NPO) will replace Co-Packaged Optics (CPO) as the mainstream, because CPO yields are poor and failed units cannot be repaired; CPO won't truly go mainstream until 2030. That ordering explains why Luxnet has scheduled development of its 1.6T NPO packaging equipment for completion in Q2 2027. We broke down this roadmap debate in full in 2026 OCP APAC Summit | CPO / NPO / XPO Panel: Not a Roadmap Fight, but the 409.6T Bottleneck 18 Months Out.

2. Revenue and Financials: Record Revenue, Narrowing Profit

Metric	2026 Q2	QoQ	YoY
Revenue	NT$1.289B (quarterly record)	+8.0%	+18.5%
Gross margin	22.4%	−2.98pp	—
Operating margin	18.3%	−2.66pp	—
Net income	NT$184M	−19.8%	+94.7%
EPS	NT$1.29	−20.4%	~+93%

First-half totals: revenue NT$2.483 billion (YoY +15.2%), net income NT$414 million (YoY +34.4%), EPS NT$2.91 (YoY +32.9%). July revenue of NT$509 million set another monthly record (YoY +31.0%), bringing the seven-month total to NT$2.991 billion (YoY +17.6%).

These three numbers only make sense read together:

First, the QoQ decline came from taxes and recognition, not the core business. Revenue rose QoQ and operating margin held at 18.3%, yet net income fell nearly 20% QoQ — the gap sits in taxes. The company assumes a 20% full-year tax rate; Q2 ran higher due to the undistributed-earnings surtax.

Second, YoY is this company's true slope. Net income rose 94.7% YoY, and EPS jumped from about NT$0.67 a year ago to NT$1.29. Gross margin a year ago was below 20%; even after "dropping," it is still 22.4% this year — the result of a product mix shifting toward 800G and CW light sources.

Third, there is a structural margin pressure that wasn't spelled out. As 800G enters volume production, the company must procure high-priced components on customers' behalf. An 800G module sells for several times a 400G one, but these purchased parts are low-margin pass-through revenue. That is why revenue hit a record while gross margin loosened first. We ran the same BOM logic in 800G vs. 1.6T Optical Modules: A Generational Leap — the BOM Teardown Showing Who Really Captures the Margin.

3. Technology and Business Highlights: Three Lines, One Vertical Integration

In this call, Luxnet split itself into three plants and three lines, each with a clear role:

• Hejiang plant: the entire building becomes an InP chip fab — 3,000 wafers/month, about 5,000 m².

• Jilin plant: packaging base, including NPO optical engine packaging.

• Bade plant: DCI module assembly and test.

Light sources: CW is the main battlefield; EML retreats to high-performance niches. The company clearly sees Continuous Wave (CW) lasers as the future mainstream, while EML (Electro-absorption Modulated Laser) stays in high-power, long-reach, single-mode applications. This matches LightCounting's 2026 view: silicon photonics modulators are eating EML's market, but the InP laser itself is not being replaced — it is getting even tighter as CW light source demand surges.

CW product progress runs like this: 2-inch wafers for 20/40/70/100 mW CW and 100G EML; 3-inch since 2025 for 70/100 mW CW and 100/200G EML, with 400 mW (DFB + SOA) to follow; 4-inch and 6-inch only after 2028. This year wafers will reach 4-inch.

The real moat lies in wavelength count: the company's CW lasers already cover 16 wavelengths, while most of the market does only single or 4 wavelengths. Multi-wavelength is harder on yield and alignment, but margins are much higher — exactly the logic by which peer LandMark Optoelectronics turned "multi-wavelength" into a moat. Compare with Earnings Highlights: LandMark Optoelectronics (3081) | 2026 Q2 — Gross Margin Hits 57%, Record EPS of NT$4.62 as SiPh Lasers Turn "Multi-Wavelength" into a Moat.

On EML, it's a combination of "outsourced epitaxy + in-house process." Epi wafers come from Optoway, the team comes from Source, and Luxnet runs the EML process, with volume production starting in 2H26 and scaling in 2027. High-gain APDs are likewise slated for volume production in the second half.

Packaging is already paving the way for NPO. Beyond existing CoC and CoB, the company is developing NPO packaging, expected to contribute meaningfully to revenue in 2028. In process capability, die bonding already reaches 0.3 μm; flip chip will be developed this year, along with the Fiber/Lens array attachment NPO requires, with FAU/Lens active alignment accuracy below 0.1 μm.

Fiber-to-chip coupling is a bottleneck shared by every CPO/NPO player. We discussed why this step determines yield in [CPO Breakdown 3/6] The Most Fragile Link in CPO Is the Laser.

Modules: 800G in volume, 1.6T NPO slated for 2027. 800G is in volume production this year, with next year's demand above expectations. 1.6T NPO packaging is being co-developed with Taiwanese and Japanese equipment makers, targeting development completion in Q2 2027, volume production in 2H27, and a real ramp only in 2028 (the company's internal view is more conservative, pointing to 2H28). After that comes 3.2T NPO, extending to CPO in 2030.

4. Management Outlook: Converting Timelines into Numbers

Luxnet gave no revenue guidance on this call, but it did give three sets of verifiable hard numbers.

Set 1: Market demand. LightCounting has raised its data center CPO/NPO/transceiver forecast three times, lifting the 2031 market size from US$52 billion to US$80 billion. On the module side: in 2026, about 70 million 800G units (last year's forecast was only 35 million) and about 15 million 1.6T units; in 2027, 1.6T could reach 60 million units.

Converted to light sources it's even clearer: 2026 CW demand is about 400 million lasers (roughly 85 million modules), and EML about 160 million (roughly 18–20 million modules). One multi-wavelength module consumes several lasers — which is why chip-level capacity is worth more than module-level orders.

Set 2: DCI, the underrated line. In 2026, 400ZR remains mainstream with a little 800G starting to appear; next year the two combined reach about 2 million units, rising to 3 million by 2031. Broken down, 400G is flat to declining while 800G grows significantly: 1.0/1.3/1.5 million units in 2027/2028/2029. Current share leaders: Eoptolink in 400G, Innolight in 800G, Marvell in ZR. The company's own view is more aggressive: in 2027, revenue from new-generation DCI products will match existing DCI, meaning the DCI line more than doubles in 2027; in 2028, new products will be several times existing ones. Next-gen DCI is already in R&D, with small volumes appearing in 2027/2028.

Set 3: Capacity and capex. This was the most information-dense part of the call.

• Cleanroom: 4,000 m² in 2024 → +950 m² at Hejiang in Q1 2025 and +1,500 m² at Jilin in Q3 → 1,200 m² at Zhudong in 2026 (production starts Q3) plus 500 m² of office → the 5,500 m² Bade plant purchased in May 2027 is completed by year-end, bringing the cumulative total to 13,000 m² → another ~10,000 m² added in 2028.

• Wafers: annual capacity of about 3,000 wafers in 2026 and about 6,000 in 2027; die fab capacity of about 30 million chips. Actual 2025 output was 3,000 wafers and 14 million chips. From Q2 2026, the share of chips going into 1.6T rose noticeably.

• Laser back-end packaging monthly capacity: 6 million units/month by end-2026 → 12 million units/month by end-2027. In 2027, 2-inch and 4-inch will coexist depending on customer demand and substrate supply; 4-inch equipment begins arriving in Q4 2027.

• Capex: higher than originally expected in 2026 at NT$1.8–2.0 billion; conservatively NT$3 billion in 2027, with construction still to come in 2028. Depreciation periods are long. The company specifically noted that its InP products are backed by customer-consigned equipment, reducing investment risk.

Read the wording by tier. "2H26 will grow sharply vs. 1H" is an expectation; "order visibility extends to 2028" is a confirmation; "new products will deliver step-change revenue growth in 2027" is a target. These three tones carry entirely different levels of certainty, yet investors often read them as the same thing.

5. Supply Chain and Customer Clues

Customer concentration is this company's most obvious double-edged sword. Modules + packaging make up 85% of revenue, InP 15%. DCI is concentrated in two brands, and Luxnet remains the "exclusive supplier to a single customer" — exclusivity means strong pricing power and stickiness, but also ties this line's revenue entirely to one customer's capex cycle. Management's answer is diversification on the InP side: adding customers and product types this year to reduce risk.

The non-China supply chain dividend has been flowing for two years. The company said plainly that it began capturing the dividend in 2024 when the non-China supply chain started to take shape. US restrictions on Chinese optical modules are a sustained tailwind for Taiwan's optical supply chain — but the real bottleneck is not in modules; it's in upstream substrates, which we broke down in You Can Ban Modules, but Not Substrates: The Decisive Battle in Optical Communications Has Moved Down to InP.

How to read the industry positioning. The same week's Taiwanese earnings calls complete the puzzle: LandMark (3081) makes CW laser chips, Elaser (3450) does optical engine and external light source packaging, and Luxnet (4979) is one of the few spanning all three stages — "InP chips → back-end packaging → module assembly." Vertical integration keeps the margin in-house, at the cost of far higher capex intensity than single-stage players. Elaser's report is worth reading alongside: Earnings Highlights: Elaser (3450) | 2026 Q2 — Quarterly EPS Triples, but the Real Signal Is an External Light Source Not Yet in Volume.

6. Risks: Four Variables to Watch

1. The flip side of exclusivity is single-point risk. DCI is concentrated in two brands, and the company remains the exclusive supplier to a single customer. If that customer changes specs or delays a project, the assumption of "more than doubling revenue in 2027" shifts wholesale.

2. Capex is heavily front-loaded. NT$1.8–2.0 billion in 2026 and NT$3 billion+ in 2027, against 1H26 revenue of NT$2.483 billion, is a pace that bets the next two years of cash flow. Longer depreciation periods smooth quarterly depreciation, but this capacity must actually be filled.

3. Nearly every milestone lands in 2027–2028. 1.6T NPO packaging completes development in Q2 2027 and ramps in 2028; 4-inch equipment arrives only in Q4 2027; the Bade plant is completed only at end-2027. The rest of 2026 is essentially about execution and ramp, not a new story.

4. Revenue growth does not equal margin growth. 800G modules require procuring high-priced components on customers' behalf, which is low-margin revenue. What really drives gross margin is how fast the shipment share of InP chips and multi-wavelength CW climbs — not how many modules ship.

7. Conclusion

Luxnet's P&L this quarter reads like a tap on the brakes: margin down, EPS down. But strip out the tax recognition timing difference and what's left is a company with record revenue, operating margin still above 18%, and net income doubling YoY.

This quarter's real information content is not in the P&L but in the cleanroom floor-area table.

From 4,000 m² in 2024 to 13,000 m² by end-2027 and another 10,000 m² in 2028, this curve speaks more honestly than any line of guidance about management's confidence in 2028. Combined with two details — "order visibility to 2028" and "customer-consigned equipment" — it is reasonable to infer that this capacity is not speculative investment, but is being built on the back of customers' long-term contracts.

Three metrics to track next quarter:

1. Whether 3Q26 gross margin returns to around 25%. This is the only way to verify the "tax recognition timing" explanation. If it doesn't recover, switch to the structural explanation of "800G purchased components diluting margin."

2. Whether monthly revenue can hold above NT$500 million. July's NT$509 million was a monthly record; the claim that new DCI products, CW and EML all ramp in the second half will show up directly in the slope of monthly revenue.

3. Actual progress of Q3 EML volume production and the Q3 Zhudong plant start-up. Both are scheduled for the same quarter and are the only execution metrics that can be verified in real time in 2H26.

This article is for technology and industry trend analysis only and does not constitute investment advice.