Earnings Highlights: Elaser (3450) | 2Q26 — EPS Triples, but the Real Signal Is an External Light Source Not Yet in Volume Production

2 days ago

9 min read

Elaser posted 2Q26 EPS of NT$2.09, up 202.9% QoQ, with gross margin jumping to 31.5%. But the most important thing in this report is not the numbers themselves — it is the revenue mix. AI-related optics rose from 32.7% to 45.5% of revenue in just seven months, giving a company that used to rely on power-device packaging and testing an entirely new skeleton. And its biggest bet, ELSFP (external laser small form-factor pluggable light source), is still not in volume production — that is the real variable for 2027.

1. Three Core Conclusions

First, this is a "structural transformation" report, not a "cyclical recovery" report. First-half revenue totaled NT$5.064 billion, up only 9.9% YoY, and first-half EPS of NT$2.78 was actually down 3.1% YoY. Looking only at the half-year report, Elaser looks unremarkable. But broken down by quarter, 2Q26 revenue was NT$2.977 billion, up 42.6% QoQ and about 61% YoY, with gross margin up 6 percentage points QoQ — all the growth was compressed into a second-quarter surge. That means the inflection is real, and steep.

Second, what Elaser sells is not parts — it is "yield." Seven of the world's top ten optical module makers are Chinese companies, and Elaser works with none of them; of the remaining three US companies, two have been its customers for more than ten years, and this year it added a Japanese customer. When asked in Q&A, "A major US player says it will make COS in-house — where is your moat?", management answered candidly: buying equipment to make COS is easy; doing it well is hard. Translated into business terms, Elaser's pricing power comes from being the only third party able to package high-power CW lasers at volume-production yields.

Third, ELSFP is a valuation option — and the biggest uncertainty. President 宋天增 made a blunt statement at the call: "Without an external light source, there is no CPO." One CPO module needs 18 external light sources, each rated at least 300mW (23dBm). Yet when asked about next year's contribution, the company itself hit the brakes: "Not that fast — there may be small shipments, but we can't forecast them." That gap is itself a signal: the technical positioning is done, but the timeline is entirely tied to customers' CPO/NPO adoption pace.

2. Revenue and Financials

Quarterly consolidated P&L (NT$, consolidated GAAP basis):

• Revenue NT$2.977 billion, up 42.6% QoQ and about 61% YoY

• Gross profit NT$939 million, up 73.1% QoQ; gross margin 31.5% (presented as 32% in the company deck), up about 6 percentage points QoQ

• Operating expenses NT$307 million, up 22.1% QoQ — opex grew far slower than revenue, a second lever beyond gross margin

• Operating income NT$631 million, up 117.1% QoQ; operating margin 21.2%

• Pre-tax income NT$659 million, up 116.0% QoQ

• Net income (consolidated) NT$501 million, up 178.9% QoQ; of which attributable to the parent NT$305 million, up 204.0% QoQ and 295.4% YoY

• EPS NT$2.09, up 202.9% QoQ (1Q26: NT$0.69)

One easy misread here: there is a gap of nearly NT$200 million between consolidated net income of NT$501 million and NT$305 million attributable to the parent. The reason is that Elaser holds 51% of GEM Services (6525) and 54.23% of Centera Photonics (7917), so a sizable share of the two subsidiaries' profits belongs to non-controlling interests. Use the attributable-to-parent line for EPS, and the consolidated figures for operating momentum.

Balance sheet: cash and equivalents stood at NT$3.087 billion at the end of the first half; the current ratio fell from 201% to 189% — normal during a capacity expansion as cash goes into equipment; book value per share rose from NT$34.76 to NT$38.34.

The revenue mix is what matters. As of July 2026, AI-related optics (including optical modules) accounted for 45.5% of revenue, versus just 32.7% for full-year 2025; power devices (GEM Services) fell from 62.0% to 50.4%; traditional consumer optics fell from 5.3% to 4.1%. To be clear, the falling power-device share is not a decline — GEM Services' July revenue hit a record NT$614 million, up 32.5% YoY (NT$465 million a year earlier). The optical communications line is simply growing faster.

Management also gave the margin structure: GEM Services close to 30%, Centera Photonics just above 30%, and Elaser's own COSA business above 35%. That explains why every percentage point the mix shifts toward AI optics lifts overall gross margin.

3. Three Platforms: COSA, TRX, ELSFP

Elaser frames its growth drivers as three platforms — a much clearer cut than talking product lines.

COSA (optical sub-assembly) — today's cash cow. Capacity has nearly doubled every year since 2024, and 2026 keeps that slope. The business packages EML and CW laser chips into usable optical sub-assemblies, serving both EML and CW lasers, with CW power ranging from 70mW (chip length ~0.8mm) up to 400mW (~3mm). Management singled out the process difficulty: voids inside long chips affect laser performance and heat dissipation, and 400mW/3mm is the hardest to make. This is not marketing talk — it is the physical version of why the laser light source is the most fragile link in CPO: the light source itself hates heat, yet the architecture demands it sit close to heat sources, so its reliability ceiling is the ceiling of the entire product line.

Elaser's real value proposition is "helping laser wafer fabs get new products out the door" — helping customers scale from 3-inch to 6-inch, push 70mW to 300–400mW, and it has fully validated the entire 70mW–400mW range for silicon photonics lasers. EML and CW each make up about half of the COSA business. Take Coherent: Elaser has worked with it on co-design and development since 2024, and Coherent broke ground in Texas in the first half of this year. The demand side of this line can be cross-checked against Lumentum's FY2026 Q4 report, which turned "optics replacing copper" into numbers — when the upstream laser leader is running at full capacity, midstream packaging naturally runs full too.

TRX (optical transceivers) — the second engine, led by Centera Photonics. The division of labor: Centera designs, Elaser manufactures, handling yield, capacity expansion and stable volume production. After acquiring Centera, it redesigned its 400G AOC and pushed yield to 99%; AOC capacity utilization is currently about 70% and will keep climbing, with a move into 800G. Management is relatively cautious on 800G: it still needs to pass qualification, and next year's market share is uncertain. We have previously broken down Centera's technical foundation — Centera Photonics is really the designer of "the entire optical path"; its value lies not just in AOC, but in building out the silicon photonics optical engine across five stages.

ELSFP (external laser source) — the ticket on post-2027. Both CPO and NPO architectures need an external light source, and Elaser's ELSFP serves both. It has been co-developed with a major laser chip maker and is in engineering pilot production. Key numbers: one CPO module needs 18 external light sources, each at least 300mW (23dBm). As for CW chip sources, "we have them all" — customers ask Elaser which ultra-high-power laser vendor it recommends, and end customers also specify directly.

The CPO vs. NPO route debate recently had a hard showdown at OCP, and Elaser's "play both sides" positioning lands right at the core of that discussion — CPO / NPO / XPO panel: this isn't a route war, it's the 409.6T wall 18 months out. Whoever wins in the end, the external light source stays on the BOM.

4. Management Outlook and Capex

The company gave no revenue guidance, but it did give two hard forward-looking numbers.

Capital expenditure (machinery and equipment only, excluding land, buildings and software): about NT$260 million in 2024 → about NT$1.1 billion in 2025 → an estimated NT$1.6 billion in 2026, roughly NT$3 billion over three years. That slope is itself guidance — equipment investment leads revenue by about one to two years, so 2026's NT$1.6 billion corresponds to 2027 output capacity. Management added that upstream equipment lead times are long and tools will arrive month by month, so capacity will "ramp" rather than switch on all at once.

On August 13 the board approved issuing about NT$3 billion in convertible bonds, to buy machinery and equipment and repay bank loans (lowering the debt ratio). Note that this is a new authorization "on top of" the NT$3 billion capex above, not the same money. For shareholders it is a clear dilution warning — and a clear expansion signal: the company chose to lock in funding just as gross margin turned up, judging that demand visibility is long enough.

Language analysis: for COSA, management said "we can forecast doubling every year" (high certainty); for AOC, "utilization at 70% and will keep climbing" (direction certain, magnitude withheld); for 800G, "needs qualification, next year's share uncertain" (explicitly reserved); for ELSFP, "not that fast, can't forecast" (explicit brakes). The gap between these four tones is the company's risk ranking for the next 18 months.

5. Supply Chain and Customer Clues

Customer concentration is extremely high — and deliberately so. Of the world's top ten optical module makers, Elaser does no business with the seven Chinese companies; two of the three US companies have been customers for over ten years, and a Japanese customer was added this year. So this expansion is mainly driven by more demand from existing customers, plus one new customer. The upside of this structure is high stickiness and deep co-development; the downside is that any inventory adjustment by a major customer hits quarterly revenue directly.

Upstream shortages are real. Management described widespread shortages across the AI supply chain: DRAM, compute boards, laser chips, thermal materials and packaging are all short, and judged that shortages will be the market's main theme for the next few years. That matches our observation last week — the decisive battle has moved down to indium phosphide: module-level capacity can be shifted, but there is currently no fix at the materials level. Elaser sits in the packaging step "between chips and modules," so in theory it is less directly hit by substrate price hikes than companies that make their own chips — but it also means its ceiling is set by how many chips upstream can supply.

Peer comparison: reporting the same week, LandMark Optoelectronics (3081) saw 2Q26 gross margin surge to 57% with EPS of NT$4.62 — that is the margin level of in-house epitaxy and chip design; Elaser's COSA business at 35%+ is the margin level of outsourced packaging. The two companies sit at different positions, so their margins can't be compared head-to-head, but together they show how value is split between the "chip end" and the "packaging end" of the same supply chain.

6. Risks and Counterarguments

• Dilution risk: share-capital expansion after the NT$3 billion convertible bond lands will directly flatten the EPS growth slope.

• Customer in-sourcing threat: a major US player has publicly said it will make COS in-house. Management's rebuttal (buying equipment is easy, doing it well is hard) holds technically, but if a big customer merely pulls 30% of volume in-house as a bargaining chip, that is still real pressure on Elaser's pricing power.

• ELSFP timing risk: this is the most expensive piece of the valuation, yet the one the company is least willing to put a timeline on. If CPO/NPO adoption slips another year, this platform's contribution moves from 2027 to 2028.

• A reminder from the half-year report: 1H26 EPS was NT$2.78, down 3.1% YoY. One explosive quarter doesn't mean the full year extrapolates linearly; part of 1Q26's low base was a demand gap.

• Subsidiary profit attribution: strong consolidated numbers don't mean shareholders get more — an inevitable discount of the 51% / 54.23% ownership structure.

7. Conclusion

What Elaser really accomplished this quarter was repositioning itself from "a power-device packaging and testing company that also does optics" to "an AI optics packaging company that also has a stable power-device cash flow." The 45.5% vs. 50.4% golden cross in the revenue mix will probably flip officially next quarter.

Its business model is actually simple: in a supply chain where everyone is short of materials, own the step of "packaging hard-to-package high-power lasers at production volume," then use capacity that doubles every year to turn that step into a toll booth. The position has no flashy technology story, but it has two very good properties — it doesn't have to buy the most expensive InP substrates itself, and it doesn't have to bet on whether CPO or NPO wins.

Three numbers to watch next quarter:

1. Whether AI optics breaks 50% of revenue — the confirmation signal of the structural flip, which also directly decides whether overall gross margin can reach 35%.

2. How far AOC utilization rises from 70%, and whether 800G qualification is confirmed — the only checkpoint for whether the Centera line delivers this year.

3. The actual terms and funding timeline of the convertible bond — coupon, conversion price and issue timing will determine how painful the dilution is.

As for ELSFP, don't expect numbers in 2026. What to watch is whether it moves from "engineering pilot production" to "small-volume shipments" — that change of verb will matter more than any single quarter's EPS.

This quarter's beautiful numbers buy the present; that NT$3 billion convertible bond buys 2027.

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