Phase A — Understand the business
Lens 1 · Company Overview
Eoptolink Technology (新易盛, "Xinyisheng") designs and manufactures high-speed optical transceivers — the pluggable modules that convert electrical signals to light and back at the edge of every switch and GPU rack in a data center. The portfolio spans 100G/200G/400G/800G and now 1.6T modules across QSFP/QSFP-DD/OSFP form factors, for four end-markets: AI/ML clusters, cloud data centers, 4G/5G wireless, and transport/FTTX. The AI-datacenter 800G/1.6T line is the entire growth story; legacy telecom is a rump.
- Founded 2008 in Chengdu, Sichuan; IPO on the Shenzhen Stock Exchange ChiNext board 3 March 2016.
- Chairman & actual controller: Gao Guangrong; President: Huang Xiaolei. (Gao is the subject of a 2026 CSRC penalty — see Lens 9/10.)
- Business model: make-to-order module assembly. Eoptolink buys the two value-bearing components — the DSP/retimer (Broadcom/Marvell) and the EML laser / SiPh modulator (Lumentum, Coherent, plus its own SiPh and TFLN modulators) — and does the optical design, packaging, assembly, and test. It is a module integrator, not a chip owner.
- Contract structure: no take-or-pay disclosed; sales are PO-driven into a tight buyer set. Pricing is competitive — Chinese modules ship at a reported 20–25% discount to Western vendors, which is the lever that wins share but caps pricing power.
- Customers: direct-to-cloud and to NVIDIA. Top-five customers = 71.1% of revenue; overseas sales = 96.16% of 2025 revenue. This is one of the most concentrated, most export-dependent revenue bases in the AI supply chain.
Lens 2 · Supply Chain
Map, named end-to-end (chokepoints in bold):
Upstream chips (the leash):
- EML lasers / 200G-per-lane sources → BOTTLENECK. For 1.6T you need 200G/lane EMLs; Lumentum is currently the only volume supplier of 200G/lane EMLs. Coherent and MACOM are the other Western sources. NVIDIA's $4B March-2026 investment into Lumentum + Coherent explicitly locked up priority EML capacity, pushing non-favoured buyers' lead times past 2027. China's domestic high-speed (≥25G) optical-chip localization rate is below 15% — Zetta Semiconductor sampled a 200G/lane PAM4 EML in late 2025 but mass production is years out. This is the single-source dependency that most threatens Eoptolink.
- DSP / retimer → DUOPOLY. Broadcom + Marvell own the DSP. Eoptolink is reported to source DSPs from Broadcom; its Gen-2 1.6T OSFP uses a 3nm DSP that cuts module power ~20%.
- Silicon photonics / TFLN modulators — Eoptolink uses EMLs and its own SiPh + thin-film-lithium-niobate modulators, a partial hedge against the EML squeeze but still reliant on foundry capacity (Taiwanese foundries dominant).
Midstream (the company): optical design → packaging → assembly → test. Two manufacturing bases: Chengdu (China) and Sriracha, Chonburi, Thailand. The Thailand plant (Phase I live since H1-2023; Phase II in mass production from early 2025) is the AI-interconnect plant — built explicitly to serve North-American demand outside China export-control exposure and to ramp 1.6T.
Downstream (buyers): NVIDIA (modules into GB-class racks) and the hyperscalers direct (Amazon/Google/Meta/Microsoft buy modules directly, the model InnoLight pioneered).
Chokepoints: EML 200G/lane (single-source Lumentum) and DSP (Broadcom/Marvell duopoly) — Eoptolink owns neither. Its own moat sits in packaging, test yield, and qualification speed, not silicon.
Lens 3 · Competitive Advantages (moats)
- Cost + speed-to-qualify, not IP. The durable edge is being first through NVIDIA/hyperscaler qualification at scale and shipping at a structural cost advantage. InnoLight + Eoptolink together hold ~60% of NVIDIA's incremental 800G orders for 2025. Qualification is a real switching cost — re-qualifying a module vendor is slow and risky for a hyperscaler mid-build — but it is per-generation, not permanent. Each speed transition (800G→1.6T→3.2T) re-opens the bake-off.
- Best-in-class margins. Eoptolink runs the highest net margin in the sector (~33% in 2024, ~38% in 2025) vs InnoLight's ~20–22%. That is the strongest single tell that its product mix and execution are genuinely good — but margin leadership in a component business invites both customer price-downs and competitor entry.
- Bargaining power: weak on both sides. Upstream it is a price-taker on EML/DSP (NVIDIA out-bids it for the scarce lasers). Downstream its top-5 take 71% of revenue, so customers hold the whip. The moat is "be the cheapest qualified vendor for this generation," which is real cash today and fragile across cycles.
- Geographic moat (underrated): the Thailand base is a genuine structural advantage — it lets Eoptolink sell into US hyperscalers under tariff/export-control regimes that increasingly disadvantage China-made AI gear. That is a moat InnoLight shares but most domestic peers don't.
Lens 4 · Segments
Eoptolink does not publish clean product/geography EBITDA splits in any source I could reach — segment-level operating income is n/a. What is sourced:
- By geography: overseas 96.16% of FY2025 revenue — effectively a US-hyperscaler export business with a Chinese cost base.
- **By product (directional,
):** 800G is the 2025 revenue engine; 1.6T is the 2026 growth layer (samples to NVIDIA in H1-2025, volume ramping H2-2025→2026); 100G/200G/400G + telecom is the declining legacy tail. No segment file exists to ground this — flagged as a key open item.
- Trend: revenue accelerating violently then decelerating off a vast base — FY2024 +179% → H1-2025 +282.6% → 9M-2025 +221.7% → FY2025 +187.3%. The YoY comp is now lapping the 2025 explosion, and Q1-2026 already showed the first sequential profit decline (Lens 5).
Phase B — Measure performance
Lens 5 · Earnings Result
FY2025 (annual report): revenue ¥24.84B (≈$3.50B ) +187.29% YoY; net profit attributable ¥9.53B (≈$1.34B ) +235.89% YoY; gross margin 47.81%; overseas 96.16%.
Quarterly trail:
- Q1-2025: revenue doubled, "breaching ¥8B" trailing; 12-month revenue +220.9%.
- H1-2025: revenue ¥10.437B (+282.64%), net profit ¥3.942B (+355.68%).
- 9M-2025: revenue ¥16.505B (+221.70%), net profit ¥6.327B (+284.37%).
- Q3-2025: revenue ¥6.068B (+152.53%), net profit ¥2.385B (+205.38%); GM 46.94% (+5.41pp YoY), net margin 39.30%. But revenue dipped sequentially vs Q2 on "phased shipment pace"; net income +0.63% QoQ.
- Q1-2026 (the crack): net profit attributable −13% QoQ, missing consensus, driven by expanding FX losses. This is the first miss of the cycle and the proximate trigger of the 24-Apr selloff.
Balance-sheet / cash-flow flags: detailed working-capital, inventory, receivables and FCF data are n/a (no filing read). FX is the named earnings risk — a 96%-overseas exporter booking USD revenue against an RMB cost base is structurally short USD/long RMB on translation, and that is exactly what hit Q1-2026.
Market reaction: on 24 April 2026 the stock fell −11.67%, erasing ¥700B+ ($10.2B) of market cap in a day, dragging InnoLight and TFC with it ("Yi-Zhong-Tian" trio). The tape says the market was priced for flawless hyper-growth and punished the first FX-driven wobble brutally — a high-beta, sentiment-driven name.
Lens 6 · Earnings Calls (sentiment trend)
No transcripts in the research layer; A-share issuers don't run US-style quarterly calls, and English transcripts are sparse. From management commentary in secondary reporting, the management focus through 2025 was: 1.6T ramp progress ("smooth, volume picking up gradually in H2-2025"), Thailand Phase-II mass production, and capacity for 800G+ AI demand. The tone shift is the tell: through mid-2025 it was unqualified bullishness on demand; by Q1-2026 the narrative added FX headwinds and "phased shipment pace" hedging — the first defensive notes of the cycle, even as management kept reiterating a constructive multi-year compute-demand outlook. Treat all of this as ``, second-hand.
Lens 7 · Comps
Chinese optical-module peers. Multiples below are `` with source/date or n/a. I did not have a clean single source for a full EV/Sales · EV/EBIT · 5-yr-avg-ROE grid, so unsourced cells are marked honestly — do not read fabricated multiples into the blanks.
| Company | Ticker | Mkt cap | P/E | P/S | Net margin | Notes |
|---|
| Eoptolink | 300502.SZ | ~¥453–589B (≈$64–83B) | ~52x normalized | ~20.7x | ~38% FY25 | Highest margin; trades at ~693% premium to Morningstar FV ¥247 |
| Zhongji InnoLight | 300308.SZ | ~¥594B (≈$84B) | ~35x fwd | n/a | ~20–22% (2024) | Global #1 800G SiPh; ~50–60% of 1.6T share; SiPh mix >40% |
| Accelink | 002281.SZ | n/a | n/a | n/a | n/a | China's main vertically integrated (IDM) player — owns chips→modules |
| Coherent | COHR (US) | n/a | n/a | n/a | n/a | ~16% sector share; owns EML supply (vertical) |
| Lumentum | LITE (US) | n/a | n/a | n/a | EBITDA margin ~10–15% | Only volume 200G/lane EML supplier; NVIDIA-backed |
Read: Eoptolink and InnoLight carry comparable mega-cap valuations (~$64–84B) on ~35–52x earnings — China is pricing these two as the structural 1.6T winners. The Western names (Coherent/Lumentum) trade on the chip economics and capture more of the durable margin (they own the EML), which is the central irony of the whole chain: the module makers own 7 of the top-10 seats but the chip makers keep the money.
Lens 8 · Stock-Price Catalysts (>5% moves)
What actually moves this name:
- Earnings prints & sequential margins — the dominant driver. The 24-Apr-2026 −11.67% crash was an earnings event (Q1-2026 −13% QoQ on FX), not a demand event.
- NVIDIA order/qualification news — confirmation of 800G/1.6T allocation share spikes the stock; any hint of NVIDIA in-housing or re-allocating EML capacity (the $4B Lumentum/Coherent lock-up) pressures it.
- The "Yi-Zhong-Tian" basket beta — it trades as a 3-name complex with InnoLight + TFC; sector sentiment moves all three together.
- Regulatory / governance — the CSRC investigation and penalty of the actual controller (Lens 9/10) is an idiosyncratic overhang.
- FX / RMB — newly a first-order driver given 96% overseas revenue.
Pattern: a hyper-growth momentum stock that reacts to margin direction and NVIDIA headlines far more than to absolute demand, with violent drawdowns when priced-for-perfection meets any wobble.
Phase C — Judge people & books
Lens 9 · Management
- Track record: Chairman/actual controller Gao Guangrong built Eoptolink from a 2008 Chengdu startup to a top-3 global AI-interconnect supplier with sector-leading margins and a >$60B market cap — a genuinely strong operating record on the numbers. President Huang Xiaolei runs operations.
- Skin in the game: Gao is the actual controller (founder-led) — high alignment in principle. Specific insider-ownership % is
n/a (no insider-transactions.csv).
- Capital allocation: FY2025 — R&D ¥702M (≈$103M; notably light at ~2.8% of revenue for a "tech" leader, consistent with being an integrator not a chip-R&D house), plus a ¥10-per-10-shares cash dividend + 4-for-10 stock dividend from capital reserves. So: reinvest in Thailand capacity, modest cash return, no value-destroying M&A surfaced.
- Red flags (material): the actual controller was penalized by the CSRC in early 2026 — see Lens 10. This is a real governance black mark on the founder, not a peripheral issue.
- Archetype: founder-controlled operator — fast, execution-driven, capacity-aggressive, which fits a land-grab phase. The governance lapse is the shadow side of concentrated founder control.
Lens 10 · Forensic Red Flags
Accounting-risk surface for a module integrator (every figure /; no filing read):
- Revenue recognition / concentration: top-5 = 71.1% of revenue, 96% overseas — a small number of POs drives the whole P&L. "Phased shipment pace" language (Q3-2025) means revenue can be timing-shifted across quarters, which warrants scrutiny of cut-off and any channel build.
- FX exposure: confirmed-material — Q1-2026's miss was FX losses on USD revenue / RMB books. Watch hedging disclosure and whether FX is being parked below the line.
- Margin durability: 47.8% GM / ~38% net is extraordinary for a component assembler and is the number most likely to mean-revert as 1.6T competition and customer price-downs arrive. Not fraud — but the figure carrying the ~52x multiple is the most fragile.
- Working capital / SBC / goodwill: receivables, inventory, SBC, and intangible balances are
n/a — a real gap; cannot confirm whether cash flow tracks earnings without the filing.
Regulatory findings (required):
- SEC (EDGAR LR/AAER): none possible — Eoptolink has no CIK and does not file with the SEC. Zero SEC findings by construction, not by clean record.
- CSRC (the material finding) ``: China's securities regulator opened an investigation into actual controller Gao Guangrong in January 2026 for violating restricted-share-transfer rules, and finalized an administrative penalty in February 2026: ¥9.4986M of illegal gains confiscated + a ¥22M fine. Substance: via a family trust and his HTSC account he transferred 1.42% of total share capital, of which 0.42% breached restricted-transfer rules; separately, his failure to truthfully report holdings produced false records in the company's 2020, 2021 and 2022 annual reports. This is the single most important red flag in the file — the founder/actual controller has an adjudicated disclosure-falsification finding touching three years of annual reports. It does not implicate the operating numbers directly, but it materially lowers the trust you can place in governance and in disclosures generally.
- Non-SEC (FTC/DOJ/etc.): web search surfaced no US/other-agency enforcement against the company itself.
- Net: No SEC findings (no CIK); one material CSRC finding against the actual controller (Feb 2026, share-transfer breach + false annual-report records 2020–22) — verified via the pre-fetched regulatory file + web as of 2026-06-20.
Phase D — Project & stress-test
Lens 11 · Forward Projection
Bottom-up off FY2025 actuals (revenue ¥24.84B, net ¥9.53B, ~994M shares → FY2025 EPS ≈ ¥9.59 ). All forward lines ; consensus anchors .
Consensus anchors ``: sell-side sees earnings/revenue +35–38%/yr, net-income 3-yr CAGR ~55%; FY-ahead revenue est ~¥7.84B/quarter-run-rate region (figure ambiguous in source); 12-mo avg target ~¥575–638 vs ~¥581 spot.
| Scenario | FY2026e revenue | Net margin | Net profit | EPS (¥) | Logic |
|---|
| Bull | ~¥40B (+60%) | ~38% | ~¥15.2B | ~¥15.3 | 1.6T ramps clean, NVIDIA share held, EML allocation secured, FX neutral |
| Base | ~¥33B (+33%) | ~35% | ~¥11.6B | ~¥11.6 | Matches consensus ~+35% growth; mild margin give-back + FX drag |
| Bear | ~¥27B (+9%) | ~30% | ~¥8.1B | ~¥8.1 | 1.6T price war + EML rationing caps volume + FX losses persist → first down-EPS year |
At ~¥581 spot, base ~¥11.6 EPS ⇒ ~50x forward P/E — i.e. the stock is priced near the bull-ish end and leaves no margin of safety if the bear (margin reversion + FX) plays out. (Per --watchlist rules, no forecast.ts Brier forecast logged.)
Lens 12 · Bull vs Bear
Bull case. The AI-interconnect TAM is exploding — 800G+ transceiver shipments go 24M units (2025) → ~63M (2026), +2.6x; the optical-interconnect market doubles $5B (2024) → $10B+ (2026). Eoptolink is one of two Chinese vendors with ~60% of NVIDIA's 800G and a credible 1.6T roadmap (Gen-2 OSFP on 3nm DSP, LRO/LPO/FRO, even 6.4T NPO and 12.8T XPO demos at OFC 2026). It earns the best margins in the sector (~38%), ships 20–25% cheaper than the West, and has a tariff-insulated Thailand base. If 1.6T ramps as 800G did, FY2026–27 compounds fast and ~52x is "expensive but growing into it."
Bear case (permanent-impairment risks). (1) It owns neither chip. NVIDIA's $4B EML lock-up at Lumentum/Coherent can ration the scarce 200G/lane lasers Eoptolink needs for 1.6T, pushing non-favoured buyers past 2027 — a hard volume cap it cannot engineer around (domestic EML <15% localized). (2) Margin reversion. 47.8% GM in a module-assembly business is anomalous; customer price-downs + Chinese-on-Chinese 1.6T competition (InnoLight is #1 in SiPh and leads 1.6T share) drag it toward the sector's teens-20s. (3) Concentration + FX. Top-5 = 71%, 96% overseas — any single hyperscaler pause, plus RMB strength, hits revenue and the bottom line simultaneously (Q1-2026 already proved the FX leg).
Pre-mortem (18 months out, thesis broke): 1.6T volumes underwhelm because Eoptolink lost the EML-capacity race to NVIDIA's favoured Western suppliers; a hyperscaler capex air-pocket coincides with RMB strength; gross margin slid from 48% toward 30%; the ~52x multiple de-rated to ~20x on the first demand (not just FX) miss — a 50–60% drawdown from a name that already fell 12% in a day on FX alone.
Multiples too high? At ~52x trailing / ~50x forward base, yes on any normalization — the stock embeds sustained ~35%+ growth and ~38% margins indefinitely. Morningstar pegs it ~693% above a ¥247 fair value (an outlier, but directionally a valuation warning).
Contrarian view (what the market refuses to see): the consensus treats Eoptolink as a structural compounder; it is better understood as a cyclical, single-source-constrained, FX-exposed contract assembler at the mercy of two duopoly chip suppliers and a handful of hyperscaler buyers — extraordinarily profitable right now, but with a moat that resets every speed generation and a governance discount the market is currently ignoring.
Lens 13 · Devil's Advocate (short-seller)
- What breaks the money machine: the 200G/lane EML supply leash. NVIDIA already pre-bought the scarce lasers; if Eoptolink can't get enough, 1.6T volume is capped regardless of demand — and 1.6T is the 2026 story.
- Concentration: top-5 = 71%, NVIDIA + 2–3 hyperscalers effectively are the company. One re-allocation or one quarter's capex pause cascades straight to the P&L.
- Moat weaker than bulls think: the edge is per-generation qualification + price, not IP. InnoLight (bigger, SiPh #1, 1.6T share leader) is the more dangerous Chinese competitor; Coherent/Lumentum own the EML and capture the durable margin. Eoptolink is the least vertically integrated of the serious players.
- Worst capital-allocation / governance: the actual controller has an adjudicated CSRC penalty (Feb 2026) for restricted-share-transfer breaches and causing false records in three years of annual reports. For a foreign A-share you can't audit via EDGAR, founder disclosure-falsification is exactly the red flag that should widen the risk discount.
- What must hold for today's price: ≥35% growth and ~38% margins, every year, with EML supply secured and RMB benign. Miss any one and ~50x compresses hard.
- −20–30% growth shock: flips FY2026 to flat/negative EPS growth; at ~50x that is a multiple and earnings de-rate — easily a 50%+ drawdown (and it already lost ~$10B in a day on a milder FX shock).
- Single permanent-impairment scenario: NVIDIA + hyperscalers vertically integrate / co-package optics (CPO) and/or favour Western EML-owning vendors, structurally shrinking the merchant 1.6T+ pluggable pool Eoptolink sells into. Plausibility: medium over 3–5 yrs (CPO volume is "another two years" out per OFC 2026), low in the next 12 months.
Lens 14 · Management Questions (ordered by information value)
- What is your secured 200G/lane EML supply for 2026–27, by supplier, given NVIDIA's $4B capacity lock-up at Lumentum/Coherent — and what is your contingency if allocation is cut?
- What was NVIDIA's share of FY2025 revenue, and the top-5 customer list — and how concentrated is the 1.6T order book specifically?
- How sustainable is the 47.8% gross margin as 1.6T scales and Chinese-on-Chinese competition intensifies — what GM do you underwrite at steady state?
- What is your FX-hedging policy after Q1-2026's losses, and how much of the 96%-overseas revenue is naturally or financially hedged?
- What share of your 1.6T BOM is in-house (SiPh/TFLN modulators) vs bought EML, and what is the roadmap to owning more of the chip?
- How do you protect the merchant pluggable business against CPO / co-packaged optics adoption by NVIDIA and Broadcom over the next 3–5 years?
- What is the Thailand Phase-II capacity (units/yr by speed) and the China-vs-Thailand mix for US-bound shipments under current export controls?
- Where are you in NVIDIA/hyperscaler 1.6T qualification vs InnoLight, and what is your realistic 1.6T share target?
- Given the actual controller's CSRC penalty and the false-records finding for 2020–22, what governance and disclosure-control changes have you implemented?
- What is insider ownership today, and are there further planned share transfers/reductions by the actual controller or related trusts?
- What is your capital-allocation priority between capacity, the modest dividend, R&D (only ~2.8% of sales), and any chip-level vertical integration?
- How exposed are you to US export controls / entity-list risk, and does the Thailand base fully insulate US-hyperscaler supply?
- What is the 3.2T / next-gen roadmap and timing, and does the NPO/XPO demo work convert to volume orders?
- How should we read "phased shipment pace" — is there channel inventory or pull-forward we should normalize for?
- What single development would most change your own view of the 2027 outlook?