Rating: Cautiously Bullish | Target Price Range: RMB 330-420 | Current Price: RMB 306.01 (2026-10-15 close) | Margin of Safety: ~+8% (baseline fair value floor of RMB 330 vs. current price) | Time Horizon: 12 months
Market cap ~RMB 426.66 billion (total shares 1.39426 billion, after the 4-for-10 bonus share issue on 2026-06-11) | Sector: AI Datacom High-Speed Optical Modules (400G/800G/1.6T) | Exchange: ChiNext, Shenzhen Stock Exchange
Eoptolink is the second global powerhouse in the optical module industry: in 2025, optical module sales reached USD 3.5 billion (+189%), surpassing Coherent for the first time on the LightCounting rankings to become the world's No. 2, with deep ties to North American AI compute capital expenditure. In 2026H1, revenue was RMB 20.910 billion (+100.34%) and net profit attributable to shareholders RMB 7.529 billion (+90.98%); Q2 standalone net profit of RMB 4.749 billion hit a record high. However, the stock has retraced 49.9% from its 52-week high of RMB 610.49 to RMB 306.01, pressured by three successive waves: rumors of cloud providers' capex cuts (late July), export control speculation (early October), and profit-taking at elevated levels (mid-October). At the current price, the stock trades at only ~11.8x the sell-side consensus 2027E (net profit of ~RMB 36.2 billion) — effectively, the market is pricing in "2027 as the profit peak." Our baseline view is that 2027 is not the peak: per Goldman Sachs, shipments of 800G and above are still +84% in 2027E and +19% in 2028E, and the company's order visibility and incentive plan targets (cumulative 2026-2028 revenue of RMB 290 billion) point to sustained volume growth. Baseline fair value: RMB 330-420 (+8% to +37% vs. current price). Constraints are equally clear: the margin of safety is thin; export control expansion, customer concentration (top five clients at 72.34%), share-dilution signals from growth lagging Innolight (+182%) and the industry (+120%), and the impairment elasticity of RMB 11.655 billion in inventory together form the downside structure, with a bear-case scenario of RMB 140-185. We are cautiously bullish, with two key validation points in the coming month: the Q3 report on 10-30 (Q3 standalone QoQ direction) and the final scope of export control policy text.
Nature of the opportunity: A high-growth realization period colliding with a "peak concern" discount — the essence of this opportunity is not discovering unrecognized growth, but that the market's pricing of "whether 2027 is the profit peak" is significantly more conservative than the verifiable demand mathematics of the industry chain; evidence grade B (core financial data verified against primary financial reports; industry volume/price forecasts rely on LightCounting/Goldman Sachs/TrendForce institutional figures via secondary citation, some via media reports).
Most important marginal changes in the past 90 days: ① The FCC equipment authorization new rule was published in the Federal Register on 2026-09-10 and did not include Eoptolink (partially removing the original "control rumor" tail; the remaining tail now shifts to the export control expansion path); ② Three waves of decline pushed the stock from RMB 610.49 to RMB 306.01; on 10-15 the Chairman proposed an RMB 800 million-1 billion share buyback (~0.2% of market cap; signaling significance exceeds substance); ③ 1.6T ramped up from 2026Q2, with the company stating H2 volume is accelerating and "demand for optical modules in 2027 is already fairly clear" (verbatim from the 2026-07-19 earnings call); ④ Sell-side consensus EPS was revised down 22.1% over the past 90 days, with expectations converging along a "net margin mean reversion" path.
Primary share price drivers (by importance): ① Earnings momentum — 1.6T ramp pace and Q3/Q4 standalone revenue QoQ (Q2 base of RMB 12.572 billion); ② Policy tail risk — the scope and exemption clauses of export control text; ③ Demand slope — North American cloud providers' 2027 capex guidance and the direction of industry shipment forecast revisions.
Verifiable expectation gap: Current price implies only 11.8x consensus 2027E. This reading has two equivalent interpretations — "earnings will be sharply revised down" or "pricing existing consensus at cyclical peak multiples (~12x)." If Q3 standalone QoQ growth is ≥+10% and net margin holds at 35%, combined with export controls limited to the 3.2T generation (not ramping until 2028-2029), "peak pricing" will be falsified by the data, leaving room for a re-rating toward 17-19x. Reverse falsification conditions: expansion to 1.6T, standalone revenue turning QoQ negative, or gross margin declining QoQ for two consecutive quarters by ≥1 percentage point.
Key evidence:
The volume evidence is solid: NVIDIA data center revenue reached USD 96.2 billion in FY27Q2 (+106%), with FY27Q3 guidance at USD 108 billion; optical modules per GPU rise with each generation (800G-generation scale-out ~1:2.5-3, 1.6T-generation networking ~1:4-6, per broker estimates), and each speed doubling raises per-unit value by 2-3x. On price, disagreement must be acknowledged: per TrendForce, ~70% of 2027 capex growth flows to memory, meaning optical module revenue growth may lag overall capex growth; EML optical chip shortages (Lumentum's 2027 capacity already fully booked under long-term agreements, orders covering through 2029) are both a price-support logic and a shipment ceiling plus cost pressure. This thesis is used under a conservative framing of "certain volume growth, uncertain price elasticity": downside risks tracked via 800G/1.6T annual negotiated prices and quarterly gross margin QoQ; if LightCounting's price warning materializes while Goldman Sachs misses, the earnings assumptions in C2/C3 must be revised down in tandem.
Key evidence:
The path to RMB 50 billion revenue in 2026 is clear: H1 delivered RMB 20.910 billion; H2 needs only +39.1% QoQ (the Q2 standalone QoQ slope of +50.8% shows no decay); full-year net profit of RMB 18-20.5 billion implies a net margin of 34-36% (H1 actual: 36.0%; Q1 was dragged to 33.3% by RMB 522 million in FX losses, recovering to 37.8% in Q2). Constraints must also be faced squarely: 2026H1 growth (+100%) lagged Innolight (+182%) and industry Ethernet growth (~+120% in 2026E); third-party estimates put Innolight's 1.6T share above 50% — generational first-mover advantage sits with the rival, so share assumptions cannot be linearly extrapolated; the equity incentive plan target (2026 revenue ≥ RMB 50 billion) is a pure revenue metric with no profitability constraint, granted at a 31% discount (RMB 212/share), and can only serve as a revenue signal, not an earnings commitment. Tracking signals: Q3 standalone revenue QoQ (≥+10% would beat expectations; neutral scenario ~RMB 13.8 billion, ~+127% YoY), and whether cumulative 9M revenue reaches RMB 37 billion (74% progress toward the RMB 50 billion line).
Key evidence:
The 11.8x forward multiple admits two mathematically indistinguishable readings: the market believes 2027E earnings will be sharply revised down (equivalent to pricing at 18-22x for RMB 19.4-23.7 billion, i.e., "zero growth"); or the market is pricing existing consensus at cyclical peak convention (~10-15x) — RMB 36.18 billion × 12x ≈ RMB 311, nearly identical to the current price. Our baseline judgment is that the "2027-as-peak" premise underlying the latter does not hold: Goldman Sachs figures show 800G+ shipments still +19% in 2028; the incentive targets imply RMB 150 billion revenue in 2028; NPO/OCS ramp from 2027H2-2028 — i.e., there are still two years of volume growth beyond 2027. Under this judgment, 2027E net profit of RMB 27-31 billion × 17-19x exit multiple (see valuation chapter), baseline fair value RMB 330-420. If the peak judgment is wrong (2028 capex peaks, price cuts deepen), the current price is not cheap — this is precisely why the margin of safety is only ~8%, and why this report rates only "cautiously bullish."
Key evidence:
Triple downside constraints stacked: ① Policy — if export controls expand to 1.6T and below, or Thai fabs fail to obtain exemptions, U.S.-bound revenue (estimated at 50-70%, a stitched estimate) would take a material hit; ② Concentration — 96% of revenue is overseas, demand tied to a few North American cloud providers; on 2026-07-30 a single unverified rumor alone triggered a one-day -11.89%; ③ Financial elasticity — inventory of RMB 11.655 billion (31.8% of total assets) carries low provisioning (4.90%) during a price upcycle; if LightCounting's price warning materializes, incremental write-downs would far exceed the apparent ratio (a 20% demand cut scenario implies ~RMB 990 million in after-tax impairments, roughly one quarter's profit). The bear case is constructed on "control expansion + deeper price cuts + inventory top-ups + net margin falling to 25-30%": 2027E net profit of RMB 15.5-18.5 billion × 12.5-14x (anchored to the actual trough multiples of 12.3-16x from the last cycle), fair value RMB 140-185, -40% to -54% vs. current price.
| Metric | 2023FY | 2024FY | 2025FY | 2026H1 (latest period) |
|---|---|---|---|---|
| Revenue (RMB 100M) | 30.98 | 86.47 | 248.42 | 209.10 |
| Revenue YoY | — | +179.15% | +187.29% | +100.34% |
| Net profit attributable to shareholders (RMB 100M) | 6.88 | 28.38 | 95.32 | 75.29 |
| Net profit YoY | — | +312.26% | +235.89% | +90.98% |
| Non-GAAP net profit (RMB 100M) | 6.78 | 28.30 | 95.07 | 75.11 |
| Gross margin | 30.99% | 44.72% | 47.81% | 48.46% (optical interconnect basis) |
| Net margin (attributable NP/revenue) | 22.22% | 32.82% | 38.37% | 36.01% |
| Net operating cash flow (RMB 100M) | 12.46 | 6.41 | 77.01 | 16.16 |
| Capex (RMB 100M) | 5.54 | 14.76 | 13.20 | 12.64 |
| Free cash flow (RMB 100M, OCF − capex) | 6.92 | -8.35 | 63.81 | 3.52 |
| Cash and equivalents (period-end, RMB 100M) | 25.15 | 16.00 | 81.56 | 72.40 |
| Interest-bearing debt (period-end, RMB 100M) | ≈0 | ≈0 | 0.15 | 0.24 |
| Debt-to-asset ratio | 15.12% | 32.11% | 30.20% | 33.23% |
| Inventory (period-end, RMB 100M) | 9.63 | 41.32 | 72.34 | 116.55 |
| Accounts receivable (period-end, RMB 100M) | 7.15 | 25.39 | 44.38 | 75.48 |
| Contract liabilities (period-end, RMB 100M) | 1.38 | 0.97 | 0.91 | 2.17 |
| Top five customer concentration | 60.66% | 71.10% | 72.34% | Not disclosed |
Note: Balance sheet stock items are all period-end figures; the latest period is as of 2026-06-30; restricted cash is minimal (only RMB 56 million restricted); the company has no interest-bearing borrowings (only lease liabilities).
Reasons for metric changes (moves of YoY ≥ ±20% and company explanations): 2025 revenue +187.29% — the company explained "continued industry upturn and a substantial increase in global demand for high-speed optical modules," with unit sales of 16.03 million (+82.78%); 2025 net profit +235.89% — revenue ramp-up + gross margin +3.09pp (product mix changes and improved production efficiency) + FX gains of RMB 244 million; 2025 operating cash flow +1101.57% — "increased sales collections from business growth" (2024 was -48.56%, explained as stockpiling and tax payments); 2026Q1 contract liabilities +222.32% — increased advance payments from customers; 2026H1 financial expenses of RMB 644 million (vs. -RMB 208 million in the prior-year period, +409.74%) — increased FX losses (USD asset exposure of ~RMB 11.4 billion); inventory at end-2026H1 +61.1% and receivables +70.1% vs. year-start — proactive stockpiling "considering on-hand and expected orders and the procurement cycle"; 2025 domestic revenue -48.19% — the company focused on large global cloud service customers as domestic demand has yet to ramp up.
Q2 standalone revenue was RMB 12.572 billion (+96.9% YoY, +50.8% QoQ), with net profit of RMB 4.749 billion (+100.4% YoY, +70.8% QoQ) and a standalone net margin of 37.8% — the strongest quarter since listing: volume contributions came from continued 800G ramp and 1.6T starting to ship in Q2; margin recovery came from Q1 FX losses (RMB 522 million) narrowing sharply in Q2. H1 gross margin (optical interconnect) was 48.46%, +0.99pp YoY, holding steady and even rising despite EML price increases, reflecting generational mix (rising 1.6T share) and scale effects offsetting cost pressure. The balance sheet corroborates order confidence: inventory of RMB 11.655 billion (+61.1% vs. year-start) plus prepaid materials/equipment totaling ~RMB 2.675 billion represent stockpiling under lock-in agreements for scarce materials; contract liabilities of RMB 217 million remain elevated. Against management commentary (high Q3/Q4 order visibility, 1.6T ramp accelerating in H2, clear 2027 demand), H1 results were "basically in line" with the company's beginning-of-year expectations (company's own assessment). No standalone-quarter consensus estimate is available, so we draw no conclusion on "beat/miss"; the direction of Q3 standalone QoQ (~RMB 13.8 billion in the neutral scenario) is the next falsifiable checkpoint.
Business model in brief: An order-driven manufacturing business with ~98% direct sales and 96% overseas revenue; core products are high-speed pluggable optical modules for AI datacom. No subscription or recurring revenue (performance obligations at end of 2025 correspond to zero revenue). Revenue is dual-driven by cloud providers' capex cycles and product generations (400G→800G→1.6T, with per-unit value jumping roughly 2-3x). Light on intangibles (goodwill + intangibles combined ~1.5% of total assets); production capacity spans a dual-base setup in Chengdu + Thailand.
Cash content of earnings test: OCF/net profit attributable to parent for 2023-2026H1 sequentially at 1.81→0.23→0.81→0.21; FCF/attributable profit at 1.00→-0.29→0.67→0.05. The pattern is clear: in years of high profit growth, cash conversion weakens instead — in 2026H1 the combined increase in inventory + receivables + prepayments was ~RMB 10.1 billion, nearly equal to the H1 revenue increase, with expansion heavily consuming working capital. This is the classic profile of an expansion phase where "profits are real but eaten by reinvestment": with order-driven volume ramping, advance stocking is necessary to lock in scarce EML chips and capacity, which cannot simply be equated with deteriorating earnings quality. However, it means the current 36% net margin does not convert fully into distributable cash, and valuation should apply a cash discount to "book profit." Positives: non-GAAP/attributable profit has long been ≈100% (99.8% in 2026H1), SBC is only 0.6% of attributable profit, and there is no room for aggressive Non-GAAP adjustments — the earnings caliber is clean.
Return on capital: 2025 ROIC ~73% (NOPAT = total profit of RMB 10.866 billion × (1 - effective tax rate of 12.08%) ÷ average invested capital of ~RMB 13.2 billion), far above WACC; ROE rose from 13.4% in 2023 (trough of the last cycle) to 72.6% in 2025. It must be noted this includes cyclical windfall and tax-benefit components: the Thai subsidiary enjoys BOI tax exemption (2025 revenue RMB 21.9 billion, net profit RMB 1.409 billion), and the Western Development 15% preferential rate contributed ~RMB 197 million in tax savings. The company expects to enter the global minimum tax (Pillar Two) framework from fiscal 2026, so the 12.08% effective tax rate may rise.
Capex quality: capex/D&A was 7.0x in 2024 and 3.3x in 2025; 2026H1 capex was RMB 1.264 billion (3.4x the prior-year period's ~RMB 370 million), directed at Thailand Phase II and Chengdu expansion — primarily growth investment. The company does not disclose a maintenance/growth split, so maintenance outlays cannot be precisely isolated, but fixed assets + construction-in-progress growth (+22.5%/+70.5% vs. start of year) aligns directionally with order growth; no "capital black hole" characteristics for now.
Moat & quality red flags: The real advantages are cost efficiency and delivery speed — net margin of 36.0% above Innolight (~32.7% in 2026H1), delivery capability demonstrated by headcount expanding from ~2,000 to 8,000+ over three years, status as a key Amazon supplier plus NVIDIA qualification in 2025. Red flags: customer concentration (top five at 72.34% and rising year by year), single-point upstream dependence (largest supplier at 23.87% of purchases; EML optical chips outsourced), and ~RMB 11.4 billion of USD exposure with no hedging arrangements in place (FX swung from gains to large losses in 2026).
Track record of walking the talk: pragmatic (operational level). ① Thai capacity: the 2024 annual report pledged "further expansion of the overall scale of the Thailand factory" → the 2025 annual report confirmed Phase II came online in early 2025, and the Thai subsidiary posted 2025 revenue of RMB 21.909 billion — delivered. ② Capacity & 1.6T: the 2025 annual report planned to "continuously increase 800G and 1.6T capacity" → 2026H1 fixed assets +22.5% vs. start of year, capex of RMB 1.264 billion, 1.6T already shipping in volume — in progress. ③ Earnings guidance caliber: on the 2026-07-19 call, management self-assessed "Q2 results broadly in line with beginning-of-year expectations" → Q2 attributable profit of RMB 4.749 billion set a single-quarter record — delivered. The company never gives quantitative guidance, IR responses are highly templated, with no record of "over-promising and under-delivering" — nor of over-promising at all.
Shareholder friendliness: neutral to weak. Payout ratios for FY2023-2025 were 15.96%→11.24%→10.45% in sequence, declining year after year, with retained cash of RMB 7.2 billion+; no history of cancellation-style buybacks; a RMB 1.65 billion private placement in 2020. Positive signals came in two 2026 actions: a restricted stock incentive plan binding the core team (granted at RMB 212/share; first grant of 5.1852 million shares completed on 2026-09-07) and a chairman-proposed RMB 0.8-1.0 billion buyback on 2026-10-15 (price cap ~RMB 459, for employee incentives) — but the buyback equals only ~0.2% of market cap, and the 31%-discounted incentive grant sparked pricing controversy at the draft stage. Overall verdict: strong operational execution, weak shareholder-return awareness; governance improvement is at the "signal shown, awaiting verification" stage.
Risk signals: ① Controlling shareholder/Chairman Gao Guangrong improperly transferred 1.42% of shares in March-April 2023, and family trust shareholdings were not truthfully reported, causing false statements in the shareholder disclosures of the 2020-2022 annual reports; in February 2025 the CSRC imposed fines and confiscations totaling ~RMB 31.4986 million — a major governance flaw. ② In September-October 2025, a further ~1.15% of shares was transferred via negotiated placement, with media estimating cash-out of ~RMB 3.7 billion. ③ Related party Chengdu Hongchen Photonics (with shareholding by relatives of the controlling shareholder and general manager) transacted ~RMB 136 million of related-party purchases with the company in 2025, an over-2,000x increase from less than RMB 60,000 in the prior period, precisely amid chip shortages; IR did not directly address the inquiry. These items constitute a substantive source of governance discount and may be amplified by international investors in the H-share offering pricing.
| Segment | Revenue share | Gross margin | YoY | Business logic in one line |
|---|---|---|---|---|
| Optical interconnect products | 99.72% (2025FY) | 47.81% | +188.07% | High-speed AI datacom modules; absolute profit engine |
| Others (materials, etc.) | 0.28% | ~45.6% (est.) | +47.93% | Negligible share |
| Overseas | 96.16% (2025FY) | 48.36% | +251.03% | Direct sales to North American cloud providers; supply shortage + generational dividend |
| Domestic | 3.84% | ~34.01% (est.) | -48.19% | Domestic AI demand not yet ramped; deliberate contraction |
The profit engine is overseas optical interconnect products (96.16% revenue share × 48.36% gross margin; absolutely dominant contributor to company gross profit). Overseas unit sales in 2025 were 12.926 million units at an ASP of ~RMB 1,545/unit (estimated from disclosed volumes); in 2026H1 overseas ASP rose further to ~RMB 2,103/unit — the structural upgrade to 1.6T and 800G is directly lifting per-unit value. The overseas vs. domestic gross margin gap is ~14.4pp (≥10pp): overseas is direct sales to supply-constrained AI cloud providers (generational dividend + qualification barriers), while domestic is a more competitive market — the pricing-power differential for the same product across two markets is clear. The company does not disclose revenue mix by speed (400G/800G/1.6T); management's IR line is "800G is the main shipping product, 1.6T is ramping continuously, silicon photonics share is rising notably" — the only public window for tracking share and generational transition; given limited precision, this needs to be probed repeatedly in quarterly reports and earnings calls.
Accounting red flags: ① Surge in related-party purchases (severity: medium) — purchases from Hongchen Photonics rose from RMB 59,576 in the prior period to RMB 136 million (over 2,000x increase), with a period-end payables balance of RMB 45.42 million; IR did not directly address pricing fairness (2025 annual report related-party transaction note). ② New large "long-term employee compensation payable" without detail (medium) — jumped from 0 to RMB 816 million in 2025 and to RMB 1.201 billion at end-2026H1; the notes do not disclose the accrual basis or beneficiaries, impairing verifiability of the liability structure. ③ Declining impairment provision ratio amid surging inventory (low) — 2025 inventory balance +73.5% while the provision ratio fell from 7.67% to 6.86%, and further to 4.90% at end-2026H1; a demand reversal would amplify downside earnings elasticity. ④ Effective tax rate of 12.08% heavily reliant on tax incentives (low) — Thai BOI exemption + Western Development preference total ~RMB 197 million; upside risk once Pillar Two takes effect.
Cross-period consistency: ① OCF/attributable profit (1.81→0.23→0.81→0.21) matches the stocking cadence; management's explanations have been consistent. ② Top-five customer share (60.66%→71.10%→72.34%) rising each year while the largest customer fell from 36.79% to 22.97% (customer diversification but rising overall concentration), consistent with the company's "customer structure optimization" narrative. ③ FX gains boosted profit for three consecutive years (-1.08→-1.79→-3.18 hundred million RMB) before swinging to a large loss in 2026Q1 (+5.22 hundred million RMB) — the company explained only the period's direction, without detailing management or hedging of the ~US$1.14 billion exposure; a "company unexplained" item. ④ R&D expense ratio trending down (4.32%→4.66%→2.83%→2.10%) — R&D spending growth persistently below revenue growth; the company attributes this to scale effects, but in the context of generational competition (rivals' R&D spend ~2.4x ours), it warrants ongoing monitoring.
Overall verdict: No evidence of earnings-manipulation-grade accounting tactics; the earnings caliber is clean (non-GAAP ≈ attributable, minimal SBC). But the surge in related-party purchases and the undisclosed-item liability require annual-report-inquiry-level ongoing verification, and the low inventory provision is the largest financial-statement risk point if the cycle reverses.
Current price RMB 306.01 (close of 2026-10-15), market cap RMB 426.66 billion. PE(TTM) 32.52x (TTM attributable profit RMB 13.119 billion) — 0th percentile over 1 year, 9th over 3 years, 34th over 5 years; static PE 44.78x; PB 17.57x (75th percentile over 5 years, to be read alongside 72.6% ROE); PS(TTM) 12.08x. During an earnings explosion, the PE percentile is mechanically depressed by high growth; the "one-year-low percentile" mainly reflects a surging denominator rather than cheapness, and percentiles serve only as a floor reference — this name has hit a structural demand inflection, so the primary valuation anchors are forward PE/PEG and ramp scenarios: 2026E forward PE ~20.7x (consensus attributable profit RMB 20.6 billion) / 22.5x (own build RMB 19.0 billion); 2027E forward PE ~11.8x (consensus RMB 36.2 billion) / 15.0x (own build midpoint RMB 28.5 billion); PEG ~0.27-0.39 (forward PE ÷ 2027E growth, cross-check only).
Peer benchmarking (as of 2026-10-15):
| Company | PE(TTM) | Forward PE | Latest growth | ROE |
|---|---|---|---|---|
| Eoptolink | 32.5x | 2026E 20.7x / 2027E 11.8x (consensus) | 2026H1 revenue +100.3% | 72.6% (2025FY) |
| Innolight | 37.8x | 2026E ~26x / 2027E ~17x (est.) | 2026H1 revenue +182.5% | 39.2% (2026H1) |
| Coherent | 71.0x (GAAP distorted) | forward ~30.9x | FY2026 revenue +22.5% | 8.5% (FY2026) |
| Fabrinet | 30.1x | — (TTM basis) | FY2026 revenue +35.7% | 21.3% (FY2026) |
Note: Coherent's GAAP figure is distorted by acquisition amortization and restructuring; the market actually prices it at ~30.9x forward. Fabrinet at ~30x with +35-40% growth serves as a reference for "steady-growth valuation" in North American optical interconnect.
Market cap of RMB 426.66 billion ÷ 2027E consensus attributable profit of RMB 36.18 billion = 11.8x. Equivalent reading one: the market is pricing 19.4-23.7 billion of profit at 18-22x (zero profit growth in 2027). Equivalent reading two: the market is pricing current consensus at the customary cyclical-peak multiple (~12x; 36.18 billion × 12 ≈ RMB 311 ≈ current price). The two readings are mathematically indistinguishable; the difference lies in whether the market disbelieves the "earnings" or the "multiple." Our judgment: the current price requires zero profit growth in 2027 or a peak discount, whereas reality shows industry shipments +84% (Goldman Sachs), sell-side consensus +75.3%, company profit growth of +235.9%/+91.0% over the past two years, and clear order visibility per management — "peak pricing" is mathematically inconsistent with demand. At the same time we acknowledge the market's pessimism has reasons: consensus EPS has been cut 22.1% over the past 90 days (net margin mean-reversion path), ~70% of the 2027 capex increment flows to memory, and discussion of CPO eroding pluggable share post-2028 is heating up. Two-way check conclusion: demand math supports the high growth of 2026-2027; whether it supports the duration beyond 2028 is the crux of the debate — hence the valuation conclusion must be presented as a scenario distribution, not a single-point target.
Zero-growth EPV is ~RMB 132.9/share: normalized FCFF ~RMB 18.7 billion (base = 2026E attributable profit of RMB 19.0 billion; bridge: + D&A ~RMB 1.2 billion − maintenance capex ~RMB 1.5 billion − zero-growth working capital change of 0) ÷ WACC 10.5% + net cash of RMB 7.216 billion. Book asset value ~RMB 17.4/share (including RMB 5.18/share of net cash) — under an asset-light model, purely formal. About 43.4% of the current price is supported by zero-growth EPV and ~56.6% is growth option — a high growth-option share does not in itself constitute an overvaluation conclusion (true of most growth stocks): this option has been validated by two prior inflections at 400G and 800G, not pure narrative; whether it deserves this price is answered by the ramp scenarios below. EPV here serves as a downside floor reference (in a cyclical drawdown the floor itself shrinks — ROE fell to 13.4% in 2023; beware the floor's cyclicality).
| Scenario | Probability | Fair value range | Construction |
|---|---|---|---|
| Bear | 0.22 | RMB 140-185 | Export controls extended to 1.6T, or 2027 capex downshift + deepening annual price cuts + inventory top-ups: 2027E attributable profit RMB 15.5-18.5 billion × 12.5-14x (anchored to last cycle's trough realized multiple of 12.3-16x) |
| Base | 0.53 | RMB 330-420 | 2027 revenue RMB 80-95 billion (industry shipments +84% × share ±2pp × ASP annual decline 8-12%) → attributable profit RMB 27-31 billion × exit PE 17-19x |
| Bull | 0.25 | RMB 520-645 | Continued capex upgrades + 1.6T share recovering to 16-17%: 2027E attributable profit RMB 33-36 billion × 22-25x; range covers Goldman's RMB 633 and consensus target median of RMB 600 |
Exit multiple anchoring rationale (base case 17-19x): ① Innolight 2027E forward ~17x (estimated in this report from its market cap and own-build earnings) — the current pricing of the other half of the duopoly; ② Fabrinet TTM 30x, Coherent forward 30.9x — North American steady-growth valuation references; ③ well below the optical module sector's 5-year median PE of 44x — pricing in the combination of "high growth delivered but the market persistently applying a cyclical discount." Why not 10-15x (peak convention): the base case assumes ~+15-20% growth still in 2028 (Goldman: 800G+ shipments +19% in 2028; incentive targets imply 2028 revenue of RMB 150 billion; NPO ramping from 2027H2 and scaling in 2028) — if that assumption is falsified, the exit multiple should converge toward the bear case, which is precisely the essence of the bear/base divide. Two discipline notes: we did not anchor to the company's current trading multiple (current 11.8x 2027E is for reference only); the ramp calculation (2027E earnings × exit multiple) is fully consistent with the scenario table. Transition zone: between the two tiers lies a "growth downshift" path (2027E attributable profit RMB 21-24 billion × 12-15x ≈ RMB 180-260), whose probability is embedded within the bear/base boundaries — a fall from RMB 306 into that zone would mean base-case failure, not a "cheaper" add-on opportunity.
Own earnings forecast (reconciliation anchor):
| Period | Revenue | Attributable net profit | Key assumptions |
|---|---|---|---|
| 2026E | RMB 51-56 billion | RMB 18-20.5 billion | 800G mainstay + 1.6T ramping in H2; gross margin 46-48%; net margin 34-36%; FX disturbance narrowing |
| 2027E | RMB 80-95 billion | RMB 26-32 billion | Industry shipments +84%; share 15%±2pp; ASP annual decline 8-12%; net margin 32-34% |
Comparison: management gives no quantitative guidance (incentive targets of RMB 50/90/150 billion revenue for 2026/2027/2028 are pure revenue metrics); sell-side consensus is 2026E RMB 20.6 billion / 2027E RMB 36.2 billion — our 2027E is ~6-28% below consensus, mainly due to a more conservative net margin assumption (32-34% vs. the current 36%); if profitability holds, consensus is achievable. Reconcile against this table at the next earnings report.
Undervalued (margin of safety ~+8%, vs. base-case fair floor of RMB 330), with quality and price assessed separately: quality is "strong but cyclical" (global #2 + 72.6% ROE + peak net margin, but customer concentration and profits containing cyclical windfalls); price sits in the "peak-fear" pricing zone — if 2027 growth delivers, it is clearly undervalued; if the peak thesis holds, it is merely fair. Probability-weighted center ~RMB 380 (~+24% vs. current price); upside to base-case center +22.5%, to bear-case center -46.9%, to bull-case center +90.4% — odds skewed upward but with a fat left tail. Two additional discount factors: ① the H-share offering (board-approved; up to 8% of post-issue share capital + 15% greenshoe) — if priced at a discount, mechanical dilution of ~7-9% per share, not yet reflected in the fair range (timetable and pricing undetermined; tracked as a risk item); ② governance discount (controlling shareholder's penalty history + related-party purchases) constrains the slope of valuation repair.
Global optical module sales exceeded USD 23 billion in 2025 (+50%, LightCounting 2025-12 quarterly update, incl. ~USD 1.1 billion of AOCs), of which Ethernet optical modules (AI datacom being the absolute majority) accounted for ~USD 17 billion, +82%; the 2024 total was ~USD 15.3 billion. Growth forecasts are still being revised rapidly upward: LightCounting raised its 2026 Ethernet growth forecast from +65% to +73% within the year (2026-09), and in 2026-10 further raised it to +120% under the latest basis (corresponding to ~USD 37.4 billion). Further out: LightCounting forecasts Ethernet optical modules reaching USD 80 billion in 2031; Yole forecasts datacom optical modules reaching USD 112.3 billion in 2031 (2025–2031 CAGR ~30%); Goldman Sachs forecasts the 800G-and-above segment at ~USD 45.2 billion in 2026 → USD 129.5 billion in 2028 (CAGR ~69%).
Quantified chain of the demand inflection (drivers × units per card × penetration × share): Driver variables — NVIDIA data center revenue FY26Q3 USD 57 billion → FY27Q1 USD 81.6 billion (+85%) → FY27Q2 USD 96.2 billion (+106%) → FY27Q3 guidance USD 108 billion; the nine major cloud providers' 2026 capex exceeds USD 957 billion (+106%), 2027 to exceed USD 1.5 trillion (+60%) (TrendForce, 2026-10-14). Units per card — the number of optical modules per accelerator card jumps by generation: at 800G (GB200/GB300, in-rack scale-up over copper), scale-out is ~1:2.5–3; at 1.6T (Rubin platform), two-tier networking ~1:4, three-tier ~1:6 (broker/media estimates, moderate reliability); doubling of speed simultaneously lifts unit value 2–3x (1.6T unit price is roughly 2–3x that of 800G). Penetration pace — 800G shipments ~7.5 million units in 2024 → 18–20 million in 2025 → more than doubling in 2026E; 1.6T ramping to the million-unit level in 2025 → ~8.6 million units on a sales basis in 2026E (LC) / 25–32.8 million units on a demand basis (Eoptolink estimate/Goldman Sachs) → becoming the shipment mainstay in 2027 (800G+ combined 144 million units, +84%). Cross-check: Goldman Sachs 800G+ shipments 2026E 78.22 million units (+166%) → 2027E 144 million (+84%) → 2028E 171 million (+19%). Company share — Eoptolink's 2025 optical module sales of USD 3.5 billion (+189%), ~15% of the total (backed out from 3.5/23.4); 25–30% share in 800G (third-party estimates); the company self-describes as "800G/400G/200G/100G Market Leader" (official IR), which differs in wording from the third-party "global No. 2, behind Innolight" — both are presented side by side, with the third-party estimate taken as authoritative. The chain's key vulnerabilities lie in pricing and duration beyond 2028: under TrendForce's basis, ~70% of the 2027 capex increment flows to memory (DRAM+NAND share rising from 47% to 68%), so optical module revenue growth may lag total capex growth; LightCounting's (2026-04) warning of prices "falling sharply before end-2026" directly conflicts with Goldman Sachs' "1.6T holding above USD 700, 800G declining low-single digits annually" — under consensus volume growth, price assumptions are the industry forecast's biggest point of divergence.
Upstream — InP EML/DFB laser chips are the tightest link across the industry: Lumentum's EML shipments are 25–30% below demand, 2027 capacity is fully booked by non-cancellable long-term agreements, and orders extend to 2029 (company basis), with LTA prices of ~USD 7.5–9.5 per chip; Mitsubishi Electric is the second supplier; on domestic substitution, Yuanjie Semiconductor's 100G EML is in volume production and 200G is in qualification. Midstream — module packaging and testing (the segment where the company operates); core competitive factors are silicon photonics/EML solution integration, automated capacity and yields, and speed of customer qualification response; silicon photonics penetration ~47% in 2025 → ~85% by 2031E (Yole). Downstream — the NVIDIA ecosystem and the five hyperscale cloud providers (the ultimate payers); Broadcom's AI order backlog exceeds USD 73 billion. Value distribution: during the shortage period, gross profit is concentrated at both ends — "upstream scarce chips + midstream modules"; the company has relatively weak bargaining power over upstream (must lock volume via LTAs + prepayments, ~RMB 2.675 billion of prepaid materials/equipment in 2026H1) but is episodically strong versus downstream (gross margin rose to 48%, a record high, amid supply shortage), though cloud providers' centralized procurement and the custom of per-generation annual price cuts imply risk of gross margin reversion when the cycle turns.
Supply is highly concentrated among Chinese leading vendors: LightCounting's 2025 global Top 10 includes 7 Chinese companies with a combined share of 60%+ (over 70% at 800G and above); brand-basis CR2 is ~38% (Innolight USD 5.33 billion + Eoptolink USD 3.5 billion). Current supply-demand: significantly undersupplied but easing at the margin — LightCounting stated in 2025-12 that demand exceeded supply by more than 2x for most products, revised in 2026-04 to demand exceeding supply by ~30%, with the shortage expected to ease by end-2026; the 1.6T gap is sharpest (2026 demand of 25–32.8 million units vs effective capacity of ~15 million units). Entry barriers: 1–2 year cloud provider/NVIDIA qualification cycles, EML LTA lock-ups (new capacity booked out to 2027–2029), automated capacity at the hundred-million-unit scale with yield ramping, and R&D plus in-house silicon photonics capability under rapid iteration. Competition and substitution threats: within the duopoly — in 2026H1 Innolight revenue +182% vs Eoptolink +100%; Innolight has re-opened the gap with first-mover 1.6T volume (third-party share >50%), with 800G share Innolight 40%+/Eoptolink 25–30%; an industry-wide expansion wave of ~RMB 40 billion (media estimates, unconfirmed) raises 2027 price-war concerns; alternative routes — CPO (NVIDIA Spectrum-X Photonics 200G/lane in full volume production since August 2026 as the world's first, Quantum-X already deployed with GB300; TrendForce expects ~35% penetration by 2030), LPO/LRO (transitional solutions), DAC/AEC (in-rack copper interconnect diversion) — CPO's share of high-speed ports remains small in 2026–27, a genuine threat for 2027+.
The industry is in the middle-to-late stage of an upcycle: demand explosion + supply shortage (EML gap, insufficient 1.6T capacity) support continued upside through 2026–2027; but LightCounting cautions that optical module sales often peak ahead of the cycle while cloud capex is still growing (supply-chain over-stocking), and the stock price has already run far ahead. Leading-indicator checklist: ① the five major cloud providers' quarterly capex guidance wording; ② NVIDIA's quarterly data center revenue and guidance; ③ EML lead times/LTA prices and capacity bookings; ④ module makers' prepayments and inventories (stocking intensity as a proxy for order confidence); ⑤ LightCounting's supply-demand gap readings (2x → ~30% → convergence by end-2026); ⑥ 1.6T demand/capacity ratio; ⑦ CPO switch shipment ramp slope. Regulation: ① new FCC equipment authorization rules promulgated on 2026-09-10 did not list optical module companies ("optical communication concerns may ease significantly" per Industrial Securities Communications) — but watch whether the rules allow scope expansion; ② export-control expansion path (Morgan Stanley's scenario: starting from the 3.2T generation, with a 65% US-content exemption) is the largest policy tail risk; ③ tariffs — optical modules (HTS 8517.62) were previously on the exemption list, with high policy volatility; ④ on the import side, high-end DSPs/optical chips face supply-cutoff risk under US export controls. Neutral-to-negative: deep binding of the demand side (US cloud providers) and supply side (Chinese capacity) makes hard decoupling extremely costly, and Thailand capacity provides a buffer, but policy tails suffice to cap valuation.
| Company | Revenue scale | Latest growth | Gross margin | ROE | Position/differentiation |
|---|---|---|---|---|---|
| Eoptolink | 2025: RMB 24.8 billion; 2026H1: RMB 20.9 billion | 2026H1 +100.3% | 48.46% (2026H1) | 72.6% (2025FY) | Global No. 2; highest net margin in the industry, key Amazon supplier + NVIDIA qualification |
| Innolight | 2025 optical modules USD 5.33 billion; 2026H1 RMB 41.778 billion | 2026H1 +182.5% | 48.46% (2026H1) | 39.2% (2026H1) | Global No. 1; core across the NVIDIA/Google ecosystem, 1.6T first mover (share >50%) |
| T&S Communications | 2026H1 RMB 2.828 billion | Net profit +33.9% (2026H1) | ~60% | N/A | Upstream passive components + packaging OEM "picks-and-shovels" player; lower growth but higher gross margin |
| Accelink | 2026H1 RMB 6.610 billion | +26.1% (2026H1) | Datacom segment 21.85% | N/A | Global No. 4; state-backed IDM, main battleground domestic and telecom, notably lower gross margin |
| Coherent | FY2026 USD 7.12 billion | +22.5% | 38.5% (GAAP) | 8.5% | Vertically integrated IDM; growth far below the duopoly — counter-evidence to "inflection dividends concentrated at the top" |
One of the two global AI datacom optical module duopolists, with a rising share trend (outside top 5 in 2023 → No. 3 in 2024 → No. 2 in 2025) but facing marginal pressure in 2026 (growth trailing Innolight and the industry). Moat sources: ① cost and expense efficiency (net margin of 36.0%, ~3pp ahead of Innolight); ② delivery and rapid iteration (headcount 2,000 → 8,000+ in three years, 1.6T volume delivery, 3.2T/6.4T NPO in development); ③ customer structure positioning (key 400G/800G supplier to Amazon, NVIDIA qualification in 2025, leading among emerging compute customers not directly supplied by NVIDIA). Weaknesses: R&D intensity only ~40% of Innolight's, reliance on external EML optical chips, North America policy exposure as the biggest tail. Positioning verdict: leader (No. 2), with two-way possibilities — "catching up with No. 1" or "being caught by No. 3" — and the generational transition (the 1.6T→3.2T and CPO form-factor debate) is the window for a share reshuffle.
A generic financial framework for the optical module industry easily misses three defining swing factors, presented separately:
Product generations and ASP structure: The industry transitions roughly every two years — 400G→800G→1.6T→3.2T — with unit value jumping ~2–3x per generation; the company's ASP rose from ~RMB 981/unit in 2024 to ~RMB 1,545/unit in 2025 and ~RMB 2,103/unit overseas in 2026H1 — half of the growth came from "price" (generational mix), not pure "volume". This determines two tracking points: ① revenue growth ≠ shipment growth; the race between annual price cuts (10–20% annual declines on mature products is industry practice) and generational upgrades determines the direction of gross margin — 2026H1 gross margin of 48.46% (+0.99pp YoY) shows the generational dividend temporarily outpacing price erosion; ② the volume ramp timing of 3.2T (Innolight basis: 2028–2029) determines when the "last generation" of this product cycle arrives.
Customer qualification and concentration: Cloud provider/NVIDIA qualification takes 1–2 years; qualification is both a barrier and a dependency — the company's top-five customer concentration is 72.34% (rising year by year), with the largest customer at 22.97%; Amazon is the key 400G/800G customer, and NVIDIA qualification was obtained in 2025. Rising customer concentration coexists with "declining share of the largest customer", reflecting "fewer large customers buying more; customer count growing but head-concentration deepening". The company does not disclose shipments by speed tier or revenue by geography — these two data gaps are the biggest obstacle to external share tracking, leaving only IR call language ("800G mainstay, 1.6T ramping, silicon photonics rising") cross-checked against third-party rankings.
Technology route debate (the 2027–2028 watershed): Pluggable (the company's current main business) vs CPO (NVIDIA Spectrum-X in volume production, optical engines integrated on the switch side) vs NPO/LPO (transitional form factors) vs OCS (optical circuit switching). The company's roadmap: NPO expected to commercially ramp in 2027H2 and scale in 2028, "becoming one of the main products"; OCS volume commercial demand in 2028; already participating in CPO layouts; released 6.4T NPO and 12.8T XPO solutions in March 2026. In-rack scale-up copper interconnect and inter-rack optical interconnect will coexist long term; CPO's material erosion of pluggables is limited in 2026–27 (still a small share of high-speed ports), but the architectural choice of the 3.2T generation after 2028 is the key event determining whether the company's second growth curve (NPO) materializes or share is lost — a long-term tracking variable more important than quarterly results.
Overall assessment: Cautiously bullish (confidence 0.58), 12-month horizon. The current price of RMB 306.01 is ~8% below the lower bound of the base fair-value range (RMB 330); the core logic is the expectation gap between "peak concerns being priced in" and "2027 growth delivery". The reasons for limited confidence are equally clear — a fat left tail (bear case RMB 140–185, probability ~0.22), consensus estimates still on a downward revision path, and two of the three catalysts behind the stock's decline (capex rumors, export controls) not yet officially disproven.
Strategy implications: The current price suits phased position building rather than a one-time heavy position — before the two near-term validation points (Q3 report on 10-30, export-control text) land, policy and sentiment dominate over fundamentals; if the Q3 report shows single-quarter revenue QoQ ≥+10% and net margin holding above 35%, the position can be raised to the base level; conversely, if QoQ turns negative or gross margin declines consecutively, wait for pricing of the RMB 180–260 transition zone to complete before re-evaluating.
Risk disclosure (ranked by impact): ① export-control expansion to 1.6T and below, or the Thailand plant failing to obtain an exemption — bear-case scenario, -40% or more; ② downward revision of North American cloud providers' 2027 capex guidance (≥2 companies, ≥10%) compounded by deeper annual price cuts — base case invalidated, converging toward the RMB 180–260 transition zone; ③ elasticity of additional inventory write-downs (RMB 11.655 billion in inventory) if orders reverse (provision ratio only 4.90%); ④ structural share dilution (growth consecutively trailing Innolight and the industry); ⑤ governance discount (record of penalties against the actual controller, related-party procurement) and H-share issuance dilution (mechanical impact ~7–9%); ⑥ FX (USD exposure ~RMB 11.4 billion, unhedged).
Catalyst timeline: 2026-10-30 Q3 report (single-quarter QoQ direction; whether cumulative first-three-quarters revenue reaches RMB 37 billion); 2026Q4 export-control/FCC follow-up texts and the RMB 0.8–1.0 billion buyback implementation process; 2027-01 AOI patent litigation oral arguments (involving legacy 400G/100G products, limited direct exposure, guard against expansive interpretations of any injunction); 2027-04 annual report (reconciliation against the RMB 50 billion revenue incentive target, quantified 2027 demand statements); 2027H2 NPO commercial ramp validation. Price signals: if the stock recovers to ~RMB 420 (upper bound of base fair value), evaluate taking profits; a fall below RMB 240 (a further ~27% discount to the base lower bound) would imply the market pricing a deep bear case and require re-examining the entry thesis.
Data and basis notes: Financial data come from the company's 2023–2025 annual reports and 2026 Q1 and interim reports (as of 2026-06-30); prices and valuation multiples are as of the 2026-10-15 close; industry data come from LightCounting, Goldman Sachs, TrendForce, Yole, and other institutions (some cited via media, as noted); forecasts and scenarios are this report's own estimates, including subjective probability judgments, and do not constitute investment advice.