Fair-Value Report — Zhongji Innolight Co., Ltd. - H Shares (3308.HK)
Insider Signal
Buyer: The Capital Group Companies, Inc. — one of the world's largest active fund managers (manager of the American Funds complex), not a custodian, prime broker, index manager or nominee. This is a genuine, economically-motivated institutional investor.
Filing: HKEX Disclosure of Interests, reason 1101/1103 (on-/off-exchange purchase). Disclosed ~Oct 2, 2026 for a trade on Sept 30, 2026: 369,100 H-shares bought at an average price of HK$1,069.1312, value ≈ HK$394.6 million.
Stake change: ≈5.68% → 6.27% of the class (+0.59pp). This is an incremental top-up by an already-substantial shareholder, not a first-crossing of the 5% disclosure threshold.
Why this passes the gate: the IPO priced H-shares at HK$980 on July 30, 2026. Capital Group's Sept 30 purchase was executed ~9% above the IPO price, two months after listing, at a genuine secondary-market print — ruling out a placing/cornerstone-lockup artifact. No pledge, control-default, intra-group reshuffle, or employee-scheme characteristics apply. Verdict: genuine open-market accumulation by a real, conviction-driven buyer.
Verified Snapshot
FX: CNY/HKD ≈1.1703, HKD/CNY ≈0.8544 (Oct 5, 2026). Sources cross-checked in Chinese primary filings (CNINFO-sourced releases via cnr.cn, stockstar.com, eastmoney.com) per the primary-language rule for PRC-incorporated H-share issuers; English aggregators (stockanalysis.com) used only for H-share-specific price/mechanics.
Market-cap caveat: stockanalysis.com's HK-listing page (HK$1.11T) and SZ-listing page (CNY 951bn) each apply one listing's price to what appears to be an inconsistent share-count basis. The figures above use the actual H/A share split applied to each listing's own price, then converted — treat as the more reliable figure, but reconcile again once SZSE reopens Oct 8, 2026 and a contemporaneous A-share print exists.
Business
Zhongji Innolight is the world's largest maker of datacom optical transceivers (21–23% global market share, 2025), supplying high-speed optical interconnects for AI/cloud datacenters. It is reported to supply approximately 80% of NVIDIA's GB200-platform 1.6T optical modules, making it NVIDIA's largest single optical vendor among its three primary suppliers (Innolight, Coherent, Fabrinet) — a dominant, structurally advantaged position in the current AI-infrastructure buildout. H1 2026 optical-transceiver segment gross margin was 46.59% (+6.6pp YoY) on 1.6T/800G mix shift. R&D intensity is modest (2.75% of revenue), consistent with a scale-manufacturing rather than R&D-heavy model.
Concentration risk is material and rising: top-5 customers were 75.98% of FY2025 revenue, rising to 81.9% by Q1 2026 (largest single customer ≈24–25%); one third-party estimate (not a company disclosure) puts NVIDIA alone near 35% of revenue. Overseas revenue is 94.8% of the total (up from 86.8% in FY2024) — the business is almost entirely a bet on continued US/global hyperscaler AI capex flowing through a small number of counterparties.
Valuation
Method: profitable, fast-growing operating company → three-scenario probability-weighted DCF (owner-FCF basis) is the primary lens, cross-checked against global optical-component peers and a reverse-DCF read of what the current price implies. Modeled in RMB (the operating currency) on the combined A+H share base, then converted to HKD/CNY per-share for comparison to each listing.
WACC (CAPM, scripts/wacc.py)
| Input | Value |
|---|---|
| Risk-free (CNY 10Y) | 1.70% |
| Beta (relevered, hardware/comms-equipment, cyclical+concentration-adjusted) | 1.30 |
| Mature-market ERP | 4.20% |
| Country/political risk premium | 2.50% |
| ↳ elevated above the generic China CRP (≈0.9–1.0%) to reflect name-specific risk: active US DoD Section 1260H "Chinese Military Company" listing (June 2026) and a Sept 28, 2026 Senate bill naming Innolight by name for a federal-procurement bar — not generic market risk | |
| Cost of equity | 9.66% |
| Cost of debt (post-tax) | 2.98% |
| WACC | 9.53% |
Three-scenario DCF (scripts/dcf.py) — fair value per share
Common inputs: revenue base RMB65.23bn (trailing-twelve-months through H1 2026, since FY2025's RMB38.24bn is already stale versus the in-year ramp); operating margin paths anchored to the actual H1 2026 operating margin of ≈38–39% (net margin 32.7%, grossed up at an assumed 15% effective tax rate consistent with HNTE preferential treatment common for PRC tech-hardware exporters — not independently confirmed from a disclosed tax-expense line, flagged as an assumption); tax 15%; SBC ≈0.5% of revenue (immaterial for a PRC manufacturer without a heavy equity-comp culture); net cash RMB5.51bn; 1,178.275m diluted shares; weights 25% bull / 50% base / 25% bear (standard default — elevated geopolitical risk is already captured via the WACC political-risk premium above, so scenario weights were not additionally skewed to avoid double-counting).
| Scenario | Yr1 growth | Yr1 op. margin | Terminal g | FV/share (CNY) | FV/share (HKD-equiv.) | TV % of EV |
|---|---|---|---|---|---|---|
| Bull (25%) | +55% | 39%→42% | 3.0% | 686.47 | 803.9 | 75.4% |
| Base (50%) | +35% | 37%→27% | 2.5% | 269.71 | 315.6 | 69.2% |
| Bear (25%) | +5% | 25%→18% | 2.0% | 100.82 | 118.0 | 49.6% |
| Probability-weighted | 331.68 | 388.2 |
Current H-share price HK$1,111.00 is −65.1% above the weighted fair value (HK$388.2); current A-share price CNY808.44 is −58.9% above the weighted fair value (CNY331.68). Even the explicit Bull case (HK$803.9 / CNY686.47) sits below today's price — every scenario this model can construct, including one with 1.6T-driven margin expansion to 42% and 55% Year-1 growth, still falls short of the market price. Terminal value is 50–75% of enterprise value across scenarios, so the result is inherently sensitive to the long-run growth/margin assumptions — appropriate caution given this is a component manufacturer, not a software compounder.
Reverse DCF (scripts/reverse_dcf.py) — what the price implies
At HK$1,111.00 (CNY 815.86 blended), holding base-case margins, WACC and terminal growth fixed, the market is pricing in a flat 38.0% revenue CAGR sustained for 10 straight years (reinvestment held at 15.2% of revenue). That compounds the RMB65.2bn revenue base roughly 20-fold, to over RMB1.2 trillion, within a decade — a scale and duration of sustained hardware-component growth with few, if any, precedents (semiconductor capital-equipment and fiber-optic-equipment cycles historically decelerated sharply well before a decade of compounding at this rate). The margin-solve direction has no finite solution within the script's search ceiling — i.e., there is no flat operating margin, however high, that reconciles the price with the base-case growth path alone; the price requires growth duration, not just margin, to be this generous.
Peer comparison (median-anchored)
| Name | Price | As-of | Mkt Cap | Trailing P/E | Fwd P/E | Rev. growth (YoY) | Gross margin | Fwd P/E ÷ growth |
|---|---|---|---|---|---|---|---|---|
| 3308.HK (target, H) | HK$1,111.00 | Oct 5 | ≈HK$1.13T | 47.1 | 16.1 | +134% | 46.6% | 0.12 |
| 300308.SZ (target, A) | CNY808.44 | Sep 30 | ≈CNY961bn (combined) | 44.1 | — | +182% | 42.0% | — |
| Eoptolink 300502.SZ | CNY389.00 | Sep 30 | CNY542bn | 41.4 | 18.7 | +100% | 48.5% | 0.19 |
| Coherent COHR | $337.04 | Oct 2 | $66.0B | 81.8 | 35.8 | +22.5% | 38.5%/40.2% adj. | 1.59 |
| Lumentum LITE | $1,085.42 | Oct 2 | $97.4B | n/m | 49.9 | +83.2% | 47.4%/50.4% adj. | 0.60 |
| Fabrinet FN | $463.69 | Oct 2 | $16.6B | 35.5 | 25.5 | +35.7% | 12.2% adj. (contract mfr.) | 0.71 |
| Median (ex-target) | 41.4 | 25.5 | +59% | 45.0% | 0.65 |
This is the key methodological conflict in this report, and it should not be papered over. On growth-adjusted forward P/E, 3308.HK's H-shares (16.1x for 134% growth) look cheap relative to every global peer — cheaper than Eoptolink, far cheaper than Lumentum and Coherent. A pure peer-relative read would call 3308.HK undervalued. The DCF above says the opposite: fair value is roughly a third of the current price. Per the valuation framework's own discipline, when every comp in a hot sector looks "cheap" relative to an absolute DCF, the right read is usually sector-wide dislocation, not a DCF error — the entire AI-optics complex (Lumentum, Coherent, Eoptolink, Innolight) may be pricing a shared, multi-year hypergrowth narrative that the group could de-rate together if AI capex growth decelerates, exactly as happened to fiber-optic equipment names in 2000–2001 and semiconductor-capital-equipment names in prior cycles. The cheap-looking multiple is not, on its own, a reason to buy.
Conclusion
Rating: Speculative Sell. The insider signal is real — Capital Group is a genuine, conviction-driven global active manager, buying meaningfully above the IPO price in the open market, topping up an already-substantial stake. The underlying business is also genuinely excellent: dominant NVIDIA 1.6T optical-module share, H1 2026 revenue +182% and net profit +242% YoY, net-cash balance sheet, and an order book that reportedly extends into 2027. None of that is in dispute.
What is in dispute is the price. A disciplined three-scenario DCF — including a Bull case with 55% Year-1 growth and margin expansion to 42% — cannot reach today's price; the reverse DCF shows the market needs a flat 38% revenue CAGR for a full decade to justify HK$1,111, a bar with few precedents for a hardware component maker at this scale. That gap, combined with acute and already-realized binary policy risk (DoD 1260H listing, a Senate bill naming the company directly for a federal-procurement bar, two separate double-digit single-day crashes in 2026 alone), argues against initiating or adding exposure at current levels — even though the signal itself, and the business behind it, are both real. This is a "signal real, price wrong" case.
Key Risks
- Policy/export-control (primary, two-sided): further escalation (an actual FCC import ban, expanded procurement restrictions, or entity-list action against customers dealing with Innolight) could impair revenue quickly; conversely, resolution/de-escalation could remove the discount embedded in this report's elevated WACC and close some of the DCF gap.
- Customer concentration: top-5 = 81.9% of Q1 2026 revenue; loss or diversification-away by one or two hyperscalers would be disproportionately damaging.
- Cyclicality / commoditization: 1.6T margins are currently exceptional (46.6% gross); AI-capex cycles and optical-component history both suggest margin compression as capacity is added and competitors (Eoptolink, Coherent, Fabrinet) scale up.
- Terminal-value sensitivity: 50–75% of DCF enterprise value sits in the terminal year across scenarios — small changes to long-run growth/margin assumptions move fair value substantially.
- Dual-listing data quality: combined market cap and A-H premium/discount could not be cleanly reconciled while SZSE was closed for National Day (reopens Oct 8, 2026); re-check before acting on precise valuation levels.
- Tax-rate assumption: the 15% effective rate (HNTE preferential) used to back into the current operating margin is an assumption, not a disclosed line item — if the true rate is higher, current margin (and thus every scenario) is modestly lower than modeled.
Sources
- HKEX Disclosure of Interests filing, The Capital Group Companies Inc., disclosed ~Oct 2 2026 for trade dated Sept 30 2026 (via futunn.com news aggregation of the filing)
- stockanalysis.com/quote/hkg/3308 and /quote/she/300308 (price, market cap, P/E) — Oct 5 2026 / Sept 30 2026
- Zhongji Innolight H1 2026 interim results — cnr.cn, stockstar.com, huxiu.com, ithome.com (Aug 21–22 2026)
- Zhongji Innolight FY2025 annual results — cls.cn, Sina Finance, Eastmoney (Mar 30–Apr 1 2026)
- H1 2026 balance sheet detail — Sohu filing summary (Aug 2026)
- Customer concentration — Eastmoney caifuhao series (Feb 2026)
- NVIDIA supply-chain share — SemiAnalysis newsletter; hellochinatech.com
- IPO terms (HK$980, July 30 2026 listing; over-allotment Aug 26–27 2026) — EBC, BigGo Finance, Tencent/QQ, RTHK
- Policy risk timeline — 10jqka.com.cn, esmchina.com (DoD 1260H, June 2026); cls.cn, ithome.com (FCC rumor, Aug 2026); 10jqka.com.cn, 21jingji.com (Senate bill crash, Sept 28 2026)
- Peer data — stockanalysis.com (COHR, LITE, FN, 300502.SZ); SEC 8-K filings for COHR/LITE Q4 FY26 results
- Analyst targets — marketscreener.com (A-share consensus); various broker-note aggregation (Goldman, Citi, Morgan Stanley, Bocom, Nomura)
- China 10Y yield — tradingeconomics.com (Oct 2 2026); Damodaran mature-market ERP/China CRP — aswathdamodaran.substack.com Data Update 4, 2026
No redactions applied.
Published via wiis. Originals are encrypted at rest; this page reflects the redacted version. report abuse.
Discussion
No comments yet.