Applied Optoelectronics, Inc. (Nasdaq: AAOI) company deep dive, first edition. Two PDFs, Chinese (230 pp) and English (296 pp), each with 16 chapters plus 4 appendices and 40 figures and 151 data tables, table-for-table consistent across the two versions; every computed number is derived in Appendix D. Data as of the close on 2026-09-22; financials to 2026-06-30 (10-Q filed 2026-08-06); filing cutoff 2026-09-22, later filings excluded.
AAOI sells two kinds of hardware by the unit: 100G to 1.6T optical modules for hyperscale data centers, and CATV amplifiers sold through a distributor to cable operators; in Q2 2026 they were 56.1% and 42.0% of revenue. In two years the company doubled its share count and got capacity in Texas, Taiwan and Ningbo in return: from 2025 through H1 2026 it raised $1.547bn net through ATMs, 97.3% of its external funding. At 30 June 2026 the cost side already carried $480.2m of construction in progress and prepayments on the balance sheet, not yet depreciating, plus leases signed but not yet commenced with $173.3m of ten-year base rent. The revenue side had not one contract that locks in future revenue: contract liabilities were zero, RPO is not disclosed, and of Customer C's announced $324m of orders at most $12.8m had been delivered by 30 June. The CATV half is sold in through a single distributor, Digicomm, recognized on shipment and collected in about 9 months, with $211.0m receivable at 30 June; the end customer, Charter, states that its network evolution completes in 2027. The report answers three questions: how fast, and in what proportion, capacity bought with equity turns into revenue and cash; how far management's figures should be discounted after five product-level forward calls, none delivered in full; and what delivery ratio the share price already assumes.
What the report delivers
| Chapter | Specific result |
|---|
| 2, 3 | Who buys each leg, from which budget, and how fast they pay (data center 57–88 days, CATV about 9 months); six quarters of net ATM proceeds matched quarter by quarter to capital spending; the seven lags from equity to collected cash, calibrated on cases that have already happened; nine years of ≥10% customers and a quarterly reconstruction; the Amazon warrant's accounting count reconciled separately from its legal vesting; narrowing down Customer C; the press-release order ledger and six substitutes for RPO; a breakdown of Q3 guidance and a table of how management's figures drifted |
| 4, 5 | Digicomm at 97–100% of CATV revenue, with receivable days of 256–329 on a FIFO basis; how lender BOKF carves the Digicomm receivable out of its borrowing base; Charter's upgrade spending and the precedent of CATV revenue falling 79% in five quarters from Q1 2023; upstream suppliers and financiers; the capacity ledger item by item ($480.2m on the balance sheet awaiting service, $173.3m of leases not yet commenced, $248.8m of purchase options on two phases and the company's exit route); the three-location capacity map and the delivery clock; about $67.4m a year of new fixed cost once assets enter service and leases commence |
| 6 | A slice of the chain: of four key inputs only the laser chip is made in-house; six independent sources on the InP laser shortage; the speed mix (800G at 11.9% of data center revenue) and management's 800G and 1.6T figures against actuals; the timing of LPO and CPO; the Chinese cost curve; tariff and geopolitical exposure; a falsifiable table of five moat claims |
| 7, 8 | Segments and margins quarter by quarter; all-in capex after the XBRL tag change; the tax line between GAAP and non-GAAP (an add-back in 20 of 22 quarters); the financials split into cause, process and result; nine earnings-quality checks; seven XBRL tagging errors; seven equity distribution agreements round by round (the same $100m of net proceeds took 5.76m shares in 2025 and only 550k in Q2 2026); interest-bearing debt and the redemption-notice window on the 2030 convertible; four share-count bases and EV; a monthly funding ledger, the financing toolkit and the range of further dilution |
| 9 | The charter and its 66⅔% threshold; holder structure; the control provisions inside the Amazon warrant; the board's ties to the ATM agent; incentives tied only to the share price; the timing of two severance and indemnification enhancements; twelve months of insider trades one by one ($124.5m sold, zero bought, 82.6% of the amount outside 10b5-1 plans); the change of auditor; shareholder votes |
| 10, 11 | Five peers on one basis over eight quarters, with industry-weighted growth; a substitute mirror for CATV; capex intensity, SBC and returns; share-count change against the sources of expansion funding; the valuation table and AAOI's own 5-year percentile; a reconstruction of sell-side consensus; five kinds of evidence: price reverse-engineering, event study, peer dislocation, options-implied distribution and holder structure |
| 12, 13 | The rating framework (one DCF yardstick for the basket and the stock); valuation inputs; the defining variables and annual results of three scenarios; target-date share count and net cash on one ledger; probability bases by bins of the Q2 2027 realization rate, with ±10pp tests; 22 single-variable ±30% tests; discount-rate, β and terminal-value variants; a line-by-line reconciliation with the market-implied view; certainty layers, bull and bear side by side, a scenario-to-action map, conditions for the view and for invalidation, and a review calendar |
| 14–16 | 18 risks ranked by observability, risks from a common source and the funding amplifier; five figures that are easy to misread; falsification conditions for all seven judgments, a catalyst path with eight dated nodes in 12 months, tracking indicators and proxy readings for the linchpin variable, and the settlement method |
| Appendices | A terms and definitions; B sources and 37 known limitations; C disclaimer, conflict disclosure and a revision log of 92 items; D derivations and calculations |
Review and revisions
The valuation model was first reviewed from four angles (rating framework / input evidence / mechanics and one ledger / bull-case rebuttal); the finished text then went through two rounds of full-text review (round one on four tracks: financials and valuation / contracts, customers, counterparties and governance / tradability / presentation and wording; round two: a skeptic's rebuttal with a tradability re-check / English-edition consistency). Together they raised 186 challenges, and the 92 substantive revisions are published in Appendix C, Table 150. One of them moved the target price. The first review round found two problems: the convertible's redemption had been judged on the 2027-01-15 path price, whereas the indenture runs the price test over the 30 trading days before the redemption notice, and notice can be given from 2026-10-19 at the earliest; and probabilities were assigned to bins of the Q2 2027 realization rate while scenario prices were taken at the bin edges. The model now re-derives conversion timing and make-whole shares in all three scenarios from the indenture and prices each scenario at the conditional mean of its bin; together with two further model changes in the same round, the target moved from $77.09 to $93.02. The direction of the rating did not change, but the number of single-variable ±30% tests that flip it rose from two to seven.
Sources and verifiability
Primary sources: SEC EDGAR (CIK 0001158114) 10-Ks for FY2019–FY2025, the 10-Qs and earnings releases; 8-Ks with exhibits (Microsoft's two SOWs, the Amazon transaction agreement and warrant, the 2030 convertible indenture, the BOKF loan agreement, the Hightower and Ningbo leases, two cleanroom general contracts); seven equity distribution agreements and eight 424B5s; DEF 14A; 133 raw Form 3 / 4 XML filings and 60 Form 144s; 22 13G / 13G/A filings; XBRL companyfacts. Also counterparty and peer filings from Charter, Comcast, Harmonic, Lumentum, Coherent, Fabrinet and others; the interim reports of Zhongji Innolight and Eoptolink; FINRA short-interest data, the Nasdaq options chain and 13F aggregates, and the US Treasury yield curve; earnings calls from public transcripts. Every claim carries an [A]/[B]/[C]/[D] evidence grade; 37 known limitations and data gaps are listed item by item in Appendix B, Table 149.
How this report was produced, and disclaimer
Compiled independently by the TopX research team from public information, then reviewed and revised before release by professional investment managers and fund managers with asset-management experience. Reviewers take part in a personal capacity, do not represent their respective employers, and provide no investment advisory services through TopXEA; TopXEA holds no investment-adviser licence in any jurisdiction. Primary filings (SEC, exchanges, company disclosures) are the first-priority source, key figures require corroboration from at least two independent sources, claims carry [A]/[B]/[C]/[D] evidence grades, and known limitations are disclosed item by item inside the report. This is research content: it does not constitute investment advice, an offer or a solicitation and is not tailored to any particular investor. Ratings and target prices are analytical judgments, not instructions to buy or sell; past data and forward projections do not indicate future results.
Conflicts of interest: (1) positions held by the publisher or its affiliates in the subject security: none; (2) business relationship with the subject company or its affiliates: none; (3) the publisher also operates the sale of Expert Advisors (automated trading software, including the EA VIP membership), parameter presets and strategy source code, the TopxAI AI-model API relay service (ai.topxea.com), and TopxAI API credit cards (prepaid credit codes), which have no relationship to the subject; (4) this report is paid for by readers, not by the subject company.
Please note before purchase: digital goods are non-refundable once delivered.
