Phase Space AI

03 Valuation

Applied Optoelectronics [AAOI]

Applied Optoelectronics [AAOI] — Valuation, Street Context & Factor Scorecard

Task 3 | investment-memo skill v1.4.0 | as of 2026-07-27 | spot $97.82

---

2026-07-29 — METHODOLOGY MIGRATION (supersedes parts of this document)

Written 2026-07-27 under the retired six-Gate / E[R]-vs-cash-hurdle framework; migrated onto the Criteria framework. The analysis below is retained as evidence; the following supersede it. The authoritative valuation output for this name is now §3A/§3B of 05_Trade_Construction.md.

Data basis — the correction that did NOT bite here

	OLD	NEW
Revenue basis in the screen / scan / rescore path	last fiscal year, FY2025, 210 days stale	TTM, 4 quarters to 2026-03-31
Revenue	$455.7m	$472.6m (+3.7%)
EV / Sales	14.54x	14.02x
Demonstrated revenue CAGR	26.9% (annual-FY)	26.0% (TTM-consistent)
Terminal EBIT margin	20.0% hardcoded	18.45% derived (semis/optical cohort median; AAOI is pre-profit)
Implied-path margin	+5.5pp	+2.6pp

The last-FY error that understated MU by 136% and SanDisk by 76% understates AAOI by only 3.7%, because AAOI's inflection is annual rather than intra-year. The correction that actually moved this name was the terminal margin, and it cost 2.9pp of margin. Corrected inputs come from reports/scan/AAOI_analysis.json and are used rather than recomputed.

What is superseded

• §6's probability-weighted value of $48.15 and E[R] for a long = −50.8% are retired, along with the 4.7% cash hurdle. The reverse-DCF implied-path test is now the ownership test: the price requires a 23.4% five-year revenue CAGR against 26.0% demonstrated — a margin of +2.6pp, a narrow PASS.
• §6's exit multiples are explicitly disallowed as five-year exit anchors. They were anchored on "AAOI's own trading history (1–2x sales pre-2024)" with a bull case at 7.0x. A 1–2x sales multiple was set when AAOI was a flat-revenue business; applying it to a 26% grower is the same defect as anchoring a 30% grower on 1–7% growers. The growth-matched anchor is 42.8x EV/EBIT (n=16, of which 5 are semis/optical — the best sector fit of the four names in this batch), with a sector-restricted cross-check at 64.1x (n=5).
• Sensitivity is now run over the EXIT MULTIPLE, never over scenario probabilities.
• A 12-month target is now required and is $124 (+41.3% vs spot $88.12), range $115–$134 on the consensus revenue range, built at AAOI's own trailing-one-year median EV/Sales of 10.32x. The anchor is declared identified but fragile — the one-year window is entirely a re-rating period.
• Price is now $88.12 (2026-07-28, scan), not the $97.82 used below.

NOT superseded — and this is the point of the document

The accounting-quality evidence stands in full and is independent of every revenue, multiple and framework question above. Digicomm International at 53.1% of revenue and 71.6% of receivables at ~264-day implied DSO, named in none of the 18 quarterly earnings releases (4Q21–1Q26) — re-verified independently for this migration, zero matches across all 18 Ex-99.1 files on disk. Company-wide DSO 106 → 180 days; AR +418% on revenue +249%; AR + inventory of $505.2m against TTM revenue of $507.0m; accruals +0.159, the first positive year in five; eight registered offerings in fifteen months and share count +193%. This is now scored under the Downside Criteria and it is the memo's real output. §3's factor scorecard is likewise retained in full.

---

1. Capital structure and starting multiples

---

2. Transcript Mention-Frequency Table (REQUIRED)

Corpus: 8-K Ex-99.1 earnings releases, 18 quarters, 4Q21–1Q26. One pinned source across the entire series, never mixed. Chosen because Alpha Vantage was exhausted and because these releases are first-party, complete, and remarkably constant in length (2,391–3,086 words), which makes normalisation stable. All figures below are per 10,000 words, never raw counts.

Corpus limitation stated up front: an 8-K Ex-99.1 is a press release, not a call transcript. It contains no Q&A, so the prepared-remarks-vs-Q&A split that the method normally reports cannot be computed here. Every mention below is therefore effectively "unprompted" — which strengthens the emerging-term signal but removes the ability to distinguish management priorities from analyst curiosity. Where a term is absent, that may reflect the genre rather than management attention; this is called out per term.

The four findings that carry weight

Finding 1 — the 800G qualification arc is a decay story, not a ramp story. Reading the actual sentences in sequence:

• 3Q23: "we shipped our first samples of 800G datacenter products to two different customers"
• 1Q25: "growing demand for our 400G and 800G products… actively working to build out our capacity"
• 2Q25: "new customer qualification efforts for 800G and 1.6Tb transceivers"
• 3Q25: "We continued to make progress on securing 800G product qualification. We are nearing what we believe are the final stages of 800G p[roduct qualification]"
• 4Q25: zero mentions of 800G
• 1Q26: "strong customer engagement around our 800G transceivers and 1.6 Tb products"

qualification peaked at 13.9/10k in 2Q25 and collapsed to 3.5 — pure boilerplate — by 4Q25. A term that decays immediately after management says it is "nearing the final stages" is a negative signal, not a positive one. Nearly three years after first samples, the language has moved backwards, from "qualification" to "engagement". No qualification win has been announced.

Finding 2 — the "AI" ramp is a boilerplate rewrite, not a business change. Every AI hit was inspected in context. In 2Q25 the standing "About Applied Optoelectronics" paragraph was rewritten from a generic description to "the building blocks for AI datacenters, CATV and broadband fiber access networks". By 1Q26 the company's self-description had become "a leading provider of advanced optical and HFC networking products that power AI". The company inserted "AI" into its own boilerplate identity. The 1Q26 spike to 15.1/10k is largely this. This is exactly the failure mode references/mention-frequency.md warns about, caught by inspecting surrounding text rather than trusting the count.

Finding 3 — the corpus structurally excludes financing language, so the caller's financing-leading-indicator test cannot be run on it. dilution, ATM, offering and warrant are zero in all 18 quarters; convertible appears only inside balance-sheet line items and non-GAAP reconciliation boilerplate. Reporting "no financing signal" would be wrong — the correct statement is that this corpus cannot carry that signal. The signal was therefore taken from an independent corpus, the EDGAR filing index, where it is emphatic: eight registered offerings in fifteen months and a +193% share count. The caller's hypothesis was right; it simply lives in a different dataset.

Finding 4 — Digicomm, at 53.1% of revenue, is never named once. Nor is DOCSIS (one mention in 18 quarters), the technology cycle that generated the majority of the growth. The company discusses CATV constantly in the abstract (17–26/10k, stable) while never identifying the counterparty or the cycle driving it. Disclosure occurs only where Regulation S-K compels it.

---

3. Factor & Anomaly Scorecard (REQUIRED)

Direction assessed for a contemplated long position.

Synthesis

Nine of ten computable factors point down; one — momentum — points emphatically up, and it is the strongest factor in the literature. This is the central tension of the name and it is exactly what the screen's composite of −2.56 was capturing.

The quality block is unambiguous and mutually reinforcing: positive accruals, +114% asset growth, F-score 4/9, and weak gross profitability are not four independent readings — they are four views of the same fact, that this company is converting equity issuance into working capital rather than converting revenue into cash.

For Quality Criteria purposes this is the empirically favourable profile for a short: high-accrual, low-F-score, high-asset-growth names are where short theses actually work. For Momentum Criteria purposes the momentum reading is a serious obstacle — though tempered by the fact that the trend has already broken (44% of the high, below the 50-day).

---

4. Consensus, Street positioning, and the required bridge

4.A What the Street thinks

Sourcing honesty: Alpha Vantage's 25/day quota was exhausted before this build, so consensus figures come from public aggregators rather than a primary feed. The FY2026 revenue consensus in particular is reported as a range because sources genuinely disagree. The 2Q26 revenue consensus of $194.1M and the FY2026 EPS of $0.91 are the two most consistently reported figures and carry the bridge below.

4.B The bridge: house vs Street, decomposed

House FY2026E revenue $742.9M vs Street $850M–$1.0B → a −13% to −26% variant.

But the sharper and more falsifiable version of the disagreement is about margin, not revenue:

Consensus FY2026 EPS of $0.91 on ~86M diluted shares implies roughly $78M of net income, which requires approximately $100M of EBIT. Against consensus revenue of ~$950M and an opex base running at ~$235M, that requires a gross margin of roughly 35%.

The company has guided 2Q26 gross margin to 29–30%. It has never printed above 30.0% on a full-year GAAP basis in the last nine years, and its all-time annual high is 43.5% back in 2017 on a completely different mix.

And the higher-margin tranche is the one rolling off. Gross margin rose from 24.8% to 30.0% precisely as CATV mix rose from 35.2% to 53.8%. As CATV normalises, mix works against margin.

Is the gap about numbers or about the multiple? Decomposed properly: it is about both, and the numbers half is the load-bearing half. Consensus needs ~35% gross margin and ~$950M of revenue; the house model has 29.5% and $742.9M. Even if one grants the Street's revenue and applies the company's own guided margin, EPS lands far below $0.91. That is a numbers disagreement, not a multiple disagreement — which is what makes it a legitimate Consensus Criteria variant rather than a valuation complaint.

4.C Why might the Street be right, and what is the risk to the house view

1. The 800G qualification could land. If it does, datacenter revenue re-rates and the mix/margin argument inverts — 800G/1.6T carry higher ASPs than legacy datacenter product.
2. Nine straight quarters inside guidance. The Street is extrapolating a real and improving execution record. Betting against it has lost money for two years.
3. The sector regime. LITE trades at 33.7x sales. If the market is right that optical is structurally re-rated, AAOI at 15.5x is cheap within its group, not expensive.
4. Six analysts is a thin consensus on a $7.8B company — the target dispersion ($0.44 to $1.33 EPS) is enormous, which means "consensus" here is a weak anchor in both directions.

Named risk to the house view: the single event that breaks this memo is a named hyperscaler 800G/1.6T qualification win with disclosed volume. It is checkable at every earnings date and is written into the invalidation triggers.

4.D Options-market read

• Realised volatility: 135.9% (1-year), 141.0% (3-month). Implied 127–148%.
• IV ≈ realised. There is no volatility edge in either direction — options are fairly priced and simply enormously expensive in absolute terms.
• Implied move to 21 Aug (covers the 6 Aug print): ATM straddle $29.01 on $97.82 = ±29.7%.
• The 18-Sep $60 put returned a quote with no IV and no Greeks — no valid two-sided market. Deep-OTM liquidity is unreliable.

---

5. DCF

Full build in AAOI_Model.xlsx, tab DCF (live formulas, verified by reading back computed cells in Excel).

WACC 13.5%, terminal growth 2.5%, tax 21%.

Result: enterprise value −$463M, equity value per share −$1.66.

This DCF is reported, but it is explicitly not the valuation anchor, and saying otherwise would be dishonest. A discounted cash flow model is not meaningful for a business with negative unlevered free cash flow in all five forecast years. What the DCF legitimately establishes is narrower and still useful:

No cell in the entire WACC × terminal-growth sensitivity grid (11.5–15.5% × 1.5–3.5%) produces a value anywhere near $97.82. The highest cell in the grid is $1.75. For a DCF to justify today's price, the operating assumptions — not the discount rate — would have to change fundamentally.

The base case is not a strawman: it has revenue growing +63% in FY2026 and reaching $974M by FY2030, gross margin holding at 28%, and EBIT turning positive in FY2029. It is a reasonable case, and it still produces no value, because the company's opex base of ~$235M against a ~28% gross margin requires roughly $840M of revenue simply to break even at the EBIT line.

---

6. Scenario valuation — the primary anchor

Exit multiples are anchored to AAOI's own trading history (1–2x sales pre-2024), not to the peer median of 10.2x, because every peer generates positive operating cash flow and AAOI does not. The bull case at 7.0x is still below the peer median — the bull case is deliberately not strawmanned.

• Probability-weighted value per share: $48.15
• Scenario-weighted E[R] for a long: −50.8%
• Cash hurdle (portfolio_book.json): 4.7% → E[R] less hurdle: −55.5%

The most important line in this table is the bull case. A genuine hyperscaler qualification win, revenue nearly tripling to $1.35B by FY2028, a 9% EBIT margin, and a 7.0x sales multiple — everything the bulls are underwriting, granted in full — returns +8.5% from today's price. The stock already prices in the bull case.

Base-rate check (required)

• Base FY2026E growth of +63% off $455.7M: Chan/Karceski/Lakonishok (2003) find firms growing >60% at ~$0.5B scale sustain it for a median of one to two years, with five-year persistence statistically indistinguishable from zero. The model decays growth to +5% by FY2030E — that is the base rate, not pessimism.
• Bull FY2028E of $1.35B implies +43%/yr for three consecutive years. That is a top-decile outcome for hardware at this scale. It is named as such and weighted 25%, not 50%.
• Gross margin is modelled down (30.0% → 28.0%) on the CATV-mix argument, against a Street that needs it up to ~35%.

---

7. Comparable companies

Two honest observations that cut in opposite directions:

1. AAOI is the only name in the set with negative operating cash flow, and it is the smallest, most concentrated, and most issuance-dependent. That justifies a discount.
2. AAOI is not the most expensive name in the group. LITE trades at 33.7x. Per references/trade-construction.md: "If the whole group is expensive together, that is a sector-regime question, not a stock-specific short thesis." The valuation, by itself, is therefore explicitly NOT the thesis here — the thesis is the cash-conversion and cycle-composition mechanism in Sections 3 and 12 of the research document. Valuation only translates that mechanism into expected return.

---

Model: AAOI_Model.xlsx (6 tabs, live formulas, verified via AppleScript read-back with all workbooks closed first). Sources: SEC EDGAR XBRL + R-files (CIK 1158114); Alpaca Markets equities and options; public aggregators for consensus (Alpha Vantage quota exhausted — flagged, not estimated).