Five record quarters, a factory booked into mid-2027 and America’s only scaled domestic AI-transceiver plant have taken this stock from $18 to $233 and back to $103. The product the entire 2027 story rests on — 800G transceivers — was 7% of last quarter’s revenue, and three customers were 92%. The price now underwrites roughly three-quarters of a five-fold capacity ramp before it has shipped. We underwrite a little more than half of it, and arrive close to, but below, the market. Our band is $60–125 against $103.39.
The growth is real, but the stock is priced off products that barely exist yet. Q2 2026 revenue of $191.9m grew 86% year on year — the fifth consecutive record — yet 800G transceivers, the product the 2027 story rests on, contributed $12.8m, or 7% of revenue. Guidance requires 800G to grow roughly five-fold sequentially in Q3 and the company to ship more than $500m in Q4 alone. A reverse DCF shows the current price already assumes roughly $4.1–4.9bn of durable annual revenue at a mid-30s gross margin — about 70–80% of management’s mid-2027 exit target, sustained indefinitely.
Three customers are 92% of revenue, and the equity is the funding source. One cable-TV customer was 42% of Q2 revenue and two hyperscalers another 50%. The ramp is financed by serial share issuance: roughly $1.03bn of at-the-market stock sold in the first half, a further $600m programme filed 21 August — the proximate cause of the latest 30% drawdown — and a share count up about 49% year on year. Operating cash flow was negative $73.8m in the first half despite the headline “return to non-GAAP profitability”, which itself leaned on a $14.3m non-GAAP tax benefit. Adjusted EBITDA was roughly zero.
The strategic position is genuinely scarce, which is why the bull case deserves respect. AAOI is the only US-based maker of AI optical transceivers at scale, makes its own lasers — the industry bottleneck — has capacity booked through Q2 2027, holds a single $200m-plus order for 1.6-terabit modules, and owns a credible co-packaged-optics laser franchise carrying 55–65% gross margin on management’s numbers. Potential US restrictions on Chinese transceivers would hand it share. Our bull and superbull cases — $148 and $267 — are not fanciful and carry a combined 27.5% weight. The modal path lands at $78 and the weighted value at $86.
Start with the disclosure this company makes that almost no company makes: per-product revenue, per month if you ask, and customer percentages to the point. In the second quarter of 2026, on $191.9m of record revenue, three customers were 92% — one cable-TV operator at 42%, two hyperscalers at 26% and 24% — and the top ten were 99%.
Now the product mix, which is the central fact of this report. Of the $191.9m, CATV equipment was $80.6m, 400G transceivers $48.4m, legacy 100G $41.2m — and 800G, the product on which the entire 2027 investment case rests, was $12.8m. Up ten-fold year on year and doubling sequentially, and still 7% of the company. Management’s mid-2027 exit target requires that gold sliver to become roughly $217m per month.
Neither concentration figure is hidden — both were volunteered on the call — and the company has lived the consequence before. When its then-largest hyperscaler cut orders in 2018–19, revenue and the stock collapsed and took six years to recover. What is different this time is that the book is contracted: capacity is fully booked through Q2 2027, more than $324m of 800G and 1.6T orders are disclosed including a single $200m-plus 1.6T award, and customer demand was described as running 20–40% above what AAOI can supply. The question this report prices is not whether the demand exists. It is how much of a booked but unshipped five-fold ramp to underwrite before it ships.
The financing is the business model. The ramp’s capital — $565.5m of capex in the second quarter alone, including $280m of equipment prepayments — is funded by selling stock: roughly $1.03bn of at-the-market sales in the first half, and a third $600m programme filed on Friday 21 August, after which the shares fell about 30% in days. Selling equity at roughly 8x sales to build capacity customers have pre-booked is value-accretive arithmetic if the orders hold — and it also guarantees a supply of shares into every rally, which is why the price behaves the way it does.
| Measure | Value | Note |
|---|---|---|
| Reference price, 1 Sep 2026 close | $103.39 | The last completed session. Every multiple on this page scales with it |
| Market capitalisation | ~$9.6bn | On ~92.8m diluted shares, convertible notes treated as equity per the company’s diluted convention |
| Enterprise value | ~$9.2bn | Less $416m of net cash — $508.8m of cash and restricted cash against $92.8m of debt excluding the converts |
| EV / FY2026E revenue | 8.4x | On management’s ~$1.09bn. 20.1x on trailing revenue |
| EV / FY2027E revenue | 3.5x | On our $2.6bn base case — if it ships |
| Non-GAAP gross margin | 29.8% | Q2 2026. Guided to exit 2026 at 32–33%; the 40% aspiration is a decade old |
| H1 2026 operating cash flow | −$73.8m | Investing −$633.7m, financing +$980.3m, of which $1.03bn was ATM stock. The cash flows tell the real story |
| Receivables / inventories | $314.0m / $278.8m | A very long ~147 days of sales outstanding; inventories +35% sequentially, mostly ramp raw material |
| 52-week range | $18.50–$233.67 | 56% below the May high |
| Realised volatility, 1-year | ~140% | Against roughly 84% implied. Position size matters more than entry price |
Derived inputs. Enterprise value, net cash and every multiple in this table are ours, computed from the Form 10-Q figures for the quarter ended 30 June 2026 as reported and the 1 September closing price. The diluted share count of ~92.8m follows the company’s own guidance convention and treats the $129.1m of convertible notes as equity; the conversion terms were not re-verified for this report. PP&E of $697.1m is up 85% in six months, with a further ~$330m of other assets dominated by equipment prepayments — the bet is being placed now.
Implied terminal EBIT margin 25.8% · implied terminal EV/EBITDA 8.0x · FY2026–32 revenue CAGR 30.0% · terminal value is 89% of enterprise value.
The number above is one path, priced honestly — not the published conclusion. This panel runs a full discounted cash flow through whichever scenario card in Section 03 was last loaded, the base case on arrival at $78. The published conclusion blends the four scenario values — $15, $78, $148 and $267 — at fixed judgment weights of 27.5 / 45 / 22.5 / 5, giving the $86 weighted value quoted throughout this report. We state that blend in Section 03 rather than compute it on this panel, deliberately: on this name the distribution is the message, not the point estimate, and collapsing four paths into one number inside the model would hide exactly the thing the model exists to show. Outcomes from $15 to $267 with meaningful weight on both tails is the honest shape of a company scaling five-fold into contracted demand on two customers’ order books.
The engine is one set of arithmetic applied four ways. Revenue runs between the FY2027 and FY2032 endpoints on the base case’s interior shape; gross margin fades onto the terminal setting with the base case’s offsets — three points below terminal in 2027, converged by 2031; operating cost fades from 13.3% to 10.2% of revenue, consistent with management’s $70–80m-per-quarter guide growing with scale; capex follows the published build ratios; working capital absorbs 28% of every incremental revenue dollar, which is what 147-day receivables actually cost; cash tax is 18%. The −$764m of second-half 2026 free cash outflow is committed spend and is charged in every scenario, including the bear — that money is being spent now, whichever future arrives.
The terminal discount rate steps down once the build-out is over, and that choice is load bearing. We discount the explicit years at 13.5% — a rate that prices 140% realised volatility and a financing model that sells equity continuously — but the perpetuity at 11%, because holding the build-out rate forever forces a roughly 6x terminal EV/EBITDA on a grown, mid-30s-margin business, which is a bear case dressed up as prudence. On base settings the implied terminal EV/EBITDA is 8.0x, printed in the readout so you can judge it directly. Move the two rates independently and watch which one the answer actually cares about.
Reading the inputs. Terminal EBIT margin is an output, not an input: it is whatever the terminal gross margin leaves after the 10.2% operating-cost endpoint, and it prints in the readout. The interior revenue shape and the cost, depreciation and capex ratio paths are fixed, calibrated to the base-case table in Section 03 — the sliders move the endpoints, the rates and the dilution, which is where the argument actually is. Shares are fixed at 92.8m diluted on arrival with all future burn charged to the cash flows, treating the at-the-market programmes as NPV-neutral financing; if you think stock gets sold below fair value, the share slider is the instrument for that view. The tag beside the result reads against the published $60–125 band: any value inside it is consistent with the published conclusion and prints fairly valued; outside it, the gap to the $103.39 last price.
The number above is not a fact about Applied Optoelectronics. It is what our assumptions produce, and on this name the most consequential assumption is the FY2027 revenue endpoint, because it stands in for how much of a booked, five-fold, two-customer ramp you are willing to underwrite before it ships. We underwrite $2.6bn — roughly 55–60% of management’s trajectory. The market, on the reverse test in Section 03, is underwriting 70–80%. That gap is the entire disagreement, it is one slider, and the 5 November print moves it far more than anything we publish.
Same model, different inputs. Click any card to load it. The spread from bear to superbull is eighteen times, and we publish it at full width rather than trimming the tails, because on a company spending the money before the revenue arrives, both tails are real: the value destruction in the bear is savage precisely because roughly $1.3bn will have been spent into softness, and the superbull is management’s own numbers, which they have hit five quarters running.
Each card loads into the model above; the card values print from the same engine: $15, $78, $148 and $267. Blended at 27.5 / 45 / 22.5 / 5 — weights we fix in this report rather than expose as controls — they give the $86 weighted value. The base case is modal, not conservative — it is the single most likely path on our reading, and 27.5% of the distribution sits below it.
| Assumption | Bear | Base | Bull | Superbull |
|---|---|---|---|---|
| FY2027 revenue | $1.9bn | $2.6bn | $3.4bn | $4.3bn |
| FY2028 revenue | $2.0bn | $3.6bn | $5.2bn | $7.2bn |
| FY2032 revenue | $2.05bn | $5.25bn | $8.4bn | $12.9bn |
| Share of management’s mid-2027 exit run-rate | ~33% | ~55–60% | ~75–80% | ~100% |
| Terminal gross margin | 30% | 36% | 38% | 40% |
| Implied terminal EBIT margin | 19.8% | 25.8% | 27.8% | 29.8% |
| Build-out WACC | 13.5% | 13.5% | 12.5% | 12.0% |
| Terminal WACC / growth | 12.5% / 2.0% | 11.0% / 3.5% | 10.75% / 3.5% | 10.25% / 3.5% |
| Implied terminal EV/EBITDA | 5.5x | 8.0x | 8.3x | 9.0x |
| Diluted shares | 92.8m in every scenario; all future burn is charged to the cash flows instead. See the risk table below | |||
| Value per share | $15 | $78 | $148 | $267 |
| H2’26 | 2027 | 2028 | 2029 | 2030 | 2031 | 2032 | |
|---|---|---|---|---|---|---|---|
| Revenue ($m) | 745 | 2,600 | 3,600 | 4,200 | 4,650 | 5,000 | 5,250 |
| Gross margin | 29.7% | 33.0% | 34.8% | 35.5% | 35.8% | 36.0% | 36.0% |
| EBIT ($m) | 69 | 513 | 853 | 1,046 | 1,185 | 1,290 | 1,355 |
| EBIT margin | 9.3% | 19.7% | 23.7% | 24.9% | 25.5% | 25.8% | 25.8% |
| Capex ($m) | 750 | 700 | 480 | 340 | 350 | 370 | 385 |
| Free cash flow ($m) | −764 | −504 | +114 | +580 | +760 | +875 | +956 |
Why 2027 is $2.6bn and not management’s number. 2026 lands at their ~$1.09bn — they have guided accurately for five straight quarters and near-term guides have earned belief. $2.6bn is the Q4 2026 run-rate of roughly $475m annualised plus about 35% growth, which is 55–60% of the mid-2027 exit run-rate of roughly $6bn annualised the company is targeting. The discount reflects three things that are not execution: DSP and TIA supply that Broadcom allocates and AAOI does not control, a fourth-source queue position at the key 1.6T account, and 20–30% annual ASP erosion per speed class that is the physics of this industry. The margin path credits 1.6T and laser mix but stops at 36%, because a company that has never printed 35% has not yet earned 40%. Working capital absorbs 28% of incremental revenue — the 147-day receivables reality — and tax is 18%.
For reference, the consensus this sits against is unusually scattered: aggregators quote AAOI’s 2027 revenue anywhere from roughly $1.65bn to well above $3bn, and published price targets run from $57.50 to $260. A spread that wide is itself information — the market has not converged on how much of the ramp to underwrite either, which is precisely the question the scenario weights encode.
Base-case cash flows, varying the build-out rate against the terminal rate. Every cell is reproducible in the panel above.
| Build-out WACC ↓ / terminal WACC → | 10% | 11% | 12% | 13% |
|---|---|---|---|---|
| 11.5% | $99 | $87 | $79 | $72 |
| 12.5% | $93 | $83 | $74 | $68 |
| 13.5% base | $88 | $78 | $70 | $64 |
| 14.5% | $84 | $74 | $67 | $61 |
FY2027 revenue against terminal gross margin, with the FY2032 endpoint scaling with the 2027 figure and everything else at base. The whole plausible argument about this stock lives inside this grid.
| FY2027 revenue ↓ / terminal GM → | 31% | 33% | 35% | 37% |
|---|---|---|---|---|
| $1.9bn bear-like | $41 | $47 | $54 | $60 |
| $2.3bn | $50 | $57 | $65 | $73 |
| $2.6bn base | $56 | $65 | $74 | $83 |
| $3.0bn | $65 | $75 | $85 | $95 |
| $3.4bn bull-like | $74 | $85 | $97 | $108 |
On the revenue-and-margin grid, the current price of $103.39 sits above every cell except the bull-like corner — $3.4bn of 2027 revenue at a 37% terminal margin, or close to it. A single-path cash-flow reading would call that overvaluation. It is not that simple, because the single path is not the asset: the booked order book, the policy option and the laser franchise give the right tail real weight, and the probability-weighted value of the whole distribution is $86, 17% below the price — inside the ±20% band we require before calling anything mispriced. A cross-check on an exit multiple instead of a perpetuity — base-case 2030 EBITDA of roughly $1.45bn at 9–15x, discounted at 13.5% — spans $84–139 per share. The honest statement of value is therefore a $60–125 band centred near the mid-$80s, with today’s price in the band’s upper half.
Working backwards from the roughly $9.2bn enterprise value at an 11% steady-state cost of capital and 3.5% growth: the market requires roughly $4.1bn of durable annual revenue at a 16% free-cash-flow margin — which is what a ~36% gross margin funds — or about $4.9bn at a 14% margin. Against $456m of revenue last year, and against management’s ~$6bn mid-2027 exit-run-rate ambition, the price is paying for 70–80% of the target becoming the permanent state of the business. That is more than our modal 55–60%, but it is not irrational: capacity through Q2 2027 is booked, $324m-plus of orders are disclosed, and the policy option costs the buyer nothing. The disagreement between our $86 and the market’s $103 is a difference of degree — how much of a booked ramp to underwrite before it ships — not a claim that the market has lost its mind. The burden of proof shifts decisively at the 5 November print.
A risk charged in the cash flows and again in the discount rate is charged twice, and double-counting is how a forensic review turns into a bear case by accident. This table states where each one sits.
| Risk | Priced in | How, and only there |
|---|---|---|
| Ramp execution and component supply | Revenue scenarios and weights | The 55–60% base haircut to management’s trajectory carries DSP/TIA gating and queue position. Not also charged in the discount rate |
| Customer concentration and order cuts | Bear scenario and its 27.5% weight | The single most probable non-base outcome gets more than a quarter of the distribution. No additional cash-flow haircut, no exit-multiple penalty |
| ASP erosion and Chinese competition | Cash flows | Revenue growth fades hard after 2028 and gross margin is capped at 36% base against the 40% target. Not also in the discount rate |
| Dilution and the financing treadmill | Cash flows | Shares fixed at 92.8m diluted with all future burn charged to the DCF — the ATM treated as NPV-neutral financing. Inflating the share count on top would charge the same dollars twice |
| Equity risk and volatility | Discount rate | The 13.5% build-out WACC. Nothing else |
| CATV cycle cresting | Cash flows | The post-2027 revenue fade assumes the DOCSIS 4.0 wave crests in 2027–28. No separate haircut |
| Valuation-relative risks — concentration and ramp versus peers | Already in the price | The discount to Lumentum and Credo compensates them, which is why they are deliberately not charged again inside the model. See Section 05 |
Probability-weighted value $86 against $103.39 — the shares sit 17% above it, inside the ±20% band, in the upper half of a $60–125 range whose floor is set by the revenue-and-margin grid’s bear-adjacent cells and whose ceiling is the exit-multiple cross-check. The band is the model run across a reasonable input range, not a separate opinion, and every end of it is reproducible in the panel above. On the arithmetic, we rate the shares fairly valued. The distribution, not the point, is the message: outcomes spanning $15 to $267 with meaningful weight on both tails, and the whole disagreement with the market compressed into one question — how much of a booked, unshipped, two-customer ramp to pay for today.
Applied Optoelectronics reported second-quarter results on 6 August 2026. Five things from the call matter more than the rest. First, on management’s telling, capacity, not demand, is the constraint: both the CEO and CFO repeated that near-term revenue is bounded almost entirely by production capacity and key component availability, with the CFO blunt that if they could produce more 800G they could ship more today, demand described as 20–40% above supply, and the factory booked through Q2 2027. Second, the 800G ramp is later and steeper than the market assumed in spring: $12.8m in Q2, guided to roughly five times that in Q3, with about $330m of combined 800G and 1.6T revenue arithmetically required in Q4 — a figure the CFO confirmed as directionally right — inside a Q4 whose total capability was put above $500m. Third, 1.6T is weeks away with $200m-plus booked, final qualification at a major hyperscaler expected within two to three weeks of the call, shipments from late Q3, more than $70m landing in Q4 gated by DSP, TIA and material supply rather than capacity, and Q1 2027 targeted to double Q4. Fourth, margins improve on mix, not yet on execution: 29.8% non-GAAP gross margin was in line, Q3 is guided flat despite ~40% sequential revenue growth because expedite fees and early-ramp costs eat the leverage, and the road to 40% runs explicitly through 1.6T, then CPO lasers at 55–65% and ELSFP modules above 50% in late 2027. Fifth, a quantified wobble: one customer cannot source enough 100G switches — a memory-shortage knock-on — removing $20–25m from Q3; without it, sequential growth would have exceeded 50%. Recovery as memory normalises is a testable claim within two quarters.
| Metric | Q2 2026 | Q1 2026 | Change | Why it matters |
|---|---|---|---|---|
| Revenue | $191.9m | $151.1m | +27% q/q, +86% y/y | Fifth straight record, in line with guidance — guidance credibility intact |
| Datacenter revenue | $107.7m | $81.4m | +32% q/q, +140% y/y | The AI engine; 56% of total |
| CATV revenue | $80.6m | $66.8m | +21% q/q, +44% y/y | Record; Q3 guided $100–110m on the DOCSIS 4.0 wave |
| 800G revenue | $12.8m | ~$6m | ~2x q/q | Tiny base, guided ~5x sequentially in Q3 — the whole thesis in one line |
| Non-GAAP gross margin | 29.8% | 29.2% | +60bp | In line; still ~10 points below the long-term promise |
| Non-GAAP net income | +$5.5m | −$4.9m | swing +$10.4m | The headline milestone — but see the earnings-quality note below |
| GAAP net loss | −$22.8m | −$14.3m | wider | GAAP and non-GAAP moving in opposite directions |
| Adjusted EBITDA | −$0.5m | +$1.0m | ~zero | The cleanest read: the business is at operating breakeven |
| Cash and restricted cash | $508.8m | $449.4m | +$59m | Rose only because $538.8m of new ATM stock was sold in the quarter |
| Capex incl. prepayments | $565.5m | ~$68m | 8x q/q | The bet is being placed now; second-half intensity guided higher still |
| Q3 guide | $255–290m / $0.11–0.26 | — | +42% q/q mid | The EPS midpoint sat below the ~$0.28 consensus — a margin story, not a demand story |
The bridge from a $22.8m GAAP loss to $5.5m of non-GAAP income includes a $14.3m tax benefit plus government subsidy income, alongside the usual stock compensation and one-offs. Management itself attributed the beat largely to foreign tax benefits and subsidies and expects them to continue. Adjusted EBITDA of −$0.5m is the honest denominator: operationally, the quarter was breakeven, not profitable. That does not negate the revenue inflection — it does mean the “returned to profitability” headline should be carried at half weight, and it is why this report anchors on cash flow throughout.
Tone was confident to the point of exuberant on demand — customers pressing for faster delivery “every week” — and notably candid on constraints: the 100G hit, the DSP and TIA gating and the expedite fees were all volunteered with numbers. Analyst concerns clustered on the potential US restriction of Chinese transceivers (answered: it can only help AAOI, though nothing is signed), on whether the 800G ramp is tooling-ready (answered: capacity, mostly not yet installed, rises from roughly 200k to more than 650k units a month by year-end), and on Chinese laser competition, where the claim that 300mW DWDM CPO lasers are two to three-plus years away for Chinese entrants was the most bullish and least verifiable statement on the call. Positive surprises: the EPS beat, a $5.7m tariff refund, CATV upside and the Mediacom win. Negative surprises: the 100G air-pocket, the flat Q3 margin guide, and the sheer scale of the capex and its equity funding.
What to watch next: confirmation of 1.6T qualification and first shipments, due in September; the Q3 print on 5 November against the $255–290m guide, where the implied Q4 matters more than the Q3 number — above $450m keeps the base case alive, below $400m breaks it; exit-2026 gross margin against the 32–33% marker; the 100G recovery; the pace of sales under the new $600m programme, visible in the Q3 10-Q; and any concrete US action on Chinese transceivers.
Applied Optoelectronics designs and manufactures the optical hardware that moves data through networks, in two very different businesses that happen to share a factory floor. The first is datacenter optical transceivers — pluggable modules at 100G, 400G, 800G and now 1.6 terabit that convert electrical signals to light and back at each end of the fibre links inside AI and cloud datacenters, sold directly to hyperscale operators. Two unnamed hyperscalers were 50% of Q2 revenue; company history and warrant disclosures point to Microsoft and Amazon as the long-standing anchor accounts, though the identities are inferred rather than confirmed. The second is cable-TV broadband equipment — most recently 1.8 GHz Quantum Bandwidth amplifiers and QuantumLink management software that cable operators, one very large MSO plus new wins such as Mediacom, are deploying for DOCSIS 4.0 upgrades. Small telecom and FTTH sales round out the mix. Crucially, AAOI is vertically integrated: it has fabricated its own indium-phosphide lasers since 1997 in Sugar Land, Texas — the component that is the industry’s binding constraint — and manufactures in Texas, Taiwan and Ningbo, China. Roughly 2.5% of revenue from certain customers is netted away as contra-revenue for warrants granted in exchange for future business, the Amazon-style alignment structure, at $1.2m in the quarter.
| FY2023 | FY2024 | FY2025 | Q1’26 | Q2’26 | Q3’26 guide | FY2026E | |
|---|---|---|---|---|---|---|---|
| Revenue ($m) | 217.6 | ~249 | 455.7 | 151.1 | 191.9 | 255–290 | ~1,090 |
| Growth y/y | −2% | +15% | +83% | +~90% | +86% | +130% mid | +~139% |
| Non-GAAP gross margin | ~26% | ~25% | ~29% | 29.2% | 29.8% | 29–30.5% | ~30–31% |
| GAAP net income ($m) | −56.0 | −187* | loss | −14.3 | −22.8 | — | loss |
| Non-GAAP net income ($m) | loss | loss | ~breakeven | −4.9 | +5.5 | +10.1 to +24.0 | ~+45–70 |
| Diluted shares (m) | ~29 | ~40 | ~57–66 | ~76 | 88.2 | ~92.8 | rising |
2023–2024 from company filings; the 2024 GAAP loss* includes non-cash convertible and impairment items. 2025–2026 from the Q2 2026 release and call; FY2026E is our base estimate consistent with management’s “around $1.1bn”. Share-count expansion — from 56.8m weighted-basic in Q2 2025 to 84.9m outstanding by 20 August 2026 — is a core part of the story, not a footnote.
AI cluster architectures are optics-hungry: each accelerator generation ships with more optical links per GPU at higher speeds. Industry forecasters put datacenter ethernet optical transceivers at roughly $20–25bn in 2026, heading toward $40bn-plus by the end of the decade at a 20–30% compound rate, with 800G and 1.6T the entirety of the growth. Nvidia has publicly reaffirmed pluggable optics through 2027 even as co-packaged optics develops — and CPO itself multiplies demand for exactly the high-power lasers AAOI fabricates. Against a ~$22bn 2026 market, AAOI’s roughly $620m of expected 2026 datacenter revenue is a ~3% share; management’s mid-2027 exit target implies 15–20% of the then-market, which is the central reason we discount the target. The market is dominated by the Chinese vendors Innolight and Eoptolink — together roughly half the 800G market — then Coherent, with AAOI qualified as the fourth source at one major hyperscaler for 1.6T. The CATV business rides a separate cycle: the North American DOCSIS 4.0 upgrade wave, worth $325m-plus a year to AAOI on management’s estimate — strong now, but an upgrade cycle that will crest, plausibly in 2027–28.
| AAOI | Coherent | Lumentum | Fabrinet | Credo | |
|---|---|---|---|---|---|
| Price, 1 Sep 2026 close | $103.39 | $272.03 | $868.95 | $402.07 | $206.63 |
| Off 52-week high | −56% | −38% | −20% | −46% | −33% |
| Market cap (approx.) | ~$9.6bn | ~$42bn | ~$60bn | ~$14.5bn | ~$36bn |
| NTM revenue (approx.) | ~$1.7–2.2bn | ~$9bn | ~$5.3bn | ~$6.0bn | ~$1.5bn |
| EV / NTM sales (approx.) | ~4.2–5.4x | ~5.0x | ~11x | ~2.3x | ~23x |
| NTM gross margin (approx.) | ~30–32% | ~40% | ~48–50% | ~12–13% | ~65% |
| Revenue growth, last reported quarter y/y | +86% | ~+50–90% | +109% | +45% | ~+100%+ |
Prices and drawdowns are exchange data at the 1 September 2026 close, the same session as every other figure on this page. Market caps, forward revenue, sales multiples and margins are approximations derived from each company’s most recently reported quarter and its own guidance, all reported in August 2026 — Lumentum FQ4 revenue ~$1.0bn guiding $1.225–1.275bn, Coherent FQ4 ~$1.98bn, Fabrinet FQ4 $1.32bn — and are flagged with a tilde accordingly.
Is AAOI cheap or expensive against this set? Both, depending on the year you price. On the next twelve months it screens expensive: the group’s lowest gross margin and highest customer concentration at a sales multiple similar to Coherent’s and roughly double Fabrinet’s. On calendar-2027 base-case earnings of roughly $4.2 a share it screens cheap — about 24x for the fastest grower in optics. That is not a mispricing; it is the market charging an execution discount: the 2027 earnings are not yet earned, rest on two customers and a five-fold ramp, and will be diluted by an open $600m programme. Our judgment is that the discount to Lumentum and Credo fairly compensates the concentration and ramp risk — which is exactly why we deliberately do not penalise those risks a second time inside the model’s discount rate. What would expand the multiple: 1.6T shipping at scale, a third hyperscaler, a Chinese-transceiver restriction, or the first clean free-cash-flow quarter. What would contract it: any order cut, gross margin stuck below 31%, or ATM sales accelerating into weakness. The whole optical complex re-rated violently in 2026 and corrected violently in the last month — every name in the table is well off its high, and the group trades far above its own history.
| Dimension | Score | Assessment |
|---|---|---|
| Intellectual property — laser fabrication | 8/10 | In-house InP laser fab since 1997, one of few globally at scale, now producing 100mW CW-WDM CPO lasers with 200mW and 300mW DWDM on the roadmap. The industry’s bottleneck component, and the single strongest element of the franchise. The claim that Chinese rivals are 2–3+ years behind on high-power CPO lasers is management’s, and unverifiable |
| Regulation & geography | 8/10 | The only US-based scaled AI-transceiver manufacturer, with Texas capacity expanding and tariff dynamics already pushing hyperscalers to dual-source domestically. A formal restriction on Chinese transceivers would convert this into a decisive advantage — this score is 10/10 if that happens and closer to 4/10 if trade policy normalises. Priced in the scenario weights, not twice |
| Switching costs | 6/10 | Qualification cycles run 6–18 months and hyperscalers rarely swap mid-generation — sticky within a speed class. But every new speed class reopens the contest, and 2018 proved requalification is a right customers exercise |
| Distribution & relationships | 6/10 | Direct, decade-plus relationships with two top hyperscalers, warrant structures aligning both sides, and in CATV a near-sole-source position in 1.8 GHz amplifiers with QuantumLink pulling equipment through |
| Customer loyalty | 6/10 | The two anchor accounts stayed through the 2018–24 trough and came back at scale — genuine loyalty, to a point. The same history shows the loyalty is conditional: orders were cut once and can be again |
| Brand | 5/10 | Credible and specific inside the customer set — the domestic vendor that makes its own lasers. Invisible beyond it, and irrelevant to the buying decision against price and qualification |
| Scale | 5/10 | Meaningful and growing — 800G capacity from ~200k to 650k+ units a month by year-end, ~2,400 added employees in a quarter — but Innolight ships multiples of AAOI’s volume and Coherent is five times its revenue. Scale is the gap being closed, not the moat |
| Cost advantage | 4/10 | Vertical laser integration should confer one; a 29.8% gross margin against Chinese rivals’ comparable-or-better margins at lower ASPs says it has not yet. US manufacturing is a cost disadvantage that policy may or may not convert into a price umbrella |
| Network effects | 2/10 | Essentially absent, as for every hardware vendor in this industry. QuantumLink software attach in CATV is the only faint trace |
Widening: every high-power laser milestone, every month of the domestic-sourcing push, and each 1.6T qualification deepens the two advantages that are real — the fab and the flag. CPO adoption, often framed as the threat that kills pluggables, would multiply demand for exactly the lasers AAOI makes at 55–65% margin; the company is engaged with four of five major CPO players on management’s account. Narrowing: everything else. Innolight and Eoptolink set the price umbrella, 20–30% annual ASP erosion is structural, and each speed transition reopens every socket. The honest summary: a narrow moat made of one excellent asset and one policy option, attached to a commodity-adjacent manufacturing business. That is why the bear case carries 27.5% and why the superbull — where the laser franchise becomes the business — is real but weighted at 5%.
| Competitor | Where they are stronger | Where AAOI is stronger |
|---|---|---|
| Innolight (Shanghai-listed) | The 800G volume leader — scale, cost, cadence, and first place in most hyperscaler allocations. Sets the market price | US soil, in-house lasers, and eligibility for any procurement Chinese vendors are excluded from. AAOI does not out-compete Innolight; it offers what Innolight cannot |
| Eoptolink (Shenzhen-listed) | Second-largest 800G supplier at Chinese cost structure; together with Innolight roughly half the market | Same asymmetry: geography, lasers, warrant-aligned direct hyperscaler relationships |
| Coherent (COHR) | ~5x AAOI’s revenue, ~40% gross margin, its own InP fab, broad portfolio, datacom optics growing 60%+ | Focus and speed at the two anchor accounts, US-only manufacturing story, and a five-quarter record of out-growing it from a small base |
| Lumentum (LITE) | The margin benchmark — ~48–50% gross margin, $60bn rating, deep laser IP of its own, 109% growth last quarter | Little head-to-head in transceivers today; Lumentum is what the superbull says AAOI’s laser business could grow into — and the market pays 11x sales for the finished article |
| Fabrinet (FN) | Contract-manufacturing scale and consistency at 2.3x sales | Different business: AAOI owns its product, its lasers and its customer relationships rather than assembling others’ |
We ran a deliberately hostile read of the release, the call and the balance-sheet data, then ran the same exercise in the other direction. This is not a fraud pattern and we are not implying one — the disclosure here is unusually granular, the wobbles were volunteered with numbers attached, and the one metric management guides has been hit five quarters running. What the concerns describe is a company financing an enormous, concentrated bet with its own equity, and reporting profits that are, for now, tax-shaped.
| Concern | Sev. | Evidence, and why it costs a shareholder money |
|---|---|---|
| Extreme customer concentration | 9/10 | Three customers were 92% of Q2 revenue — 42 / 26 / 24 — and the top ten 99%. The 2018–19 precedent shows what one order cut does: revenue collapsed and took six years to recover. A single procurement decision at either hyperscaler, or one MSO’s budget, moves everything. The single highest-severity item in this report |
| Massive dilution as the funding model | 8/10 | ~$1.03bn of ATM stock sold in H1 2026, a fresh $600m programme filed 21 August, share count up ~49% y/y and up ~3x since 2023. The 21 August filing preceded a ~30% drawdown within days. Every future dollar of capex is a claim on more issuance until free cash flow turns |
| Profits are tax-shaped; operations are at breakeven | 7/10 | The celebrated +$0.06 non-GAAP EPS contains a $14.3m tax benefit and subsidy income; adjusted EBITDA was −$0.5m; H1 operating cash flow was −$73.8m. The revenue inflection is real and the profit inflection has not happened yet |
| The ramp is a five-fold ask on a $12.8m base | 7/10 | 800G must grow ~5x sequentially in Q3 and the company must ship >$500m in Q4 — more than FY2025 in one quarter — gated by DSP and TIA supply that Broadcom allocates, tooling mostly not yet installed, and ~2,400 employees hired in a quarter. Any one bottleneck slips the plan |
| Working capital consumes the growth | 6/10 | $314.0m of receivables (~147 days of sales) and $278.8m of inventories (+35% q/q) against $191.9m of quarterly revenue. Hyperscalers pay slowly; the DSO alone is why 28% of every incremental revenue dollar is charged to working capital in the model |
| ASP erosion and the Chinese price umbrella | 6/10 | 20–30% annual per-speed-class price declines are structural; Innolight and Eoptolink hold ~half the 800G market at lower cost. AAOI’s growth requires unit volumes to outrun pricing forever, or policy to intervene |
| CATV is a cresting upgrade cycle carrying 42% of revenue | 5/10 | The DOCSIS 4.0 wave is worth $325m+/yr now and is guided to $100–110m in Q3 alone — and upgrade cycles end. A 2027–28 crest is in our fade; a 2026 crest would hit the base year |
| Q3 guidance leaned on non-operating income | 4/10 | Flat sequential gross margin on +40% revenue, EPS guided below consensus at the midpoint, with tax benefits and subsidies expected to continue doing the lifting |
| Key-component dependence on a competitor’s parent | 4/10 | DSPs and TIAs come substantially from Broadcom, which allocates industry-wide and whose own CPO systems compete with pluggables long term |
| Governance friction, minor but present | 2/10 | Related-party facilities and a history of convertible restructurings during the trough years; nothing current that rises above housekeeping |
| Strength | Evidence | Why the market may be underweighting it |
|---|---|---|
| Contracted, visible demand | Capacity fully booked through Q2 2027; disclosed orders above $324m including a single $200m+ 1.6T award; demand described as 20–40% above supply | The order book is treated as press-release fodder because 800G is only 7% of revenue — but the bookings are specific, dated and large relative to the company |
| The only scaled US transceiver manufacturer | Texas fab and assembly expanding; hyperscalers already dual-sourcing domestically on tariff arithmetic; a restriction on Chinese vendors would reallocate share by decree | A free option the buyer pays nothing for at our $86; the market prices some of it, which is part of the gap to $103 |
| In-house laser fab at the industry bottleneck | InP fab since 1997; 100mW CW-WDM CPO lasers shipping to 4 of 5 major CPO players on management’s account; 55–65% margin on CPO lasers, 50%+ on ELSFP | Invisible inside a 29.8% blended margin today; if it scales it changes the company’s margin structure and multiple — the superbull in one line |
| Five consecutive record quarters, guided accurately | Revenue guidance met or beaten five straight; +86% y/y; the near-term guide has earned belief | Execution credibility is real and recent, even if 40% gross margin remains a decade-old promise |
| Warrant-aligned hyperscaler relationships | Contra-revenue warrant structures with anchor customers — the Amazon playbook — binding both sides’ economics | Read as dilution; better read as customers paying for skin in the game |
| CATV near-sole-source position | 1.8 GHz amplifier leadership, QuantumLink attach, Mediacom added, $100–110m guided for Q3 | Dismissed as legacy; it is the profitable cash engine funding the AI ramp’s opex while equity funds its capex |
Concentration, dilution and earnings quality would individually justify a heavy discount; the contracted book, the policy option and the laser fab would individually justify a premium. They are the same facts seen from both sides: the concentration exists because two of the world’s largest technology companies pre-booked a factory, and the dilution exists because the company is building what they booked. Our resolution is structural rather than rhetorical — the concerns set the bear scenario and its 27.5% weight, the strengths set the bull tail, and neither is charged a second time in the discount rate. If you own this, the two lines to watch are the Q3 10-Q’s ATM disclosure and any change in the customer-concentration percentages.
| Area | Score | Assessment |
|---|---|---|
| Guidance accuracy | 4/5 | Five consecutive quarters of revenue guidance met or beaten, with granular per-product colour volunteered. The one asterisk: long-range aspirations — 40% gross margin above all — have been promised for the better part of a decade without being reached |
| CEO track recordDr. Thompson Lin, founder, 1997 | 3.5/5 | Founded the company, built the laser fab, kept both anchor hyperscalers through a six-year drawdown, and has now delivered the sharpest ramp in the company’s history. The same record includes the 2018 concentration crash on his watch and a long stretch of value destruction between the peaks |
| Operational execution | 3.5/5 | Five records, a ~2,400-person quarter of hiring, 800G capacity tripling in six months, 1.6T qualified to within weeks — genuinely impressive throughput. The flat Q3 margin guide shows the cost of speed: expedite fees and early-ramp inefficiency are being paid in real time |
| Capital allocation | 3/5 | Selling equity at ~8x sales to build pre-booked capacity is defensible, arguably shrewd — and it is also the third financing regime in a decade, after the converts and the trough-era restructurings. The test of this cycle’s allocation arrives when the capex must produce free cash flow in 2028 |
| CFO credibilityDr. Stefan Murry | 3/5 | Candid to a fault on the call — volunteered the $20–25m 100G hit, the DSP/TIA gating and the Q4 arithmetic. The non-GAAP presentation leans on tax benefits he expects to continue, and the “profitability” framing outran the adjusted-EBITDA reality. Note for context: the widely-circulated “CFO transition” headline dates from 2014 — Murry has held the seat since; treat recycled headlines as the data-hygiene warning they are |
| Dilution management | 2/5 | The count is up ~49% y/y and ~3x in three years, programmes are announced without ceremony, and the $600m filing landed on a Friday into a fragile tape. Shareholders fund the strategy; the strategy has yet to fund the shareholders |
| Insider alignment | 2.5/5 | Founder-scale ownership by Lin remains, but insiders have been net sellers of roughly $65m over the past year, dominated by 10b5-1 programme sales including CFO sales of roughly $12m a year, most recently at $144.79 on 10 August. Planned and disclosed — and still a signal that the people with the most information monetise into strength |
| Transparency | 4/5 | Per-product revenue, per-customer percentages, monthly capacity figures, named bottlenecks and quantified wobbles — among the most granular disclosure of any company we cover. The non-GAAP tax framing is the one deduction |
Management tells you the truth about operations and dresses up the truth about profits. The operational candour — bottlenecks, misses and all — earns the near-term guide real credibility, which is why our 2026 estimate is management’s number. The profit presentation is why our model runs on cash flow and charges every ramp dollar. Both readings are priced, neither twice.
| Event | Timing | Upside | Downside |
|---|---|---|---|
| 1.6T qualification and first shipments | September 2026 | Final qualification at the major hyperscaler lands on the two-to-three-week schedule given on 6 August, first revenue ships in late Q3, and the $200m+ order begins converting. The single nearest catalyst | A slip of even a month compresses the >$70m Q4 1.6T assumption and hands the sceptics the first missed date of the ramp |
| Q3 2026 results | 5 November 2026 | Revenue inside the $255–290m guide with 800G at roughly 5x sequentially, the implied Q4 above $450m, and gross margin holding ~30% despite ramp costs. Five records become six and the base case strengthens | The first revenue miss in six quarters, an implied Q4 below $400m, or margin below 29% — any of which breaks the “capacity is the only constraint” narrative that the entire multiple rests on |
| The Q4 arithmetic itself | Guided with Q3; reported Feb 2027 | >$500m shipped in a single quarter — more than FY2025 — with ~$330m of 800G+1.6T. Achieving it validates the mid-2027 exit trajectory and forces every model, including ours, higher | This is the quarter where tooling, hiring, DSP supply and two customers’ deployment schedules must all hold at once. It is the bear case’s best single shot |
| US action on Chinese transceivers | Open-ended; live all period | Any restriction, tariff escalation or procurement rule reallocates share toward the only domestic scaled producer — the free option in the price converts to revenue by decree | Explicit normalisation of trade removes the option value and re-exposes the cost disadvantage of US manufacturing against the Chinese price umbrella |
| 100G air-pocket recovery | Q4 2026 – Q1 2027 | The $20–25m memory-shortage deferral returns as guided — a small, clean, testable claim that validates management’s explanation | Non-recovery reveals the deferral as demand loss dressed as logistics, and re-prices the credibility of every other explanation on the call |
| ATM pace under the $600m programme | Q3 10-Q, mid-November | Measured issuance into strength, or a pause, signals the equity spigot is discretionary | Aggressive sales into the drawdown confirm the treadmill reading and cap every rally with supply |
| CPO laser and ELSFP milestones | H1–H2 2027 | Volume orders from any of the four engaged CPO players, or ELSFP ramping toward the 400k-modules-a-month 2028 target at 50%+ margin — the superbull’s ignition sequence | Milestones slipping quietly out of the narrative, as decade-old margin targets once did |
| Q4 results and initial FY2027 guidance | February 2027 | A 2027 revenue guide at or above $2.6bn puts the base case on management’s own paper, with mix and margin colour to match | A first guide below ~$2.3bn resets the market’s 70–80% underwrite toward ours, with the price adjusting the difference |
The tape it trades on, for completeness. At the 1 September close the shares sit roughly 3% above the 200-day average near $100 and about 14% below the 50-day near $120, after a −56% drawdown from May’s $233.67 — the fourth drawdown of similar depth in the stock’s history. Momentum is washed out rather than broken: RSI in the low-40s, MACD negative, one-year realised volatility around 140% against roughly 84% implied. The levels the market has actually traded: support near $100, then $88–90, then the pre-breakout shelf at $76.50; resistance at $113–115, then $122–125, then $150–155. Technicians would note the $88–100 zone is where accumulation has repeatedly appeared — and that the open $600m programme means every rally meets authorised supply. We publish this as context, not method: nothing in the band derives from it.
BULLCapacity booked through Q2 2027, disclosed orders above $324m including a single $200m-plus 1.6T award, demand running 20–40% above supply, and a $20–25bn market growing 20–30% a year. The constraint is how fast they can build, not whether anyone buys. Five record quarters say they can build.
BEARThe product carrying that story was $12.8m last quarter — 7% of revenue. You are annualising a press release. The plan needs 800G up five-fold in Q3 and more than half a billion dollars shipped in Q4, through DSPs Broadcom allocates, on tooling that mostly is not installed, by 2,400 people hired last quarter.
BULLAnd it is contracted. This is not speculative capacity hunting demand — the customers pre-booked the factory and took warrants to align themselves. When has a company this size shown you a booked five-fold ramp with the buyers’ names effectively on it?
BEARIn 2018, roughly. Same anchor customers, same direct model, same concentration — and one procurement decision cut revenue in half and the stock took six years to recover. Today three customers are 92%. Bookings are not revenue, and hyperscaler capex is the most re-forecast line in technology.
BULLThe world changed since 2018. AAOI is now the only US-based transceiver maker at scale while Washington actively reviews Chinese optics — a restriction hands them share by decree, and hyperscalers are dual-sourcing domestically already. Plus the laser fab: the industry’s bottleneck component, in-house since 1997, now shipping CPO lasers at 55–65% margin to four of the five major CPO players. That is a Lumentum growing inside a transceiver company.
BEAROn management’s numbers. The blended margin is 29.8% — lowest in the peer group — the 40% target is a decade old, and Chinese competitors hold half the 800G market at prices AAOI cannot match without policy help. As for profits: the celebrated six cents contains a $14.3m tax benefit. Adjusted EBITDA was negative half a million. Operating cash flow was minus $74m in the half. This company has never yet turned growth into cash.
BULLBecause it is building the machine that does. The capex is the thesis — $565m in one quarter, prepaid equipment, capacity tripling — funded at 8x sales in an ATM that cost the buyers, not the company. First positive free cash flow lands in 2028 on the base path, and every quarter of records de-risks it.
BEARFunded by selling shares — a billion dollars in six months, count up 49% in a year, a fresh $600m programme filed on a Friday that cost the stock 30% in days. Insiders sold $65m alongside, the CFO at $144.79 three weeks before you could buy at $103. And the price still assumes $4bn-plus of durable revenue. The reverse DCF says the market pre-pays for three-quarters of a ramp that is 7% shipped.
CHAIRBoth cases survive contact with the evidence, which is why the distribution is wide and the rating is fairly valued rather than a fudge. The bear owns the base rates: concentration at 92%, a five-fold sequential ask, tax-shaped profits, equity-treadmill funding, and a 2018 precedent inside the same customer set. None of that is rebutted; it is why 27.5% of the weight sits on a $15 outcome and why the money is charged before the revenue. The bull owns the specifics: a booked factory, dated orders with dollar signs, a laser fab the industry needs, a policy option that costs nothing, and five quarters of guided delivery. That is why the right tail carries 27.5% too. Where the bear overreaches: treating the dilution as pure value destruction when it is priced financing of pre-sold capacity, and charging the concentration twice — once in fear, once in the discount rate. Where the bull overreaches: annualising the exit-run-rate as though mid-2027 were the permanent state, and taking the two-to-three-year Chinese laser gap on faith from the party that benefits from the claim. At $103.39 against an $86 weighted value the shares sit inside the band, in its upper half, and the 5 November print does more to resolve this than anything either side can add.
Three things, none resolvable from public information today. The hyperscalers’ 2027 deployment schedules — the orders are booked, but delivery cadence belongs to two customers’ internal plans. The Chinese laser gap — whether high-power CPO lasers are genuinely two to three-plus years away for Innolight’s suppliers is management’s claim, and the superbull leans on it. Washington — whether the reviewed restriction on Chinese transceivers becomes rule, and when. Anyone quoting a point target on this name is expressing a view on all three and calling it a valuation.
One. The 1.6T qualification confirmation, due within weeks of the August call — the nearest dated, falsifiable claim. Two. The implied Q4 guide on 5 November — above $450m keeps the base case, below $400m breaks it. Three. Any concrete US action on Chinese transceivers. Four. The ATM pace in the Q3 10-Q, which reveals whether the equity spigot is discretionary or structural.
Applied Optoelectronics makes two kinds of hardware. The first is optical transceivers — small plug-in devices that turn electrical signals into light and back, so data can travel on fibre between the computers inside AI datacenters. Amazon-and-Microsoft-sized cloud companies buy them by the hundred thousand. The second is equipment that cable-TV companies install to upgrade home internet speeds. Unusually, it also makes its own lasers — the hardest part inside a transceiver — in its own Texas factory, and it is the only American company making AI transceivers at scale. Almost everyone else makes them in China.
Mostly from very few, very large customers. Last quarter one cable company was 42% of revenue and two cloud giants were another 50% — three customers, 92% of the business. The customers have booked the factory’s entire output through the middle of 2027, which is wonderful for visibility and dangerous for dependence: if any one of the three changes its plans, it moves everything.
The whole argument is about one gap. The company plans to grow roughly five-fold by mid-2027, and the orders for that growth are genuinely booked. The share price already assumes about three-quarters of that plan succeeds permanently. We assume a bit more than half. Nobody — not us, not the market, not management — knows yet, because the product doing the growing has barely started shipping. The 5 November results are the next real test.
Right: the new top-speed product passes its final checks in September and ships on time, the November results deliver a sixth record with a strong December-quarter forecast, and Washington restricts Chinese-made transceivers — which would hand the only American maker a windfall. Any two of those together would likely send the shares well higher.
Wrong: a big customer trims orders — it happened in 2018 and the stock needed six years to recover — or the factory ramp slips on parts it does not control, or the company keeps selling new shares faster than the business grows into them. In our worst case the shares are worth about $15.
| Status | What it covers |
|---|---|
| Verified from company disclosures | All quarterly revenue, product-mix dollars, customer-concentration percentages, margins, GAAP and non-GAAP income, the $14.3m tax benefit, cash and restricted cash, debt, receivables, inventories, PP&E and prepayment balances, capex, the at-the-market programme amounts and the 21 August $600m filing, share counts, warrant contra-revenue, capacity figures and booking statements, the $200m+ 1.6T order and $324m+ order disclosures, guidance ranges, and every management quotation-in-substance in Section 04 — taken from the Q2 2026 earnings release and conference call of 6 August 2026 and the Form 10-Q for the quarter ended 30 June 2026, cross-checked against the Q1 2026 release and prior-year figures. |
| Verified from market data | The $103.39 closing price of 1 September 2026 and the daily series behind it; the $18.50–$233.67 52-week range; peer closing prices for Coherent, Lumentum, Fabrinet and Credo at the same session; and realised-volatility calculations from the price series. Analyst estimate ranges and price-target extremes are quoted from public aggregators as at end-August 2026 and attributed as approximations. |
| Our estimate or judgement | Enterprise value, net cash and every multiple in Section 01; FY2026E of ~$1.09bn; the entire scenario engine — the four revenue paths, margin fades, cost, depreciation and capex ratios, the 28% working-capital charge, 18% tax, the 13.5%/11% two-stage discounting and the 2 September valuation-date conventions; the 27.5 / 45 / 22.5 / 5 judgment weights; the $60–125 band and $86 weighted value; the reverse-DCF readings; peer-table derived ratios; and all scores in Sections 06, 07 and 08. |
| Still open | The two hyperscaler customers are inferred from company history and warrant disclosures, not confirmed by the company, and their 2027 deployment schedules are unknowable from outside. The convertible notes’ conversion terms were not re-verified; the $129.1m is treated as equity within the ~92.8m diluted count per the company’s own convention, and a stricter debt treatment would lower net cash and each scenario by roughly $1.40 a share. CPO-laser margins, the 400k-modules-a-month 2028 target and the two-to-three-year Chinese laser gap are management statements that cannot currently be verified independently; the superbull leans on them and is weighted accordingly. Third-party TAM figures are forecasts, not observations. Quarterly splits of FY2025 revenue used in year-on-year growth calculations are approximate. The scenario weights are judgment, stated in Section 03 and held fixed through this report; reasonable people can weight the same four paths differently, which is why each path is separately loadable in the model. |
Method. Probability-weighted scenario discounted cash flow, methodology version 1.2. Four explicit scenarios, each a full free-cash-flow path over H2 2026 plus six calendar years, with terminal value as a perpetuity at a stepped-down terminal WACC; committed second-half 2026 spend of −$764m charged in every scenario; working capital at 28% of incremental revenue; cash tax 18%; shares fixed at ~92.8m diluted with future burn charged to the cash flows. Discounting from a 2 September 2026 valuation date with H2 2026 at t=0.15, calendar years mid-period and the terminal at t=6.04. Scenario values of $15 / $78 / $148 / $267 are blended at 27.5 / 45 / 22.5 / 5 to the $86 weighted value; the $60–125 band spans the sensitivity grids and the exit-multiple cross-check. The reference price used throughout is the $103.39 close of 1 September 2026, the last completed session before publication, and every multiple in this report scales directly with it.