FINANCIAR
Applied Optoelectronics — from 223 to 67 dollars. What the factories must deliver by 2027
Applied Optoelectronics almost doubled its sales in a year and lost 70% of its market value in five months, because the market no longer pays for promised capacity, only for modules delivered and profit left per share.
The analysis follows four threads: what the company sells and to whom, what the figures reported to the US securities regulator (SEC) show, how the expansion was financed and why the stock fell. At the end we give three scenarios for the price a year from now, the 5 November test and a predictions table with the thresholds at which we consider ourselves wrong.
The testable thesis: sales growth is real and documented, but so far it has been bought with shareholders' money, not with profit. Value per share depends on three things that will show in the next two quarters: whether the new factories produce at volume, whether the margin returns to 30% and above, and whether the company stops issuing shares.
What the company does and who it sells to
Applied Optoelectronics, based in Sugar Land, Texas, makes optical transceivers: modules that turn an electrical signal into light and back. In an artificial-intelligence data centre, thousands of graphics processors exchange data over fibre-optic cables, and each end of a cable needs such a module. The current generation carries 800 gigabits per second, and the next, 1.6 terabits, starts shipping in 2026.
The company also has a second, older business: equipment for cable-television networks, sold to US operators. Both go through the same industrial recipe, because the company makes its own lasers, in the US, Taiwan and China.
In the second quarter of 2026, data centres brought in 107.7 million dollars, or 56.1% of sales, and cable television 80.6 million, or 42.0%. A year earlier the proportion was the other way round, according to the quarterly report filed with the SEC.
Customer concentration is extreme. The top three customers brought in 42%, 26% and 24% of the quarter's sales, and the top ten 99%. The largest is Digicomm, the distributor of the cable-television equipment, with 42.8% of first-half sales. The company does not name its data-centre customers, but in March 2025 it gave an Amazon subsidiary the right to buy up to 7.95 million shares at 23.70 dollars. Most of that right vests only if Amazon buys 4 billion dollars of products over ten years.
What the reported numbers show
| Quarter | Sales ($M) | Gross margin | Net result ($M) | Shares outstanding (M) |
|---|---|---|---|---|
| Q2 2025 | 103.0 | 30.3% | −9.1 | 62.4 |
| Q3 2025 | 118.6 | 28.0% | −17.9 | 68.3 |
| Q4 2025 | 134.3 | 31.2% | −2.0 | 75.2 |
| Q1 2026 | 151.1 | 29.1% | −14.3 | 80.2 |
| Q2 2026 | 191.9 | 27.7% | −22.8 | 84.6 |
Source: the company's quarterly and annual reports to the SEC, XBRL data. The margin is the accounting (GAAP) one. Shares are those stated on the report cover, at a date early in the following quarter.
Three things stand out in the table. Sales grow every quarter: the company has reported five quarterly records in a row. The gross margin does not rise with them; it has fallen from 31.2% to 27.7%, because new factories start slowly and cost money before they produce. And the accounting loss has deepened, even though the result the company adjusts, excluding the cost of shares given to employees and other items, turned positive in the second quarter.
Cash tells the same story. In the first half of 2026, operations consumed 73.8 million dollars, and investment in factories and equipment was 335.1 million: 169.0 in the US, 62.8 in Taiwan and 103.3 in China. Inventories rose to 278.8 million and receivables to 314.0 million, of which 211.0 million from Digicomm. The quarterly report says the company gave this customer longer payment terms than usual.
The promise: capacity, orders and 1.1 billion
On 9 March 2026 the company announced its first volume order for 1.6-terabit modules, worth more than 200 million dollars, from a large data-centre customer, with deliveries from the third quarter to the fourth, according to the company's release. At the time, management promised more than 500,000 modules of 800 gigabits and 1.6 terabits a month by the end of the year.
In August the promise rose to about 650,000 modules a month by the end of 2026, from a total capacity that was then approaching 200,000. Chief financial officer Stefan Murry said demand would exceed capacity until mid-2027, according to the second-quarter results release. On the analyst call, chief executive Thompson Lin repeated that 2026 sales would be “around 1.1 billion dollars”.
For the third quarter the company forecast sales between 255 and 290 million dollars and an adjusted gross margin between 29% and 30.5%. Simple arithmetic shows how high the bar is: the first half brought in 343.1 million, the midpoint of the third-quarter forecast is 272.5 million, so for 1.1 billion the last quarter has to bring in about 484 million. That is two and a half times second-quarter sales.
Who paid for the factories
The expansion was financed mainly by selling shares directly on the market, at the day's price. In 2026 the company completed three such programmes. The first, of 500 million dollars, ended on 2 April, with 4.8 million shares sold at an average price of 103.51 dollars. The second, of up to 600 million, started on 14 May. Together they raised 1.03 billion net in the first half.
The third programme, of 600 million, announced on 21 August, ended on 5 October: 5.69 million shares at 105.36 dollars on average, with net proceeds of 588 million, according to the company's announcement. In total, more than 1.6 billion dollars in ten months. Shareholders' equity rose from 733.9 million at the end of 2025 to 1.67 billion on 30 June.
For shareholders, the effect shows in the share count. There were 49.7 million in February 2025 and 84.6 million in August 2026, and with the latest programme there are about 90 million. The company's third-quarter forecast uses 92.8 million diluted shares. At 67.51 dollars the company is worth about 6.1 billion dollars on the market, roughly ten times sales over the last four reported quarters (595.9 million).
The money went into factories. Besides the building in Sugar Land, in September the company bought a building in Houston for 26.8 million dollars, leased five new industrial buildings in the same area and a 38,312-square-metre factory in Ningbo, China, for ten years, according to current reports filed with the SEC in August and September.
Why the stock fell 70%
The rise came in two steps. On 27 February 2026, after the 2025 results and a first-quarter forecast above expectations, the stock rose 56.9% in a single day, from 53.69 to 84.23 dollars. Then the 1.6-terabit orders and enthusiasm for everything related to AI infrastructure took it to 223.10 dollars on 13 May, according to the price series in the olivLaw database.
The fall has at least three causes. The first is dilution: each new programme was followed by declines, and on 24 August, the first trading day after the third was announced, the stock closed 13.8% lower, at 107.63 dollars (247 Wall St. described the intraday fall). The second is the margin: sales rose, but the gross margin fell two quarters in a row. The third is the sector.
The whole optical industry has de-rated since May and June. At the close on 23 October, Coherent was 51% below its 2026 peak, Fabrinet 58%, Credo 51% and Ciena 49%. Lumentum, which peaked only on 6 October, was 33% lower. Applied Optoelectronics lost more than all of them, because it started from the biggest rise and has the weakest profit base. In October, B. Riley cut its price target from 109 to 92 dollars, and the average analyst target cited by MarketBeat was 147 dollars.
The company's history also explains the market's caution. In 2017 the stock closed at 99.61 dollars on 1 August, then Amazon reduced its orders for 100- and 40-gigabit modules. The company cut its forecasts, and the stock lost 37% in October 2017 alone, according to Motley Fool. By July 2022 it had fallen to 1.50 dollars. A single large customer can make or break a whole year.
What comes next: the 5 November test
The company publishes its third-quarter results on 5 November 2026, after the market close. Four numbers matter more than the rest: sales against the 255–290 million range, the gross margin against 29–30.5%, the forecast for the last quarter and any sign of a new share-sale programme. A fourth-quarter forecast well below 484 million would in effect mean abandoning the 1.1 billion target.
Two more risks are worth watching. The first is the customer Digicomm: almost half of sales and two thirds of receivables depend on a single distributor, with extended payment terms. The second is the dependence on China, where an important part of production and new investment sits, in a sector where US tariffs have changed several times in the past two years.
A third thread may work in the company's favour. On 4 August 2026 Reuters reported that the Federal Communications Commission (FCC) is preparing to bar imports of new models of Chinese optical transceivers, and officials hope to publish the rule, which would then take effect, this year, according to the report carried by GV Wire. The same day, Applied Optoelectronics stock rose 19.4%. On 25 September four senators proposed removing InnoLight and Eoptolink products from federal national-security systems, with a five-year transition.
The stakes are high: Chinese makers hold almost two thirds of global shipments of data-centre transceivers, according to Counterpoint Research. The rule has not been adopted, however, and the bank Jefferies sees a low risk that it will be applied. Even if adopted, a ban helps only a company with spare capacity, and Applied Optoelectronics says it will have none until mid-2027. Nor is it clear whether the rule would cover modules made in China by US companies, and part of the company's production is in Ningbo.
What the olivLaw engine says (experiment)
The olivLaw forecasting engine, which runs internally, answers a different question from the scenarios below: how likely the price is to touch a threshold at least once within a window, not to be there at the end. That is why both directions can have high chances at the same time. We publish it as an experiment. We ran it on the stock's monthly history since 2013, with data up to 23 October 2026 and a starting price of 67.51 dollars, under the same rules as in the evaluation of the 500 US stocks olivLaw tracks.
| Threshold touched at least once | Within 3 months | Within 6 months | Within 12 months | Random walk, 12 months |
|---|---|---|---|---|
| +48% ($100) | 49% | 65% | 82% | 77% |
| +20% ($81.01) | 72% | 81% | 90% | 89% |
| −20% (below $54.01) | 60% | 68% | 76% | 87% |
| −41% (below $40) | 30% | 45% | 58% | 69% |
How to read it. On the upside, the engine sits close to the random walk or slightly above it. On the downside, it sits clearly below: it gives 58% for a fall below 40 dollars within a year, against 69% for the random walk.
The test on past data says how far to trust that lean. We checked the engine on Applied Optoelectronics' history: at 102 month-ends from 2016 to 2026 we asked it 2,340 questions of the kind “does the stock touch +5, +10 or +20%, or −5, −10 or −20%, within 1, 3, 6 or 12 months?”, each time with only the data available then. The Brier score, where lower is better, was 0.192 for the engine and 0.196 for the random walk. The advantage comes only from the 1- and 3-month horizons; at 6 and at 12 months the random walk came out slightly better overall.
On falls of 10% within 6 and within 12 months, however, the engine did worse than the random walk (Brier 0.171 against 0.153 and 0.115 against 0.102), because it systematically underestimated this stock's falls. That is why on the downside we rely on the random walk, not on the engine.
What the engine does not see. It learns only from the stock's past prices and their volatility: it does not read the results, the dilution or the FCC decision. Nor do the company's filings improve its forecasts: Applied Optoelectronics files a material report almost every month, so that information does not tell the months apart. The results are marked internally as “diagnostic” and do not change the scenarios below.
Three scenarios for 26 October 2027
The scenarios are defined on the same variable and at the same date: the closing price of AAOI stock on Nasdaq on the last trading day up to 26 October 2027. The three ranges cover every outcome, and the probabilities add up to 100%.
- A · Capacity delivers, above 100 dollars (30%): the factories get close to 650,000 modules a month, the 1.6-terabit modules pass qualification at several customers, the margin climbs towards 35%, and the company no longer needs a new share-sale programme. An FCC ban on Chinese models would push in the same direction.
- B · Growth, but diluted, between 40 and 100 dollars (40%): sales grow strongly, but below the 1.1 billion target, the margin stays around 30%, and the company sells more shares to finish its factories.
- C · Delays and dilution, below 40 dollars (30%): qualification or production slips, a large customer cuts its orders as in 2017, module prices fall faster than costs, and financing happens at ever lower prices.
Where the numbers come from. The statistical fan is a random walk without drift, using the stock's recent volatility of 114% a year. It gives 37% for A, 31% for B and 32% for C. With the average volatility of 2017–2025, 94% a year, it gives 34%, 37% and 29%. We move a few points from A towards the middle on fundamental grounds: to pass 100 dollars the company would have to deliver almost everything it promised, and the gain per share is reduced by dilution. We round to 30, 40 and 30.
The causal chain
Links 3, 5 and 6 are documented in the reports filed with the SEC and in the company's releases. Links 4 and 8 are measured in the price series. Link 2 is management's promise, not a verified fact. Link 7 combines reported figures (the margin, the loss) with a connection in time (the fall of the whole sector), which we cannot split exactly between causes. Link 9 is our arithmetic on the company's own numbers.
What it means for an investor in Romania
The stock can be bought through brokers that offer access to the US market, in dollars. At the current volatility, a fall of a quarter within a few weeks is not a rare event but an ordinary one: over the past three weeks the stock has lost almost half its value. The leu–dollar exchange rate adds a separate risk. This analysis is not an investment recommendation.
Predictions table
| Prediction | Horizon | Probability | How it is checked | Verification source |
|---|---|---|---|---|
| Third-quarter 2026 sales are at least 255 million dollars | by 30.11.2026 | 70–80% | GAAP sales in the third-quarter results release | sec.gov (8-K) |
| The third-quarter adjusted gross margin is at least 29% | by 30.11.2026 | 60–70% | The non-GAAP margin in the same release | sec.gov (8-K) |
| The midpoint of the fourth-quarter sales forecast is at least 400 million dollars | by 30.11.2026 | 20–30% | The range published in the third-quarter results release | sec.gov (8-K) |
| 2026 sales reach at least 1.1 billion dollars | by 31.03.2027 | 10–20% | Annual sales in the 2026 annual report | sec.gov (10-K) |
| The company announces a new share-sale programme | by 30.06.2027 | 70–80% | A current report or prospectus for an offering of new shares | sec.gov (8-K, 424B5) |
| The company reports an accounting net profit in Q4 2026 or Q1 2027 | by 31.05.2027 | 35–45% | GAAP net result in the quarterly releases | sec.gov (8-K) |
| Scenario A: AAOI stock closes above 100 dollars | 26.10.2027 | 25–35% | The Nasdaq close on the last trading day up to 26.10.2027 | nasdaq.com |
| Scenario B: AAOI stock closes between 40 and 100 dollars | 26.10.2027 | 35–45% | The same close | nasdaq.com |
| Scenario C: AAOI stock closes below 40 dollars | 26.10.2027 | 25–35% | The same close | nasdaq.com |
When we consider the analysis wrong
We fixed the thresholds in advance, so as not to interpret them after the fact.
- Our caution is wrong if third-quarter sales exceed 290 million, the adjusted margin exceeds 30.5%, and the company announces no new share-sale programme by 30 June 2027.
- The growth story is called into question if third-quarter sales come in below 255 million or if the fourth-quarter forecast is below the third quarter's.
- The customer risk materialises if one of the top three customers falls below 10% of sales in a quarter of 2027, as in 2017.
Main sources: the quarterly, annual and current reports Applied Optoelectronics filed with the SEC, the company's releases at investors.ao-inc.com and the daily price series in the olivLaw database. The adjusted (non-GAAP) figures are defined by the company and are not audited.
.jpg?width=1200)
.jpg?width=1200)
