Applied Optoelectronics: a capacity race funded by new shares

Revenue is up 86% and management says demand will outrun capacity through mid-2027. Free cash flow over the last twelve months is minus $900 million. The first number explains the capital spending. The second explains why the share count rose 20.8% in under ten months.

At the October 12, 2026 close the shares trade 7.1% above my triangulated Base value of $94, a downside of 6.6%. The Bear value of $23 is 77.1% below the price and the Bull value of $168 is 67.0% above it. Confidence is grade D (9/25), which carries a required margin of safety (MOS) of 75%. That is above the framework's 60% ceiling, so I issue no maximum buy price and no buy zone. The stock fell 8.2% on October 12 and is 35% below its August 17 close of $154.89, so the price is close to fair for a business that has not yet shown it can turn capacity into cash.

Valuation route: Semiconductors and ecosystem, subtype FAB/IDM-like optical manufacturing (in-house lasers, transceiver assembly). Economic classification: Standard (4/12, one point below the Quality Growth line). Primary method: normalized 2027 EV/EBITDA after the committed 2026 build. Secondary method: ten-year free cash flow model. Cross-check: forward P/E. Values are present fair value estimates, not 12 month price targets. All figures are in US dollars.

What Applied Optoelectronics is

Applied Optoelectronics (AOI, ticker AAOI) grows its own laser chips and builds them into optical transceivers, the modules that turn electrical signals into light for data center links at 400G, 800G and 1.6T speeds. It also sells amplifiers and other equipment to cable operators (CATV) and a small amount to telecom. Engineering and manufacturing sit in Sugar Land, Texas, Taipei and Ningbo, China, with research in Atlanta. In the second quarter Datacenter revenue was $107.7 million (up 140.4%), CATV was $80.6 million (up 43.8%) and Telecom was $3.4 million.

Management's capacity numbers are the story. Combined 800G and 1.6T capacity was close to 100,000 units a month at the end of the first quarter and is approaching 200,000 now. The targets are more than 650,000 units a month by the end of 2026 and more than 930,000 by the end of 2027, with over half of that in Texas. The company also holds more than $200 million of 1.6T orders and expects more than $70 million of 1.6T revenue in the fourth quarter.

Concentration is the other side of it. In the second quarter one CATV customer was 42% of revenue and two data center customers were 26% and 24%, so three customers were 92%. The top ten were 99%. Management says demand will outpace production through mid-2027. With three buyers behind 92% of sales, that statement describes three purchasing plans, and the customers are not named. A warrant held by Amazon covers 7.9 million shares at $23.6956. Of those, 6.6 million vest only if Amazon and its affiliates buy $4 billion of products over up to ten years.

The latest quarter

Two details do not fit the headline. Gross margin fell year over year while revenue nearly doubled. Management cites data center mix and supplier fees as near-term headwinds and said near-term margin should be 32% to 33% and tied the 40% long-term goal to 1.6T and laser products that ramp around late 2027. Second, the third-quarter guide includes a $20 million to $25 million drop in 100G revenue after a customer's switch shortage linked to memory supply.

Why GAAP earnings and headline cash flow cannot be used

GAAP earnings were negative in every year from 2021 to 2025, and non-GAAP net income turned positive only in the second quarter ($5.5 million, after a loss of $4.9 million in the first). The hardest metric for this business is free cash flow after capital spending, stock-based compensation (SBC, pay in shares) and the working capital the growth absorbs. The most generous metric is non-GAAP EPS, which excluded $28.3 million of items in the second quarter alone. Free cash flow cannot anchor a value today because it is deeply negative, so I anchor on normalized 2027 EBITDA after deducting the capital still to be spent, and I test it against a ten-year cash flow model.

Capital spending here includes equipment prepayments, as the aggregator's cash flow lines do. Operating losses ran from $41.4 million to $70.9 million in each of 2021 to 2025. There is no through-cycle year of positive earnings to normalize, so a mid-cycle EBITDA cannot be measured. The framework asks for a full cycle or at least five years of history, and this company has neither, which is a stated warning.

Balance sheet and shares

The balance sheet changed in 2026. Property, plant and equipment rose to $697.1 million from $376.1 million at the end of 2025, and other assets rose to $330.5 million from $58.5 million. The release does not explain that second jump, but first-half equipment prepayments of $289.7 million fit most of the $272.0 million increase. Inventory is $278.8 million, up 52.3% since December against revenue up 42.9% since the fourth quarter. Receivables are $314.0 million, and $211.0 million of that, about 67%, is owed by one counterparty, Digicomm International, according to the 10-Q.

Shares outstanding went from 75.0 million at the end of 2025 to 90.6 million, a rise of 20.8%. Under a $600 million at-the-market (ATM) program disclosed on August 21 and completed on October 5, the company sold 5.69 million shares at an average of $105.36, for about $588 million net. It was the third ATM program this year. With the first-half proceeds, about $1.6 billion has come from new shares in 2026.

Because the notes are far in the money, I treat them as shares and remove their principal from debt. The count moves with the value being tested (97.9 million to 103.1 million across my cases), because the warrants and notes dilute more at higher values. As a check, the company's own third-quarter guidance uses about 92.8 million diluted shares, a weighted average that includes only part of the August to October issuance.

Liquidity is the issue. The next table is my Base bridge from the June balance sheet to the end of 2026. The capital spending and working capital lines are my estimates, because management has given no total capital spending figure. It said second-half capital intensity will exceed the first half's $625 million and that funding will come from cash, operating cash flow, equity and additional debt.

The $1.0 billion rests on one cross-check. Growing from about 200,000 to 650,000 units a month of 800G and 1.6T capacity in six months, at roughly $3,000 to $3,500 of spending per unit of monthly capacity (my estimate from the first-half spending), implies $1.35 billion to $1.6 billion, or $1.1 billion to $1.3 billion after the $289.7 million already prepaid. If anything my $1.0 billion looks light, which is why the robustness table tests $1.3 billion. The working capital line assumes net working capital falls from about 41% to 30% of annualized revenue, since it was $314 million on the second-quarter run rate. If I include 2027 and 2028 cash flows from my Base model, the cumulative need beyond today's liquidity is about $676 million through 2028, which at $100.68 is about 6.7 million more shares, or 6.8% dilution. I do not add those shares to the count, because the cash flows that create the need are already in the value.

Primary valuation: normalized EV/EBITDA

Enterprise value (EV) is market value plus debt minus cash. I apply a multiple to normalized 2027 EBITDA (earnings before interest, taxes, depreciation and amortization, and before SBC), then add the net position from the bridge above, so that the build needed to produce that EBITDA is paid for. 2027 is not a mid-cycle year. It is the first year in which the announced capacity should be running, which is the best available stand-in. The multiples are assumptions: 13x in the Bear case, 18x in the Base and 23x in the Bull. They sit below what I would give a business with 50% gross margins because this company has a 30% gross margin, no profitable year and three customers behind 92% of revenue. The grid below tests them.

EBITDA is revenue times gross margin, less operating expenses, plus depreciation ($100 million, $115 million and $130 million). Each value is EBITDA times the multiple plus the net position, divided by the share count at that value. The net position is -$182 million in the Bear, -$162 million in the Base and -$346 million in the Bull, since the Bull needs more capacity and working capital. The primary Base is $111. Base revenue is the third-party consensus of $2.66 billion, which is 33% above the fourth-quarter run rate implied by management's more than $500 million. That consensus is from an August 19 snapshot, so it is stale and labeled estimated.

The capacity numbers give a sanity check. Average 2027 capacity of about 790,000 units a month is 9.5 million units a year. If every unit of data center revenue came from an 800G or 1.6T module at an average $400 (my assumption, not a company figure) and CATV gave $350 million, the Base needs 5.8 million units, or 61% utilization. The Bear needs 3.5 million units (37%) and the Bull 7.1 million (75%).

Sensitivity: value per share

The price of $100.68 falls between the 14x and 18x columns at my Base EBITDA, and at $450 million of EBITDA it needs more than 22x.

Secondary valuation: ten-year free cash flow model

The secondary method does not use the EBITDA multiple. It discounts ten years of free cash flow after taxes, capital spending, working capital and SBC, adds the same liquidity, subtracts the same second-half 2026 spending, and divides by the share count at that value. Explicit forecasts run for 2027 to 2029, then growth fades to 4% by 2036, followed by a terminal value.

The secondary Base is $62. The Bear value of $2 is the arithmetic result of about $1.6 billion of capacity earning an 8.5% operating margin, and it implies a terminal value of only 2.7 times EBITDA. I do not floor it, but it is the least reliable number in this article. The Base terminal value is 6.2 times 2036 EBITDA, which is conservative for a business at 22.5% operating margin and is a consequence of the 11% discount rate, 7% capital intensity and 21% tax rate. I gave the Base a cash tax rate of only 5% to 15% in the early years, to reflect an accumulated deficit of $527 million.

Sensitivity of the Base model: value per share

Historical cross-check and growth diagnostics

A historical multiple check is not meaningful, because the company has had no positive earnings in the five years of data I have, so the historical weight goes to a forward P/E cross-check. It uses 2027 non-GAAP EPS from my own scenarios, which are $1.61 in the Bear, $4.55 in the Base and $6.60 in the Bull (consensus is $4.60), at 16x, 22x and 28x. That gives $26, $100 and $185. These EPS figures add SBC back, as non-GAAP EPS does.

The growth valuation cross-check is a diagnostic only. The forward P/E is 21.9x on consensus 2027 EPS of $4.60, and consensus 2026 EPS is $0.68, so EPS growth of about 577% gives a PEG of 0.04. That supports nothing. GAAP EPS is negative (-$0.79 over twelve months), the growth comes from a very low base, and non-GAAP excludes real costs. Growth durability: uncertain. Cyclical distortion: high, because AI infrastructure spending is cyclical and the company sells to very few customers. Flags: LOW_BASE, PEG_NOT_MEANINGFUL on GAAP, ACCOUNTING_BASIS_MISMATCH. This check does not set fair value.

The free cash flow growth quality check says conventional growth rates are not meaningful, because free cash flow is negative and was -$900.5 million over twelve months. The useful figure is the path to positive free cash flow. In the Base it is about -$1.9 billion in 2026, -$483 million in 2027 and -$31 million in 2028, then $297 million in 2029 (7.7% of revenue) and $605 million in 2030 (14.0%). On the $9.9 billion economic market value, that is a free cash flow yield of -4.9% in 2027, 3.0% in 2029 and 6.1% in 2030. Funding the early years adds about 6.7 million shares, so 2030 free cash flow per share is about $5.75. Stress test: cutting free cash flow from 2029 onward by 20% gives a Base DCF of $48 and by 40% gives $35. Distortion flags: low base, capital spending timing high, working capital high, dilution high. Estimate reliability: low, since management has not guided to total capital spending and consensus is dated. Classification: structural with execution risk, diagnostic only.

Valuation triangulation

The primary and secondary Base values differ by 56.9% (measured against their midpoint), well above the 30% divergence threshold. I flag it as VALUATION_DIVERGENCE: Confidence falls a grade and the margin of safety rises. The two methods answer different questions. The EV/EBITDA method says what the market pays for 2027 earnings in this part of the market. The cash flow model says what the cash flows justify at an 11% discount rate. The P/E check lands near the primary, 10.4% apart, but it uses the same operating assumptions, so it is not independent evidence. I kept the 50/30/20 weights I use elsewhere on this site rather than adjusting them to this result. The unrounded values are $23.10, $94.04 and $168.17.

What is the market already pricing in?

At $100.68 and 98.4 million economic shares, market value is about $9.9 billion. After the committed 2026 spending, enterprise value is about $10.1 billion, or 16.3 times my Base 2027 EBITDA. At the Base multiple of 18x the price requires EBITDA of about $560 million, 9% below my Base. At 13x it needs $775 million, and at 23x it needs $438 million.

The cash flow view is stricter. My Base cash flows are worth the current price only at a discount rate of about 8.3%. Alternatively, at 11% the price needs revenue and capital spending about 1.68 times my Base path, which means 2027 revenue of about $4.5 billion. At my assumed $400 a unit, full use of the announced 2027 capacity produces about $4.1 billion, so that path is out of reach unless the average price per unit is above $400. A buyer today needs five things to be true: the AI optics ramp reaches consensus revenue, gross margin climbs from 30% toward 35% to 40% as the 1.6T mix grows, the capacity is filled without price concessions, customers keep placing orders at this concentration, and the market keeps paying a growth multiple after the capital has been spent.

Bear, base and bull scenarios

I assign no probabilities. The Bear case is a stalled ramp after most of the capacity has been paid for, not a collapse of AI demand, and the ratio of Bull to Bear is 7.3 to 1. The price sits just above the Base, so the upside to the Bull case of 67% is paired with a downside to the Bear case of 77%.

How the buy price is set

A Confidence D score carries a base margin of safety of 50% or more, and I start at 50%. I add 5 points for dependence on capital markets (funding the build needs about $0.7 billion more by 2028), 5 points for operating leverage and cyclicality, 5 points for customer concentration (three customers are 92% of revenue) and 10 points for valuation divergence above 30%. No positive modifier applies. The required MOS is 75%. Above 60%, the framework prefers a NO-BUY or WATCHLIST verdict to a pseudo-precise bargain price. No maximum buy price and no buy zone are issued. For orientation only, a 60% discount to the Base would give $37.61. That is not the real 75% requirement and must not be used as an automatic order or price alert.

Dividend and capital return

Applied Optoelectronics pays no dividend. Capital goes into capacity, and the funding comes from the market for its shares.

Reasons to own AAOI

• Demand shows in the guidance. The third-quarter revenue guide of $255 million to $290 million is 42% above the second quarter at the midpoint, and management expects 800G revenue to rise nearly fivefold sequentially.
• It makes its own lasers. Management says in-house lasers helped it avoid shortages, that substrate supply is secured into next year and that inventory covers the rest of 2027. It also plans to expand laser capacity by about 350% by the end of 2027.
• Operating leverage is large. Non-GAAP operating expenses of $70 million to $80 million a quarter sit against revenue that goes from $192 million to a guided more than $500 million. At a 32% gross margin and $80 million of expenses, fourth-quarter operating income would be about $80 million (my estimate).
• The mix points the right way. 1.6T is the highest-margin product, and management cites margins of 55% to 65% on laser products for co-packaged optics (CPO) and above 50% on EOSAP products later.
• Liquidity is adequate for now. About $1.0 billion of liquidity after debt, with the only convertible notes not due until 2030.

What could go wrong?

• Thesis breaker: funding and dilution. The build needs more money than the company has. If the shares fall, the next raise costs more shares, and the 20.8% rise in the share count this year is a precedent.
• Customer concentration. Three customers are 92% of revenue and one counterparty owes 67% of receivables. A design loss, a delayed deployment or a pricing negotiation at one of them changes the whole growth path.
• Capacity and cycle. Net property rose 85% in six months and inventory rose 52%. If demand normalizes after the equipment is installed, utilization and margins can reverse quickly.
• Margins. Non-GAAP gross margin was 29.8% in the quarter and the 40% goal has no date. Management also named constraints on DSPs, TIAs and lasers for 1.6T.
• Trade and geopolitics. Production sits partly in Ningbo and Taipei. Tariffs cost $1.9 million in the second quarter, there was a $5.7 million refund after IEEPA tariffs were overturned, and management said it was too early to quantify possible U.S. restrictions on Chinese transceivers.
• Competition and technology. New Chinese laser capacity, CPO and silicon photonics can change customer choices. Management put the time for new Chinese capacity to reach comparable quality at two to three years.

Management execution

Dr. Thompson Lin leads the company and Dr. Stefan Murry is chief financial officer. The record is mixed. On the positive side, revenue has set a record for five straight quarters, 800G volume more than doubled sequentially in the second quarter, and the company gives quarterly guidance and discloses customer concentration. On the other side, the operating loss in every year from 2021 to 2025 ran between $41.4 million and $70.9 million, the share count rose 20.8% in under ten months, and the plan triples 800G and 1.6T capacity in about six months. The 40% gross margin goal has no date and the company gives no total capital spending figure. Assessment: 9/20, mixed. The next test is whether capacity arrives on schedule without another large raise.

Stock Analyza scorecard

Revenue grew at about 21% a year from 2021 to 2025, which earns the full growth point. A score of 4 is the top of Standard and 5 would be Quality Growth, so a single point on the balance sheet or runway decides the engine. I gave the balance sheet its point because liquidity of about $1.1 billion exceeds the $217.7 million of debt and the convertible notes are not due until 2030, and I gave the runway one point because AI demand may last, though returns on the new capacity are unproven. Character change: changing. A small CATV and 100G supplier is becoming a capital-intensive AI optics builder that depends on the equity market and on three customers. I handle it through lower multiples and a higher margin of safety, and it means Hold is the ceiling for this valuation. Business quality is 36/100: moat and pricing power 8/20, return versus cost of capital 2/20, balance sheet 7/15, earnings and cash quality 2/15, growth and runway 11/15, management and capital allocation 5/10, dilution and governance 1/5. The valuation score is 26/100: discount to Base 14/40, protection against the Bear case 0/20, agreement between methods 3/15, historical valuation 4/10 (not meaningful, neutral), reverse valuation 3/10, data and model quality 2/5.

Confidence score

The company's own reporting is current, and I read the second-quarter release on SEC.gov. The 10-Q figures come from a filing summary page, consensus is a dated third-party figure and total capital spending is my estimate. Predictability is low because the business is cyclical and concentrated. Robustness is low because the two main methods differ by 57%, and the Bull case is 7.3 times the Bear.

What I would watch from here

Today's readings: green on economic shares (about 98.4 million), yellow on the largest customer (42%), on non-GAAP gross margin (29.8%, guided to 29% to 30.5%) and on inventory (up 35.2% in the quarter against revenue up 27.0%, 8.2 points faster). Financing reads green only because of timing: the $600 million program has just finished, and management has said it will use more equity and debt. The next review point is the third-quarter report, listed for November 5, 2026.

The three most important thesis breakers

1. Another large equity or debt raise is needed at a lower price, or capital spending runs well above the funded plan.
2. Gross margin stays near 30% while capacity triples, so the 40% goal slips beyond 2027.
3. One of the two large data center customers cuts orders, changes suppliers or digests inventory.

Stress test and independent check

I recomputed the share bridge, all three primary values, the Base cash flow model, the P/E check, the triangulation, the sensitivity tables, the yields and the reverse valuation with a second, separately written calculation. I also checked the reported inputs against the second-quarter release, the earnings call, the 10-Q summary and the market price. The results matched to the cent. The remaining weaknesses are in the inputs, not the arithmetic.

Every check on the cash flow model ($53 to $93) and both all-method checks ($90 and $81) come out below the price. The checks on the primary method alone land on both sides of it ($86 to $136). That mixed picture is why the verdict is a hold and not a reduce.

The strongest counter-thesis, for the bulls: demand exceeds supply through mid-2027, capacity triples, 1.6T orders above $200 million are in hand, and fourth-quarter revenue above $500 million is already a $2.0 billion run rate. If capacity runs at 75% and gross margin approaches 40%, the Bull case of $168 is not extreme. The response is that this is possible and it is in my Bull, but the price already pays 16.3 times Base EBITDA after the build, the Bear case is 77% below the price, and the funding need means shareholders carry the execution risk before the cash arrives.

Audit flags: consensus revenue and EPS come from a third-party page dated August 19 that I could not refresh, and the FY2027 figure is a summary number. Total capital spending, the working capital build, the 25% share of unvested Amazon warrants and the employee award count are my estimates. I read the second-quarter release on SEC.gov, but the 10-Q figures (capital spending split, warrants, customer and receivable concentration) come from a filing summary page, not the filing. Customers are not named, so I cannot say which customer holds which share. Bank acceptance payable of $33.9 million is in debt, and operating leases of $78.1 million (present value) are excluded from net debt because rent sits in operating expenses. The Bear cash flow value is close to zero and has a terminal value of 2.7 times EBITDA, so it carries little weight in my judgment even though it carries 30% in the arithmetic. No return on invested capital was computed from audited invested capital, because earnings have been negative. Treat the result as conditional, because several critical inputs are estimated and the two main methods diverge by more than 30%.

Final verdict: hold at $100.68

Applied Optoelectronics has orders, a ramp and a capacity plan that the market is willing to fund. The shares already pay 16.3 times the EBITDA I expect after the money has been spent, and the money has not been spent yet. The buyers in the $600 million ATM paid an average of $105.36, which is 4.6% above today's price.

Verdict: HOLD at $100.68, do not add. Triangulated Base value $94, downside 6.6%, Confidence D (9/25). The required margin of safety of 75% exceeds the framework's practical ceiling, so no maximum buy price is issued.

The main risk to this verdict is that the capacity arrives on time, the 1.6T and laser mix lifts gross margin toward 40%, and the company funds the rest with cash flow instead of new shares, which would put the value near my Bull case of $168. New money belongs on the watchlist. Demand is the part of this story that has already been confirmed. The unconfirmed part is how many new shares it takes to build the supply.

Sources

Primary sources include Applied Optoelectronics' second-quarter 2026 earnings release, a summary of the Form 10-Q for the quarter ended June 30, 2026, the second-quarter earnings call transcript and the report on the completed at-the-market offering. The October 12, 2026 close of $100.68, the share count, the cash flow history, the price history and consensus estimates come from StockAnalysis (forecast page) and are labeled as estimates where they are third-party figures.

The reference market price is the October 12, 2026 regular session close at 4:00 PM EDT. Balance sheet figures are as of June 30, 2026, and the common share count is as of October 12, 2026. The economic share count, the bridge to year-end 2026, normalized EBITDA, scenario values, the 800G and 1.6T unit utilization check and the robustness tests are analytical estimates, not company guidance.