Applied Optoelectronics, Inc. AAOI

Key data

The business

Applied Optoelectronics makes the laser-based parts that carry data as light through fiber. It is vertically integrated, growing its own laser chips in its own fabs in Texas, Taiwan and China, which is the edge: when transceivers are scarce, it controls its own supply and cost. Two businesses are colliding inside it. The old one is CATV, the gear cable operators buy to upgrade their networks, a lumpy, cyclical order book. The new one is data-center optical transceivers, the 800G and 1.6-terabit modules that plug into a switch and connect racks of GPUs in an AI cluster. Amazon's cloud arm signed on as a customer in March 2025, warrant and all, and is anchoring that ramp. This quarter Data Center became 56% of revenue, the largest segment. The catch is that hyperscalers can qualify a second supplier and cut your price the moment they do.

The numbers

Revenue is not just growing, it is accelerating: up 51% year on year in the March quarter, then up 86% in June.

The engine is entirely Data Center, and the mix shift is the whole story of the margin.

Set that against the long record, and the leap the market is pricing comes into focus.

Now the compounding math. Revenue went from $249 million in 2024 to $456 million in 2025, and the Street models about $1.05 billion this year and $2.96 billion next, a near-twelve-fold rise off 2024. The entire $8.5 billion of equity value rests on 2027 arriving roughly on schedule and, more importantly, arriving profitable: consensus wants $5.55 of earnings in 2027 from a company that has lost money every full year for eight. The revenue ramp I believe, the demand is visibly there and accelerating. What I do not yet see in the filings is the margin the profit forecast requires: gross margin fell 260 basis points year on year even as revenue nearly doubled, because the fast-growing Data Center dollars come in thinner than the CATV dollars they are replacing. The print that settles this whole memo is second-half gross margin. If it does not climb back toward the mid-30s while volume scales, the 2027 profit never shows up no matter how big the top line gets.

Management

The signal from the people who know the numbers best is loud and one-directional: zero insider buys in twelve months against $116 million of sales across 33 transactions. Founder and CEO Thompson Lin sold about $20 million himself in May and June 2026, and insider DeLaney sold $10.7 million. Plan status is not disclosed in the filings, so read them as discretionary until a Form 4 footnote says otherwise. At the same time the company is filling the till from every other door: a fresh $600 million at-the-market equity program authorized in August, on top of convertible notes, which is why cash jumped from $206 million at year-end to $500 million by June. Capex ran $179 million in 2025, 39% of revenue, poured into the capacity this ramp needs. Insiders selling their own shares while the company sells yours is not illegal, but it is not the posture of people who think the stock is cheap.

How it fails or surprises you

Gross margin never scales. Margin is 27.7% and falling as Data Center mix rises. The 2027 profit forecast implicitly needs mid-30s. If Q3 gross margin prints below 28% on revenue above $190 million, the volume-fixes-margin thesis is dead and the loss stays structural.

The raise never stops. Operating cash burn hit $85 million in the March quarter alone; the $600 million ATM sits on top of convertibles. A single slipped quarter forces another raise, and the last one would come at a lower price. Watch Q3 share count and any ATM drawdown.

A second hyperscaler qualifies (right tail). Amazon is signed. If one more named 10%-plus customer certifies AAOI's 800G/1.6T at volume in the next two quarters, revenue clears the $2.96 billion 2027 bar and margin scales with it. The market discounts this because the stock ran from $18 to $234 and back to $105 in a year, so it prices qualification risk, not qualification.

Closing thoughts

A specific number settles this one, so name it: second-half 2026 gross margin, and the first GAAP-profitable quarter behind it. Revenue is not the question, demand is visibly accelerating and the Amazon relationship is real. The question is whether a laser plug sold to a hyperscaler that can dual-source you ever carries enough margin to matter. The left tail is concrete and near, thin margins plus relentless dilution, and given a stock that has already round-tripped from $18 to $234, a stumble is punished hard. The right tail, a second hyperscaler at 1.6T, is genuinely large and not yet paid for. On balance the downside is the fatter tail at 14 times sales with no profit underneath, and the burden of proof sits squarely with the next two prints.

The bet is still that Amazon and the other giant cloud companies keep buying Applied Optoelectronics' optical transceivers and that the company turns the revenue into profit before dilution overwhelms you. What breaks it is margin: if second-half gross margin does not clear 30% while revenue holds above $190 million a quarter, the 2027 profit story is broken, and the Q3 report is where you find out.